Web

Anulytics
emystiIied

A Murketer's Goide to Lnderstunding
How Yoor Web Site AIIects
Yoor Bosiness


Eric T. Peterson







Purts oI this book ure sopplemented ut the uothor's Web site:
www.webunulyticsdemystiIied.com



Copyright ooq Iric T. Peterson




























Web Analytics Demystified: A Marketer's Guide to Understanding How Your Web
Site Affects Your Business

Copyright 2004 Lric 1. Peterson

Designations used by companies to distinguish their products are oten claimed as
trademarks. In all instances where the author and publisher are aware o a claim, the
product name shall appear in initial capitals. All trademarks, other than !eb .vat,tic.
Dev,.tifiea are the property o their respectie companies. No patent liability is assumed
with respect to the use o inormation contained herein. Although eery precaution has
been taken in the preparation o this book, the publisher and author assume no
responsibility or errors or omissions. Nor is any liability assumed or damages resulting
rom the use o the inormation contained herein.

All rights resered. No part o this book shall be reproduced, stored in a retrieal system or
transmitted by any means, electronic or otherwise, without the written permission o the
author.

Lditing by Maura K. lallam | lallam Creatie Serices
Coer design and original illustrations by Chris Garlotta
Published by Celilo Group Media and CaePress

International Standard Book Number: 0-943584-2-8



















TIIs book Is dedIcuLed Lo CIIoe MIcIeIIe PeLerson




ConLenLs i

Acknowledgements ..................................................................................................... vi
About the Author......................................................................................................... vi
Ioreword ....................................................................................................................vii

Chapter J Why Web Analytics Demystified? ................................................................ J
Intended Audience or 1his Book.................................................................................................1
\hat 1his Book Is Not ...................................................................................................................2
Use o Lxamples and \eb Analytics 1ips throughout 1his Book..........................................3

Chapter 2 Overview of Web Analytics ...................................................................... S
\hat is \eb Analytics` .................................................................................................................5
1echniques and 1echnologies Used in \eb Analytics ..............................................................6
\eb 1raic Data.........................................................................................................................6
\eb 1ransactional Data.............................................................................................................6
\eb Serer Perormance Data .................................................................................................
\eb Analytics 1ip: Lxamining Page \eights........................................................................9
Usability Studies...........................................................................................................................9
User Submitted Inormation and Related Sources................................................................9
\eb Analytics 1ip: Integrate lorms-based Data ................................................................11
\here 1his is Going.......................................................................................................................11
1he Continuous Improement Process......................................................................................12
1he Process, Deined................................................................................................................12
Strategies or Success................................................................................................................14
\eb Analytics 1ip: Applying the Continuous Improement Process ..................................16

Chapter 3 Different Web 1raffic Data Sources........................................................... J7
1he 1wo Most Common Sources plus One Uncommon One..............................................1
locus on \eb Serer Log liles as a Data Source ....................................................................18
Adantages o \eb Serer Log liles as a Data Source......................................................22
Disadantages o \eb Serer Log liles as a Data Source.................................................23
Best Practices \hen Implementing a \eb Serer Log lile Solution..............................2
Lxamples o \hen to Use a \eb-Serer Log-lile Application.......................................29
Lxceptions, Caeats, and so on ..............................................................................................30
locus on Page 1ags as a Data Source.........................................................................................30
Adantages o Page 1ags as a Data Source ..........................................................................33
Disadantages o Page 1ags as a Data Source .....................................................................35
Best Practices \hen Implementing a Page 1agging Data Collection Solution .............39
Lxamples o \hen to Use a Page-1agging Data Collection Solution.............................41
1ags into Logs............................................................................................................................42
About Real-1ime` Data Collection and Reporting ................................................................43
Cookies and 1heir Use in \eb Analytics Applications ...........................................................44

Chapter 4 Web Analytics 1erminology ...................................................................... 46
A Small Group o Mostly Useless 1erms ...................................................................................4
lit ................................................................................................................................................4
Click-1hrough............................................................................................................................4
A Slightly Larger Group o Useul But Conusing 1erms.......................................................48
Page View....................................................................................................................................48
Visit ..............................................................................................................................................50
ii ConLenLs

Visitor or Unique Visitor ......................................................................................................... 51
Reerrer ....................................................................................................................................... 52
A Medium-Sized Group o 1ruly Useul 1erms....................................................................... 53
Conersion Rate ........................................................................................................................ 53
Abandonment Rate................................................................................................................... 54
Attrition....................................................................................................................................... 54
Loyalty, lrequency and Recency ............................................................................................ 55
1he Pyramid Model o \eb Analytics Data.............................................................................. 56
lits .............................................................................................................................................. 5
Page Views.................................................................................................................................. 5
Visits ............................................................................................................................................ 58
Unique Visitors.......................................................................................................................... 58
Unique Identiied Visitors ....................................................................................................... 59

Chapter S Content Organization 1ools...................................................................... 60
Metrics Content Organization 1ools Should Proide ............................................................. 62
Lxamples o Dierent Content Organization Strategies ........................................................ 62
Directory Structure Content Organization........................................................................... 63
Inormation Architecture Content Organization................................................................ 63
Business Objectie Content Organization ........................................................................... 65
Risks Associated \ith Content Organization........................................................................... 68
\eb Analytics 1ip: \hy Are Content Groupings Important .............................................. 69

Chapter 6 Process Measurement 1ools ..................................................................... 70
An Important Note about Process and Conersion Rates ..................................................... 1
Metrics Proided by Process Measurement 1ools.................................................................... 2
Lxample Uses or Process Measurement 1ools........................................................................ 3
\eb Analytics 1ip: \hen to Require Registration................................................................... 6
Risks Associated with Process Measurement 1ools.................................................................
About A,B 1esting ........................................................................................................................ 8

Chapter 7 Visitor Segmentation 1ools........................................................................8J
Metrics Proided by Visitor Segmentation 1ools..................................................................... 82
Lxamples o Uses or Visitor Segmentation 1ools .................................................................. 83
Risks Associated with Visitor Segmentation 1ools .................................................................. 84
\eb Analytics 1ip: Use Segments to Collect Demographic Data......................................... 85

Chapter 8 Campaign Analysis 1ools.......................................................................... 87
Metrics Proided by Campaign Analysis 1ools......................................................................... 88
An Important Note about Response Rates................................................................................ 89
Lxample Uses or Campaign Analysis 1ools............................................................................. 91
Risks Associated with Campaign Analysis 1ools...................................................................... 93
\eb Analytics 1ip: 1rack Lery Campaign \ou Run, Careully........................................... 94

Chapter 9 Commerce Measurement 1ools ................................................................ 9S
Metrics Proided by Commerce Measurement 1ools ............................................................. 96
Merchandising............................................................................................................................ 96
Reerring Sources ...................................................................................................................... 9
Online Purchase Process ......................................................................................................... 9
ConLenLs iii

Sessionizing,Identiying Inormation....................................................................................98
Lxample Uses o Commerce Analysis 1ools.............................................................................99
Risks Associated with Commerce Analysis 1ools.................................................................. 101
\eb Analytics 1ip: 1ie Internal Searches to Purchases........................................................ 102

Chapter J0 An Introduction to the Customer Life Cycle...........................................J04
Reach.............................................................................................................................................. 105
Acquisition .................................................................................................................................... 106
Conersion .................................................................................................................................... 106
Retention ....................................................................................................................................... 108
A Note about the Aerage` Visitor ........................................................................................ 108

Chapter JJ Measuring Reach..................................................................................... JJ0
General Metrics to lelp Measure Reach................................................................................. 110
Oerall 1raic Volumes........................................................................................................ 110
Number o Visits .................................................................................................................... 112
Number o New Visitors ...................................................................................................... 114
Ratio o New Visitors to Returning Visitors ..................................................................... 116
Percentage o New Visitors .................................................................................................. 118
Lntry Pages and Contents..................................................................................................... 120
Visitor Geographic Inormation.......................................................................................... 121
Lrror Pages .............................................................................................................................. 123
Interested Visitor Share......................................................................................................... 124
Campaign Reach Metrics ............................................................................................................ 126
Impressions Sered ................................................................................................................ 126
Open Rates .............................................................................................................................. 12
Key Perormance Indicators Recommended or Measuring Reach ................................... 130
Metrics \ou Should Be Asking or Beore \ou Spend Adertising Dollars .................... 131

Chapter J2 Measuring Acquisition............................................................................J33
General Metrics to lelp Measure Acquisition ....................................................................... 133
Percent New Visitors............................................................................................................. 133
Aerage Number o Visits per Visitor................................................................................ 134
Aerage Number o Page Views per Visit ......................................................................... 135
Aerage Pages Viewed per Visitor ...................................................................................... 13
Page Stick` and Slip` ......................................................................................................... 138
Cost per Visitor....................................................................................................................... 141
Ratio o New Visitors to All Visitors.................................................................................. 143
leay User Share ................................................................................................................... 143
1op Pages and Content Requested By New Visitors....................................................... 144
Content locus`..................................................................................................................... 148
Percent o Visits Under 90 Seconds.................................................................................... 149
Aerage 1ime Spent on Site ................................................................................................. 150
Campaign Response Metrics...................................................................................................... 151
Responses and Respondents ................................................................................................ 152
Cost-per-Acquisition , Cost-per-Click............................................................................... 154
Reerring Sources and Marketing Mix`.................................................................................. 15
Reerring Domains................................................................................................................. 158
Reerring URLs....................................................................................................................... 161
iv ConLenLs

Naigation Path by Reerring Source ..................................................................................162
Search Lngines.........................................................................................................................164
Search Keywords and Phrases ..............................................................................................166
Notes on Search Lngine Marketing and Search Lngine Optimization...............................10
Key Perormance Indicators Recommended or Measuring Acquisition ..........................11

Chapter J3 Measuring Conversion............................................................................ J73
\hich Conersion Rates Should Be Measured.....................................................................14
Site-wide Conersion Rate.....................................................................................................14
lome to Purchase ..................................................................................................................14
Search to Purchase..................................................................................................................15
Special Oer` to Purchase..................................................................................................15
Lead Generation......................................................................................................................16
Naigation to Important Inormation.................................................................................1
Using Process Measurement 1ools ...........................................................................................18
Metrics to lelp Measure Conersion .......................................................................................18
Conersion Rates ....................................................................................................................19
Abandonment Rates ...............................................................................................................180
Campaign Conersion Metrics ...................................................................................................181
Campaign Conersion Rate...................................................................................................182
Cost-per-Conersion ..............................................................................................................184
Campaign Return on Inestment .........................................................................................186
Segment, Commerce and Search Related Conersion Metrics.............................................188
Segment Conersion Rates....................................................................................................188
Aerage Order Value..............................................................................................................190
AOV or New and Returning Customers...........................................................................191
Percent o Orders rom New and Repeat Customers ......................................................193
New and Repeat Customer Conersion Rates...................................................................194
Sales per Visitor .......................................................................................................................195
Searches \ielding Results to Search No Results................................................................196
\eb Analytics 1ip: low to Increase \our Purchase Conersion Rate .............................19
Key Perormance Indicators Recommended or Measuring Conersion ..........................202

Chapter J4 Measuring Retention ..............................................................................204
Metrics to lelp Measure Retention ..........................................................................................205
Number o Returning Visitors..............................................................................................205
Aerage lrequency o Returning` Visitors ......................................................................206
Ratio o Returning Visitors to All Visitors .........................................................................208
Loyalty metrics.........................................................................................................................208
lrequency o Visit...................................................................................................................210
Recency o Visit.......................................................................................................................212
Visitor Segmentation Metrics or Retained Visitors...............................................................213
Actiity o Retained` Visitors .............................................................................................214
Views o Key Pages and Contents .......................................................................................214
Retained Visitor Conersion Rate........................................................................................215
Customer Retention Rate ......................................................................................................21
Aerage lrequency o Return or Retained` Visitors ....................................................218
\eb Analytics 1ip: low to Retain Customers .......................................................................219
Key Perormance Indicators Recommended or Measuring Retention .............................220
ConLenLs v


Chapter JS Brining It All 1ogether Using Key Performance Indicators .................. 222
low Do I Build a KPI \orksheet ......................................................................................... 223
Use the Right Measurements................................................................................................ 224
1hink about Presentation...................................................................................................... 225
Allow or Annotation ............................................................................................................ 225
Automate Deliery ................................................................................................................. 225
1he lour Broad Business Models............................................................................................. 226
Online Commerce .................................................................................................................. 226
Adertising............................................................................................................................... 226
Lead Generation..................................................................................................................... 22
Customer Support .................................................................................................................. 22
Key Perormance Indicators by Business Model ................................................................... 228
Online Commerce Key Perormance Indicators .............................................................. 228
Adertising Key Perormance Indicators........................................................................... 230
Lead Generation Key Perormance Indicators ................................................................. 231
Customer Support Key Perormance Indicators .............................................................. 232

Chapter J6 Advanced 1opics.................................................................................... 23S
Data Integration ........................................................................................................................... 235
Custom Data Collection........................................................................................................ 23
Back-Lnd Data Integration................................................................................................... 238
Use o Personas in Design and Architecture .......................................................................... 242
low Personas Are Used....................................................................................................... 242
Personas Used in Persuasion Architecture ........................................................................ 244
\hy Is 1his Releant ........................................................................................................... 244

Bibliography............................................................................................................. 246
Books and Llectronic Documents............................................................................................ 246
\eb Sites Reerenced.................................................................................................................. 24
List o 1ables and ligures .......................................................................................................... 248

Index .........................................................................................................................2SJ
vi orewurd

Acknowledgements

1his book would not hae been possible without help rom the ollowing people:

Jim Sterne or paing the way and proiding endless encouragement
Bryan Lisenberg or adice regarding how to get this book published
1erry Lund and Dylan Lewis or their reiew o early ersions o this book
Rand Schulman or his comments and encouragement
Maura lallam, Chris Garlotta and Nik Blosser or technical assistance
Amity and Chloe Peterson or understanding that Chloe`s daddy had to write

Additionally, the author would like to thank John Breese, Dustin Robertson and the entire
BackcountryStore.com crew or allowing me to use their \eb site and data in examples
throughout this book.


Aboot the Aothor

Lric 1. Peterson has worked in the ield o \eb analytics since 1998 as a \eb deeloper,
\eb marketer and \eb analytics specialist. As the director o proessional serices or
\ebSideStory, creators o litBox, Mr. Peterson has had the opportunity to proide
analytics consulting or some o the largest and most recognized brands in the world. More
recently, Mr. Peterson has joined Jupiter Research as a Site Operations and 1echnology
analyst, adising Jupiter customers on many o the technologies and ideas presented in this
book. Mr. Peterson lies in San Diego, Caliornia, with his wie, daughter Chloe and a
menagerie o animals.

orewurd vii

Ioreword

Lric Peterson asked me to write the orward to !eb .vat,tic. Dev,.tifiea because, well, I`m
one o three people in the world who was written a ull-length book on the subject ,lurol
Inan rom Australia is the third writer,. 1he question I wanted to answer when I wrote !eb
Metric. in 2002 was "\hat is \eb Analytics All About" I wanted the aerage business
person to stop being earul o what they didn't understand and start aailing themseles o
the rich inormation at their ingertips.

\eb Analytics may seem like the counting o click-through, page iews and ,i you're lucky,
reenue. But beyond that, it is a collection o data elements that reeal the mind-set o your
marketplace. \eb analytics gies you a ront-row look into the hearts and minds o your
customers and prospects, helping you to answer questions like low do they eel`, \hat
do they graitate toward` and \hat holds their attention this week`

1he ability to strap the analytics LKG onto the world at large and see how it reacts to your
adertising, your product descriptions, and your persuasion process is simply too aluable to
stay a secret. 1o spread the word that this kind o inormation was aailable I wrote the irst
book on the subject o \eb Analytics, !eb Metric.: Prorev Metboa. for Mea.vrivg !eb ite
vcce...

I consider mysel a proessional explainer, I am a marketing guy with a degree in
Shakespeare who started out selling Apple IIe's and VisiCalc in the early 80s and neer
looked back. \hen I meet executies who just don't get it, I do my best to put it all into
terms they can understand and I get ery excited when the light bulb goes on and they see
the potential.

Because I am a published author on the subject and a consultant, these executies inariably
turn to me or help. 1hey hae excellent questions like \hat should we measure`, low
do we educate the rest o the company` and low do we conince the top brass that the
inormation we're getting o our website is ar more probatie than anything we'e had in
the past`

Inariably, corporations reach that point where they want to choose the most appropriate
technology, implement the systems and start reaping the rewards, three ery dierent steps.
1hey are ready to moe on to the next question: low Do I Get it Done` 1his is where I
all back on the two most important phrases in consulting: "It depends" and, "1rust me".

low you get it done depends on what you hae in place at the moment, how enthused your
technical sta is by the prospect o yet another initiatie, and how well the resulting reports
will be receied and put to use by the business people who hae the most to gain. It all
depends. And trust me, I'e been through this with enough companies to know that this
phrase always resonates as true.

Most clients hae already passed my book around and are prepared or more speciic
education - the down and dirty details o why some data elements will reeal great insight
viii CIupLer 1

and how others won't. \here the data stream leaks and why the ecommerce engine
disagrees with the campaign management systems. Phone calls, PowerPoint presentations,
workshops, personal attention and repeated explanations until the inormation inally sank in
and became gospel - a process that oten took much longer than any o us wanted to spend.
1hat is, until now.

Now, I will be able to hand my clients a copy o !eb .vat,tic. Dev,.tifiea and say, "Read
Chapters 4 and 11 through 16 and call me in the morning. 1hen we can get started." Now I
hae a deinitie reerence or the how` o \eb Analytics, an authoritatie work that is the
perect compliment to the what` and the why` that I hae already explained in !eb
Metric..

Regarding the author, I'e neer met anybody as bursting with enthusiasm about \eb
Analytics as Lric Peterson. le is high energy, highly ocused and a true belieer. 1he only
way I know to get his mind o the subject is to mention his daughter. le really wants the
world to understand the nitty-gritty o how it works. le really wants people to be able to
roll up their sleees and dig in and do the actual work inoled in continuous improement,
making their \eb sites more aluable to their customers and their companies. \hile I'm
looking out oer the horizon and pointing the way, he's tuning the engine and checking the
luids.

I beliee that the Internet has gien us a new window on the world, a window where we can
keep an eye on our customers and prospects. Lric Peterson makes sure you understand the
process o turning grains o sand into panes o glass and keeping that glass as clean as
possible in order to keep the picture o your customers and prospects as accurate and
aluable as possible. In !eb .vat,tic. Dev,.tifiea Lric has eectiely demystiied` the
process o making this data work or you, doing his part to share the secret with the rest o
the world.

Jim Sterne
Author: !eb Metric.: Prorev Metboa. for Mea.vrivg !eb ite vcce..
www.targeting.com
www.emetrics.org















WIy Web AnuIyLIcs DemysLIIIed? 1

CHAPTIR 1
WHY WIB ANALYTICS
IMYSTIIII?


I know that half of my advertising budget is completely wasted,
I'm just not sure which half.
Lord Leverhulme, founder of Unilever


!eb .vat,tic. Dev,.tifiea . Mar/eter`. Cviae to |vaer.tavaivg or Yovr !eb ite .ffect. Yovr
v.ive.. is intended to be exactly that, a useul guide to help you get more out o your current
online inestments. \hile there are seeral good books about rb, you would want to set up
a \eb analytics program, and hundreds o dierent endors to sell you the tools you will
need, there has been a lack o inormation about how to actually extract useul inormation
rom these applications once they`re installed.

Lnter !eb .vat,tic. Dev,.tifiea-a book that will demystiy the marketing mumbo-jumbo
that endors proide and look behind the curtain, to help you discoer that \eb analytics
doesn`t hae to be diicult or conusing at all

Ater reading this book you will take three things away with you:

Awareness o what \eb analytics is, rom a practical perspectie, and what a \eb analytics
program can and cannot tell you.

1. An understanding o which tools and statistics are actually useul to a \eb analytics
program.
2. Knowledge o which aailable statistics are most useul to your particular online
business, and how, ideally, those statistics should be used.

1he goal o this book is to elucidate which statistics, ratios, and comparisons are most likely
to help you improe the oerall quality o your \eb site or your isitors and customers.
Independent o your online business model, !eb .vat,tic. Dev,.tifiea presents each set o
measurements as they relate to the standard customer lie cycle-reach, acquisition,
conersion and retention.
Intended Audience for This Book
I you are reading this book you probably all into at least one o the ollowing groups:
CIupLer 1


1. \eb marketers, those challenged with bringing isitors to the \eb site and enticing
them to complete some aluable actiity.
2. \eb operations managers, those challenged with proiding a high-quality \eb
presence to improe the likelihood o success or the \eb marketers.
3. \eb executies, or just executies in general, those ultimately responsible or their
company`s business success.

1hese are three ery broad groups, and that you may belong to one or more o them. 1he
key to taking something away rom !eb .vat,tic. Dev,.tifiea is that you, the reader, hae
some ested interest in the success o the \eb site. I part o your job unction is to
monitor and improe your company`s online return on inestment, this book is or you.

Parts o !eb .vat,tic. Dev,.tifiea are technical, but you don`t need a background in PLRL, C
and UNIX to get alue rom this book. Sections o this book dele into the business alue
o your \eb site, but you won`t need an MBA to understand why you`d want to want to take
a particular measurement. lopeully you are lucky enough to work with two smart people,
one technical and the other business-ocused, with whom you can work with collaboratiely
on your \eb analytics program.

I you are still wondering i this book is or you, turn to any page in Chapters 11 through 14
and conirm that at least some o the statistics look amiliar to you. I you hae neer
thought about the dierence between a new isitor` and returning isitor,` wondered
about the dierence between a hit,` a page iew` and a unique isitor,` or struggled to
igure out what top paths` can tell you, then perhaps this book is not or you.

I you hae been asking yoursel those kinds o questions repeatedly, howeer, keep reading.
What This Book Is Not
!eb .vat,tic. Dev,.tifiea is not a book describing rb, you would want to establish a \eb
analytics program or your online business. It is assumed that i you are considering
purchasing this book that you hae already inested time, eort and at least some money
trying to understand where your \eb site isitors come rom and what they do when they
get there.

\e`ll waste no time describing the characteristics` o a good \eb site analyst as the only
requirement or this position is desire: 1he desire to improe the quality o one`s \eb site.

!eb .vat,tic. Dev,.tifiea does not present a comprehensie history o traic analysis
applications starting with \usage and \eb1rends Log Analyzer. 1his book does little
to compare the currently aailable applications as applications come and go, but it will
occasionally mention a endor`s speciic solution i that solution proides some type o
inormation that is unique in the marketplace.

!eb .vat,tic. Dev,.tifiea also does not attempt to describe how you can build a recency,
requency and monetary alue ,RlM, model o your isitors or customers or outline more
WIy Web AnuIyLIcs DemysLIIIed?

than just the barest o steps you would want to take to integrate multiple sources o data, this
book is a practical guide. 1his it not to say that RlM models and data integration are
unnecessary or impossible to implement. 1he decision to skirt these subjects or aoid them
altogether was based on the relatie diiculty o implementation compared to the number o
potential readers likely to beneit rom their description. lor more inormation on the RlM
model, the author points you to Chapter 11 o Jim Sterne`s book !eb Metric.: Prorev Metboa.
for Mea.vrivg !eb ite vcce... lor a cursory introduction to data integration, please see
Chapter 16 o this book.

1his book is also a endor-neutral guide. 1he author has made a conscious decision to
discuss only those metrics that should be aailable in av, well-designed \eb analytics
application. 1he concept o well-designed` is an important one, especially considering that
there were an estimated 80 companies selling some type o \eb analytics solution at the
time this book was completed. Obiously all 80 o these companies are not proiding all o
the metrics needed to build the key perormance indication reports listed in Chapter 15. I
the reader is cautious when choosing an analytics solution, preerentially choosing a endor
with A, a proen track record o success in helping customers and B, an application that is
not brand new or completely reolutionary, it is sae to assume that the metrics you need to
make use o this book will be aailable to you.
Use of Examples and Web Analytics Tips throughout
This Book
1he examples used throughout this book hae been graciously proided by the kind people
at BackcountryStore.com, the Internet`s astest growing retailer o outdoor gear. 1he author
was introduced to BackcountryStore.com through his position at \ebSideStory, creators o
litBox, and has worked closely with management at BackcountryStore.com to help them
optimize their \eb site and analytics program using many o the concepts presented in this
book. 1he author is deeply indebted to eeryone at BackcountryStore.com or their
permission to use their site and data in examples throughout !eb .vat,tic. Dev,.tifiea. More
inormation about BackcountryStore.com can be ound at www.backcountrystore.com.
q CIupLer 1




Iigure J: BackcountryStore.com home page, a shining example of how to drive
visitors to purchase online.

Unless otherwise noted, images used in examples hae been taken rom the litBox
Lnterprise and,or litBox Commerce interace, the analytics package that
BackcountryStore.com has standardized on since 1999. More inormation about the litBox
amily o serices can be ound at www.hitbox.com or www.websidestory.com.

Additionally, throughout this book the author has included seeral \eb Analytics 1ips.`
Presented in the irst person, these tips are intended to supplement the examples in order to
help you better understand exactly how this inormation is practically applied.

OvervIew oI Web AnuIyLIcs g

CHAPTIR
OVIRVIIW OI "WIB
ANALYTICS


If you build a Web site, you must have a solid reason for doing so. Once the Web
site is operational, it is crucial to understand how well it worksfailure to do so is a
waste of time and money.
Hurol Inan in Measuring the Success of your Web site


\eb analytics is a term that is easy to banter around when discussing your \eb site but
more diicult to truly understand. It is a complex interaction between human beings,
machines and code that oten generates large amounts o data that then need be mined,
manipulated and presented in meaningul ways. Don`t be misled, a \eb analytics program
can be ery diicult to implement successully-but reading this book will proide insight
that will make your lie a little easier.
What is Web Analytics
Depending on whom you ask, \eb analytics can hae a wide ariety o deinitions but is
ultimately ocused on a single goal: Understanding the online experience such that it can be
improed. A ew select examples o how \eb analytics is described:

According to Guy Creese o the Aberdeen Group, a noted authority on the subject,
\eb analytics is the monitoring and reporting o \eb site usage so enterprises
can better understand the complex interactions between \eb isitor actions and
\eb site oers, as well as leerage that insight or increased customer loyalty and
sales."
Sarner and Janowski o the Gartner Group state that \eb analytics uses a ariety
o data and sources to ealuate \eb site perormance and isitor experience,
potentially including usage leels and patterns at both an indiidual and an
aggregate leel. Data and sources may include click-stream data rom the \eb
serer log, \eb transaction data, submitted data rom input ields on the \eb site
and data in the Internet customer repository. 1he goals are to improe site
perormance, both rom a technical and content perspectie, enhance isitor
experience ,and thus loyalty,, contribute to oerall understanding o customers and
channels, and identiy opportunities and risks.`
\ebopedia rom internet.com states that ,\eb analytics is, a generic term
meaning the study o the impact o a \eb site on its users. L-commerce companies
6 CIupLer z

oten use \eb analytics sotware to measure such concrete details as how many
people isited their site, how many o those isitors were unique isitors, how they
came to the site ,i.e., i they ollowed a link to get to the site or came there directly,,
what keywords they searched with on the site`s search engine, how long they stayed
on a gien page or on the entire site, what links they clicked on and when they let
the site. \eb analytic sotware can also be used to monitor whether or not a site`s
pages are working properly. \ith this inormation, \eb site administrators can
determine which areas o the site are popular and which areas o the site do not get
traic and can then use this data to streamline a site to create a better user
experience.`

lor the sake o this book, we will deine the study o \eb analytics as such:

!eb avat,tic. i. tbe a..e..vevt of a rariet, of aata, ivctvaivg !eb traffic, !ebba.ea trav.actiov., !eb
.errer erforvavce, v.abitit, .tvaie., v.er .vbvittea ivforvatiov ava retatea .ovrce. to bet create a
geveratiea vvaer.tavaivg of tbe ri.itor eerievce ovtive.

1he underlying assumption is that the inormation gained rom this assessment will be used
to improe the oerall quality o said isitor experience, although we now know that many
times there is a disconnect between haing the data and knowing exactly what to do with it.
1he reader should immediately recognize that this deinition is suiciently broad to include a
multitude o data sources not listed aboe. 1he ollowing sections urther deine each o the
data sources listed aboe.
Techniques and Technologies Used in Web Analytics
Web Traffic Data
\ithout a doubt the most popular source o \eb analytics data is \eb traic data.
1raditionally mined out o \eb serer log iles and more recently deried rom \eb clients
ia JaaScript page tags,` traic data is oten looked upon as a goldmine o inormation.
Generically speaking, this data source is inormed by an application ,such as page tags,
making a record o the request or inormation being receied ,such as clicking on a link to
see a \eb page,, and it is this record o inormation that is analyzed.

As \eb traic data is one o the ocuses o this book, later chapters and sections will coer
this data source in much greater detail.
Web Transactional Data
Any business conducting commerce online likely has access to inormation deried rom
these transactions, similar to the \eb traic data described aboe. Data points such as
number o customers, number o orders and aerage size o transactions should be aailable
to the \eb analytics process. Like \eb traic data, this data source will be coered at
length in ollowing chapters.
OvervIew oI Web AnuIyLIcs ,

Web Server Performance Data
1he Internet is a complicated piece o technology. Designed to allow applications rom
anywhere on the planet to quickly connect to resources and make them aailable ia a ariety
o protocols, the Internet has quickly eoled rom a military,academic experiment to
become an integral part o our daily lies. 1his explosie growth has not come without a
price. \hile the number o global Internet users is increasing eery day, technology oten
lags behind this usage, especially in rural areas and deeloping nations. 1he eect this
technology lag has on the isitor experience can be examined within the context o \eb
perormance data.

\eb pages and applications are made up o constituent parts-text, script, images,
multimedia iles, banner ads, and so on-which are reassembled within a isitor`s browser
based on a system o rules called the lypertext Markup Language ,l1ML,. lor a ery
complete explanation o l1ML, please see the description proided by its inentor, 1im
Berners-Lee in his book, !earivg tbe !eb. Regarding perormance, the general idea is the
larger a \eb page is in terms o oerall kilobyte weight,` that is, the size o the page`s
component parts, the longer it will take to download and render on a isitor`s computer.

lrom these concepts comes the ery aluable 10-second rule,` oten cited by usability
experts, including Jakob Nielsen. Nielsen, in his book on Usability Lngineering states that:

One tenth o a second ,0.1, is the limit or a user to eel that a system is responding
immediately when engaged in a task.
One second ,1, is the limit or a user`s low to remain uninterrupted, allowing them
to notice a delay but not losing the eeling o direct connection with the task.
1en seconds ,10, is the upper limit or keeping users ocused on a single task.
\hen aced with delays o ten seconds or longer, users hae a tendency to engage
in side tasks or become rustrated with the lack o system response.

A good way to think about this is by imagining a conersation that you may hae with
anyone. Rather than simply engaging in regular conersation, oten beating the 0.1-second
response by talking oer each other, imagine that you had a 10-second pause between each
statement. By the end o eery pause you would likely need to take an additional second just
to reorient yoursel to the last statement made, especially as the conersation went on oer
time.

Compare this to the isitor experience on the Internet, where people commonly engage in
side tasks while waiting or pages to load. In 10 seconds, while waiting or the next page to
load, any user could potentially:

Switch to their email browser and read email
Switch to another \eb browser and sur another site
Start a conersation with a nearby person
Answer or initiate a phone call
1ake a short walk
Close their browser or simply naigate to another site
S CIupLer z


1he reason that all this is so important is that on an aerage speed connection ,56Kb
modem,, a \eb page o only 65 Kb can take eight to ten seconds to download. \hile many
o us hae broadband connections to the Internet and perormance is typically not a
concern, it is important to consider that most o the global Internet audience still connects
using a modem.

1o put the 10 second rule` into context, here is a small sample o home page weights or
popular Internet destinations:

\eb Site Page \eight Load 1ime ,28K , 56K , ISDN,
Amazon 113 Kb 34 sec , 1 sec , sec
\al-Mart 164 Kb 49 sec , 25 sec , 10 sec
Microsot Network 61 Kb 18 sec , 9 sec , 4 sec
\ahoo! 55 Kb 16 sec , 8 sec , 3 sec
LSPN 206 Kb 61 sec , 31 sec , 12 sec
1able J: Lxample page load times, measured in October 2002.

As you can see, only \ahoo! een really comes close to hitting the 10-second target or
isitors using 28K modems and only \ahoo! and the Microsot Network hit the target or
56K modems. Keep in mind, the 10-second rule` describes the bigbeva target, the absolute
maximum or keeping a isitor engaged with what they`re doing.

O course there are excuses or making mistakes like this. Our users like all this
inormation, they don`t mind the wait` or \e hae an obligation to corporate to use these
images, it is part o how we build our brand.` 1he author`s personal aorite is, My CLO
likes how the page looks and he doesn`t want to attract isitors that use modems anyway.`
1hese are all ine excuses, but keep in mind that is all that they are-excuses that contribute
to a poor oerall isitor experience.

1o make a long story longer, one o the key actors to consider when analyzing the isitor
experience on your \eb site is, low long do our pages take to download` Since the
answer is dierent or dierent types o connection, you need to take this one step urther
and ask, low do our isitors connect to our \eb site` and use that answer to answer the
irst question. \hile not adocating unattractie \eb pages, smart \eb managers must
consider download times as part o their oerall \eb analytics program.

One endor that has demonstrated time and time again their knowledge o Internet
perormance and the download times as a component o \eb analytics assessments is
Keynote Systems. At the time o publication there were seeral interesting white papers on
this subject aailable or ree, including, 1he Lconomic Impacts o Unacceptable \eb-Site
Download Speeds,` lundamentals o Internet Measurement: A 1utorial` and \eb Page
Design and Download 1ime.`

OvervIew oI Web AnuIyLIcs q

Web Analytics Tip: Examining Page Weights
\heneer I`m analyzing a client`s \eb site, the irst thing I do is check the page weights o
their home page and three or our key pages on the site. 1his allows me to establish a
baseline or likelihood o success. I page weights are high and I run into other problems, I
tend to worry more-design issues plus potentially high download times equals real trouble.
Conersely, i page weights are low, I tend to worry less about other problems.

1he logic here is simple, i you make a process needlessly diicult ava make the user wait
long periods o time to try and get through said process, said user is much more likely to
abandon the process. Use an inexpensie serice like such as Keynote System`s
NetMechanic ,www.netmechanic.com, to monitor page perormance as you deelop new
pages.

Usability Studies
Usability studies are less o a \eb analytics data point and more o a science o their own,
haing been more or less an established practice or decades in sotware engineering.
Briely, usability studies inole working with real human beings, typically likely isitors or
actual customers, to better understand how people expect to use one`s \eb site or \eb-
based application.

Usability studies are a necessary component o a \eb analytics program but this book does
not dele deeply into this topic or two reasons. lirst, while usability studies can be
rustrating, diicult and expensie to perorm, there is no substitution to seeing how people
really interact with the content and naigation systems that you hae built. One cannot
describe the eeling you get when you hae spent weeks building a sophisticated naigation
system that takes adantage o the ull spectrum o aailable technology, only to hae eight
out o ten study participants ail to understand how said system works. Conersely, one can
also not describe the eeling one gets to see real users interacting with a system you hae
designed and inding the inormation they seek in a timely manner.

Second, there are a handul o well-written books on the topic o \eb usability, allowing the
author to deer to the expertise and wisdom o others on this subject. lor more inormation
on \eb site usability the author recommends two extremely well-written books, Jakob
Neilsen`s De.igvivg !eb |.abitit, and Stee Krug`s Dov`t Ma/e Me 1biv/! 1he ormer is the
recognized bible on the subject, written by the elder statesman himsel, Jakob Neilsen. 1he
latter is arguably one o the best books written about \eb sites eer, both humorous and
useul.
User Submitted Information and Related Sources
User submitted inormation and related sources includes inormation collected ia polls and
sureys, direct user entry orms and non-\eb based systems such as CRM databases and
1o CIupLer z

LRP systems. As with all \eb analytics data sources there are a number o dierent
endors able to help collect this kind o inormation and dozens o applications to help you
collect surey data rom your isitors.

1he alue o user surey and polling to the \eb analytics process is the ability to collect
more than just raw data, you are actually able to collect user opinions and eedback. 1his is
the dierence between knowing that 8,000 people isited your home page on Monday and
knowing that our out o ie isitors thought your home page was slow to load. 1he
downside o this data type is that oten there are no clear patterns in the data, sureys can be
ery diicult to write well and oten isitors complain about unwanted pop-up windows or
being asked to complete orms that they did not speciically request.

I you are going to surey your isitors or user base the recommendation is to do so as
unobtrusiely as possible. I you are careul, this data source can add a rich component to
your oerall \eb analytics program, i you are not careul you can seriously damage the
already tenuous relationship you hae with your isitor. \ays to succeed at this include:

Partner with a endor that has experience producing well-written sureys so that
you can aoid asking too many questions, asking questions that will likely lead to
noisy` answers or producing a surey that will do more damage than good.
Surey only participants who opt-in. Lntice these isitors using some kind o
bribe` -oer discounts, access or reebies to people who will take the time to
complete your surey. Most people will take the time to oer their opinions and
eedback to any company that oers something they can use personally ,the most
useul and enticing bribe is oten simply a git certiicate to any popular online
retailer or any amount oer >10 USD,.
Use a low-impact method or conducting the surey, such as OpinonLab`s Online
Opinion System ,www.opinionlab.com,. 1his Chicago-based company oers a
data collection system based on a subtle little icon that loats in the lower-right
portion o the \eb pages you choose to instrument with their code. At the time o
this writing, OpinionLab customers included Gateway Computers, Sprint,
Oerture, Dell Computers, Noel and Intel.

Perhaps the most useul user-collected data is the orms-based type, direct user answers to
speciic questions. Depending on the traic analysis solution you hae, this type o data can
enable complex segmentation strategies such as demographic segmentation and rules-based
segmentation. Imagine being able to sort all registered isitors to your site into men` and
women` based on their answer to the question what is your gender` asked during the
registration process-and then being able to see i isitor actiities dier by gender. Carry
this one step urther and be able to see i one age group ,based on the answer to the
question when were you born`, is abandoning a key process more requently than other
age groups. Imagine the powerul analytic opportunities that this type o data creates.

It is beyond the scope o this book to describe the complex interaction between customer
records and anonymous customer isits using data rom traditionally non-\eb based
systems such as CRM and LRP systems. 1he inormation proided in Chapter 16 on data
integration approaches this topic briely, but you may also contact your speciic CRM or
OvervIew oI Web AnuIyLIcs 11

LRP endor and asking them how they recommend you integrate the data they supply with
inormation collected rom your \eb site.

Web Analytics Tip: Integrate Forms-based Data
\heneer I`m helping a client implement a traic analysis solution I am always thinking
about how to integrate orms-based data, usually with isitor segmentation in mind. My
thinking is that any strategy I can ind that brings disparate sources o data together ea.it, is
aluable.

Successul orm-based segmentation strategies that I`e implemented in the past include
ariations on demographics ,age, gender, amily income leel, presence o children in the
household, indicated preerences, indicated likelihood to purchase, and son on,.

Most adertising sales groups would rather tell potential adertisers that they had 8,000
isitors in the male, 18-to-25-year old demographic segment isit content group A` last
week and these isitors generated a combined total o 15,000 page iews` rather than we
had 2,000 isitors to content group A` last week who generated a total o 43,000 page
iews, and many o those isitors are in the 18-to-25-year old male demographic you are
targeting.`

Where This is Going
As mentioned briely aboe, the ocus o the balance o !eb .vat,tic. Dev,.tifiea will be
traic and transactional data, with perormance data included as appropriate. \hy:

1raic analysis applications, such as \eb1rends amily o products, lireclick`s
Netlame and een \usage, are ar and away the \eb analytics applications
that most online business are already amiliar with and likely own.
\hile most online business already own some type o traic analysis application,
ew o these businesses are likely using these applications to the ull extent possible,
and many are not using these applications at all ,een though they hae already
made signiicant inestment in said sotware,.
O all the data types listed aboe, traic and transactional are typically the most
oluminous and by extension potentially most useul or understanding the wide
ariety o isitor interactions possible on a gien \eb site.
It is the author`s belie that the greatest possible gains or the least possible cost
online reole around deeloping a irm understanding o what traic and
transactional data can tell us. \hile it can be prohibitiely expensie to
continuously conduct usability studies, and you risk damaging the relationship you
hae with your online isitors to surey them again and again, traic and
transactional data is collected quietly behind the scenes. 1his quiet and constant
data collection enables a undamental \eb analytics strategy, the continuous
improement process.
1 CIupLer z

The Continuous Improvement Process
1he continuous improement process, sometimes reerred to as the continuous
improement methodology or simply /aiev, is the most undamental component o a \eb
analytics program or your online business. It is an idea that has been used inormally or
some time, but has more recently been better elucidated. 1he basic tenet o continuous
improement, and by extension, \eb analytics, is that vea.vrevevt ritbovt tbovgbtfvt actiov i.
ra.tefvt. \es, it is important to hae inormation about how many people isit your \eb site
and what pages they look at when they isit but i that is all you get rom your analytics
package you will soon be asking yoursel why you made such a signiicant inestment. Jim
Sterne, one o the godathers o \eb analytics, clariies the need to measure and act quite
well in his book, .aravcea vait Mar/etivg:

1he secret to truly successul marketing is actiovabte vea.vrevevt. Measuring your
results alone isn`t enough, the key is eeding them back into your sales and
marketing processes to make continuous improements. Many marketing projects
ail in this regard.`

Supporting Sterne`s comment about marketing projects ailing to act upon what they
measure, is a recent ClickZ article by Jason Burby o the \eb business consultancy ZAAZ,
Inc,. citing the ollowing three reasons that analytics ail companies :

1. 1he goals or the analytics program are ill- or undeined.
2. Inormation generated rom analytics projects are ineectiely shared within the
organization.
3. Companies don`t take action based on the data they collect.

,See the original article on Clickz at http:,,www.clickz.com,res,analyze_data,article.php,
330121.,

lortunately, continuous improement is an established ramework to help companies aoid
all three o these problems. lundamental to the continuous improement process is the
notion that no changes are made on the \eb site without a reason or making the change
and expectations about what eect the changes should produce. By tying continuous
improement to the key perormance indicators discussed in Chapters 11 through 15 o this
book, the eects o changes can be obsered by indiiduals throughout the organization. I
you can conince your company to truly use this process you will neer hae to make
another useless measurement on your \eb site again.
The Process, Defined
1he ollowing ie steps deine the continuous improement process:
Define
lirst establish the key actiities and actions on which you want to ocus your eorts. It is
impossible to state that you want to simply increase traic to your \eb site` and
OvervIew oI Web AnuIyLIcs 1

successully work towards that goal. It is much more useul to deine a goal o increasing
isitor traic by 10 percent and isitor conersions by 50 percent rom June to July by
reining the marketing message deliered in our email campaigns.` 1he ormer is ague and
you will neer be 100 percent sure what changes aected an increase or decrease in traic.
1he latter is ery speciic, deining which parts o your site will change ,email campaigns,
and the expected changes ,10 percent increase in traic, 50 percent increase in conersions
rom June to July,.
Measure
1aking baseline measurements is the only way to understand whether any o the changes you
implement are successul. Using the aboe goal deinition as an example one would start by
recording the percentage-wise response and conersion o isitors arriing rom email
campaigning actiity in June. Ideally you will take these measurements and roll them into a
key perormance indication program similar to those described in Chapter 15 o this book.
Rolling these measurements into a regular report that is shared within the organization is
critical to ensuring that eeryone understands the changes you are about to make and that
they are able to see the eect o these changes.
Prioritize and Change
As there are oten seeral dierent solutions to a gien problem, as well as seeral dierent
component problems, it is smart business to assign priority to each component and potential
solution so that resources can best be allocated. Continuing with the aorementioned
example, it is reasonable to assume that two possible explanations or response being low are
poor targeting and poor messaging. \hile it is sometimes diicult to ind lists o
appropriate email addresses or blind campaigns ,in other words, spam, it is reasonable to
create three dierent messages and test the indiidual eicacy o each message.

It is this step that is nearly always the most time-consuming, the point where you hae to
actually sit down and create new wire-rames, design new pages, consult existing personas,
work with outside agencies, and so on. In order to make this step as painless as possible you
should consider breaking changes up into reasonable chunks` o work. I people start
making statements like, \e should redo the entire \eb site` or \e should hire an entirely
new adertising agency` in order to reach the deined goals, perhaps eeryone should take a
deep breath. 1he chances that you will be able to measure the eects o large changes like
these are not good. More oten than not when a complete redesign occurs all you do is
create a whole new suite o problems that need to be addressed.

Instead it is much better to try and break the perceied problem down into as many small
actions as possible and tackle them in order o likelihood o returning the desired eect.
Doing so is not nearly as diicult as it sounds as long as you keep the ollowing in mind:

Yov ritt ovt, be abte to trvt, vvaer.tava tbe effect of a cbavge if ,ov va/e ove cbavge at a tive.
Oftev it i. tbe .vatte.t tbivg tbat ba. tbe vo.t .igvificavt effect.

1q CIupLer z

\eb analytics is not rocket science. On the contrary, \eb analytics done well is more like
priate inestigation - by looking closely or clues and orming micro-hypotheses that can be
easily tested, oten times the larger problem is soled with a minimum o eort.
Verify
Beore you deploy your changes to the entire online unierse and risk making costly mistakes
it is wise to test your changes against a smaller audience and eriy that results are isible. In
this example it would be wise to send each o the three dierent emails to a subset o your
blind audience to test that response rates hae not gone down compared to the baseline data
measured in June. I you are making changes to your \eb site proper, it is strongly
recommended that you run A,B tests such as those described in Chapter 6 and make sure
that A, you can see the eects o the changes and B, that the eects are signiicant enough
to be propagated throughout.
Analyze
1his is the inal step, taking measurements to determine how successul the changes you
made actually were. It is this step that allows you to reine uture business strategy,
essentially learning rom your successes and ailures. Proided that you hae successully
eriied that a positie eect can be obsered based on the changes and hae rolled the
changes out to your entire site or audience, you then want to continue to measure their
eect oer time. 1he analyze step would essentially be the ater` reporting in your key
perormance indicator ,KPI, report and the percent change would then be compared to the
desired goal described in the irst step in this process to measure success or ailure.
Strategies for Success
As you can imagine this is a cyclic process - as soon as you get the results o your analysis it
is time to deine the next set o changes based on eoling business goals. A ew strategies
or success when employing the continuous improement process include:
Think Micro, Not Macro
Put another way, don`t bite o more than you can chew. A well-established \eb analytics
program will be continuous in its own right and there will be plenty o time to put all o your
business goals through this cycle. 1he biggest mistake businesses make is trying to change
too much at once and then not being sure what actually caused any obsered results. Gie
yoursel a ighting chance! 1here is more than enough ariability inherent in the Internet
itsel without adding more complexity to the mix.
Test for a Reasonable Amount of Time
\ou don`t want to spend too much time on any one step, especially the analyze` step.
1here is a tendency to keep watching, measuring and waiting in the hopes that whateer you
are watching will magically get better` on its own. 1he problem with this is that the eect
OvervIew oI Web AnuIyLIcs 1g

o most changes is seen quickly, especially on high traic sites, and so waiting and watching
essentially becomes opportunity lost. Keep the pressure on to igure out what combination
o low and presentation works best or your customers, especially i you are trying to aect
abandonment and conersion rates by changing low and design through a speciic process.
1his is not to say change your site eery day to try and ind the magic ormula. Be judicious
about making changes, your business model and existing inrastructure will likely constrain
your ability to make and deploy changes. Still, you want to ight the temptation to only
make changes once a quarter, hoping that your deined goals will be met sometime in the
next year. 1he key perormance indicators discussed in Chapter 15 will likely be extremely
helpul in determining when a reasonable amount o time has passed and that the results you
are seeing are the results you are going to continue to see.
Keep Careful Track of What Works and What Doesnt
1he single most wasteul act in continuous improement and \eb analytics is making the
same mistake oer and oer again. Assign someone the role o gatekeeper and ensure that
he or she has intimate knowledge about changes that hae been made in the past and the
eect o those changes. I the author had a nickel or eery time he heard someone say
hey, isn`t this the same change we made to this part o the site last year` he wouldn`t need
to write this book or work.

It is diicult to emphasize enough the importance o making use o the continuous
improement process in your \eb analytics program. Many o you will likely recognize the
close relationship between continuous improement and the Six Sigma process, more o you
should realize that continuous improement and kaizen are only more interesting terms to
describe the scientiic method - orming and testing hypotheses by collecting data and
objectiely analyzing said data. Most, i not all, o the greatest gains o humanity hae been
realized by some application o the scientiic method, is there any reason to expect that your
\eb site cannot beneit rom the application o a strategy this tested and aluable

Very ew online businesses are ortunate enough to simply stumble upon the correct
combination o language, presentation and low to maximize proits or optimize the isitor
experience. 1ruly successul businesses apply a methodology like continuous improement
or Six Sigma to systematically test ideas, looking or those that are good and erreting out
those that are less good.` I you take nothing else rom this book at least consider applying
the continuous improement process to whateer \eb analytics you already do.
16 CIupLer z

Web Analytics Tip: Applying the Continuous Improvement
Process
\ith eery new business I work with doing any kind o \eb analytic work, be it training,
implementation or analysis, I always spend time helping them to institute the continuous
improement methodology. 1he most important thing or eeryone to know, or at least
anyone who is serious about their online business, is that the Internet is not a static entity.
Customer preerences change, the competitie landscape changes, technology changes,
eerything changes. I you accept the act that eerything changes and that the Internet is a
dynamic entity you can immediately see the alue in being aggressie about implementing
and testing changes on your \eb site.

Case in point: One customer I work with had been requiring registration to moe through
their purchase process. 1hey were doing so because other companies did it and they were
reported to be ery successul ,I oten reer to this as the Amazon syndrome`,. \hat this
customer didn`t stop to consider was the negatie eect this would hae on isitors who
were unsure they would eer make a repeat purchase, people want to make sure that you`re a
good endor rom end-to-end beore they`re willing to commit to haing a relationship with
you.

1o make a long story short, we implemented the continuous improement process to test
the theory that remoing the registration step would hae a positie impact on the site`s
oerall conersion rate. 1urns out we were right-when they remoed the registration step,
making it optional and opt-in, the site-wide conersion rate went up more than 20 percent.
Lncouraged by this simple, but ery signiicant, win the customer kept applying continuous
improement and was able to make additional gains without outside support or
encouragement.

In addition to the continuous improement process, I also encourage my customers to track
key perormance indicators ,KPIs, such as those described in Chapter 15 o this book. 1he
combination o KPI reporting and the continuous improement process proide a
structured and powerul strategy or improing nearly any critical metric o online business
success.



TIe DIIIerenL Web TruIIIc DuLu Sources 1,

CHAPTIR
IIIIRINT WIB TRAIIIC
ATA SOLRCIS


With so much data available, what's the most important thing to measure? 1he
most important metrics is that which helps improve your site.
Jim Sterne in Web Metrics


One o the most unique things about !eb .vat,tic. Dev,.tifiea is that this is the irst book
known to its author that proides a moderately technical comparatie analysis o \eb-serer
log iles and the more recently emerged client-side page tags. 1hese page tags, sometimes
reerred to as web bugs,` hae been touted in 2003 as the uture o \eb analytics data
collection` and analysts like Matthew Berk hae declared that the \eb analytics war is oer
and page tags hae won the battle.` \hile statements such as these may be premature it is
important to understand the strengths and weaknesses o the common data sources or \eb
analytics measurement.
The Two Most Common Sources plus One
Uncommon One
1he history o \eb traic analysis typically begins with the sentence in the beginning there
were \eb serer log iles, and \eb serer log iles were good.` \hile it sounds like the
beginning o a bad joke, this is pretty much the truth-\eb traic analysis began with the
serer log ile. Because o this, \eb traic analysis is oten reerred to as log-ile analysis,`
which is oten no longer the case. \eb1rends Corporation o Portland, Oregon ,now
NetIQ,, did a great deal to promote log ile analysis, originally selling their Log Analyzer
product to the many emerging \ebmasters o the world during the Internet boom o the
1990s. Since we`e already stated that !eb .vat,tic. Dev,.tifiea is not a history book, let`s just
leae it at that.

At some point the engineers at \ebSideStory o San Diego, Caliornia, obsered that
JaaScript could be leeraged on indiidual \eb pages to transmit interesting inormation
about the page a isitor was iewing and the browser application they were using to iew the
page. In essence, the JaaScript would build a series o ariables populated with a snapshot
o isitor and page inormation and then pass that inormation to a data center ia a small
gi` image. 1he combination o JaaScript and the image request was initially reerred to as
a page tag` and page tagging started the era o client-side traic data collection.

1S CIupLer

At this point in time, log iles and page tags are the two most common \eb traic data
sources globally. Log iles still enjoy the widest usage, thanks in large part to the hard work
o \eb1rends sales orce and distribution channels through the mid-90s. loweer, page
tagging is rapidly becoming more popular, thanks in part to the application serice proider
,ASP, model that allows or aster implementations and lower in-house costs o ownership,
and the number o business sites employing client-side data collection grows daily. Both o
these data sources will be explored in great depth in this chapter.

1here is a third source o data aailable to \eb analytics reporting, that o network data
collection, which was Pioneered by Accrue Sotware, Inc., in 1996 and later copied by
Andromedia in their ARIA product and Marketwae in their lit List amily o products. A
network data collector, in the words o Bob Page, ounder and C1O o Accrue Sotware, is
the moral equialent o a protocol analyzer, except it`s tuned or l11P, and built or olks
who are content deelopers and marketing executies, instead o protocol engineers.` Put in
perhaps more riendly terms, Accrue`s network data collector sat on the same network as the
\eb serers and passiely watched data low between the \eb serer and the \eb clients
,that is, isitors` \eb browsers, making requests.

Some o the purported adantages o network data collection included the ability to measure
how oten isitors were hitting the stop` and reload` buttons in their browser, the ability
to measure a isitors eectie Internet connection speed including how long it took or an
entire page to load ,l1ML plus all component objects, and the ability to determine how
much o a \eb page was actually transmitted to determine whether pages were loaded
completely into the isitor`s \eb browser. Vendors not supporting the network data
collection oten complained that the model was not scalable and that it would hae serious
perormance issues. \hile scale is a potential problem or any analytics product, Bob Page
related to this author in a personal email that the Accrue G2 product was measuring 1.5
billion hits` a day using network collection.

\hile each o these does seem to be adantageous, the network data collection model neer
really took o in a big way. In the irst wae o consolidation, around the approximate time
that NetIQ purchased \eb1rends, Inc., Accrue purchased Marketwae and Macromedia
purchased Andromedia ,Macromedia has since ceased supporting the ARIA product,. More
recently, as Accrue`s stock price ell and stayed below the >1.00 per share minimum or
membership on the NASDAQ exchange, Accrue declared bankruptcy and was purchased by
a new company, Datanautics, which has been reported to be made up completely o the
preious owners o Accrue. \hile this author has little direct experience with network data
collection, it is sae to say that he is not alone in the world and that rer, fer people hae
direct experience with network data collection as a \eb analytics data source.
Focus on Web Server Log Files as a Data Source
According to the most recent surey by the Aberdeen Group, companies selling log ile
analysis solutions share about 41 percent o the aailable reenue in the \eb analytics
market. Companies in this market segment include the \eb1rends diision o NetIQ, SPSS
,haing purchased NetGenesis in 2001,, Accrue, IBM`s SurAid, Reenue Sciences and a
handul o others. 1he 41 market share is down signiicantly rom the 1990s when log-
TIe DIIIerenL Web TruIIIc DuLu Sources 1q

based solutions were the only game in town.` \hen the Internet craze o the mid-90s hit
and business ater business clamored oer themseles to get online, it became obious that
some solution was needed to help understand what isitors were doing on these \eb sites.
At the time, the only easily aailable data source was the \eb-serer log ile.

A \eb serer exists to do essentially one thing-sere resources to requesting clients.
Resources are objects like l1ML pages, images, multimedia iles, script iles and the like.
Requesting clients are most oten \eb browsers being used by people but can also be
automated agents like search engine spiders, perormance monitoring applications and other
related applications. A \eb serer is successul` as long as it is able to respond to all
requests or inormation in a timely ashion without losing track o requests. Problems
occur with \eb serers when try to respond to too many requests at one time, the most
common result in this case is a ailure to respond. 1o keep track o the how, when and
why` these problems occur, \eb serers oer logging capabilities.

1he \eb-serer log ile is roughly what it sounds like, a text ile that is written as actiity is
generated by a \eb serer. According to the olks at Apache, whose serers currently enjoy
more than 60 percent o the market share, in order to eectiely manage a \eb serer, it is
necessary to get eedback about the actiity and perormance o the serer as well as any
problems that may be occurring.` Microsot, whose Internet Inormation Serer ,IIS, holds
roughly 30 percent o the market share hae this to say: Microsot ,Internet Inormation
Serer, uses log iles to track inormation about eents that occur when users access IIS
\eb sites and applications. Inormation such as the number o isitors to your site or
application, the number o bytes sent and receied, and the reerring page is inaluable or
\eb managers, who can examine the data to discern trends and patterns.`

1he log iles are able to collect a ariety o data about requests or inormation coming to
your \eb serer. 1he most common pieces o inormation collected are the Internet
Protocol ,IP, address o the client requesting the resource, the authenticated user name o
the requestor ,i known,, the date and time o the request, the method o the request and the
resource being requested, the status code sent back to the client and the size in bytes o the
object sent. A typical log ile entry in common log ormat ,CLl, looks like:

127.0.0.1 - frank [10/Oct/2000:13:55:36 -0700] "GET /apache_pb.gif
HTTP/1.0" 200 2326

Most, i not all, currently aailable \eb serers oer additional ,or extended, logging
capabilities. A complete list o data aailable to common and extended logging includes:

Request Property Lxplanation
Resource requested 1he actual ile being requested ia l11P ,l1ML, GIl, JPG,
PDl, and so on,
Date Date o the request based on the serer time
1ime 1ime o the request based on the serer time
Client IP Address IP address o the browser making the resource request
Reerrer 1he URL containing the link to the resource being requested.
1his is usually only aailable i a link is actually clicked, although
not always
User Agent 1he browser or application making the resource request
o CIupLer

Serice Name Name o the serice, such as, \eb serer application
Serer Name Name o the \eb serer sering the resource
Serer IP IP address o the \eb serer sering the resource
Method GL1 or POS1. Both are methods or sending inormation to
scripts running on the \eb serer and or requesting documents
URI Query 1he query string appended to the end o the URL. Not always
present but typically used to pass inormation to dynamically
generated \eb pages
l11P Status Numerical alues describing the \eb serer response. Common
codes include 200` ,ok, and 404` ,ile not ound,. A
complete list can be ound at
http:,,www.w3.org,Protocols,rc2616,rc2616-sec6.html
\in32 Status Similar to l11P Status but reerring to Microsot`s proprietary
list o codes
Bytes Sent 1otal bytes transerred by the \eb serer
Bytes Receied 1otal bytes receied by the client browser or application
1ime 1aken 1otal time taken rom request to deliery
Protocol Version l11P protocol ersion used in request and transer
Cookie 1ext to be written to browser cookie, typically used to store
identiying or session state` inormation
1able 2: Information commonly collected by Web server log file extended file format.

1his example contains inormation in the combined log ile ,CLl, ormat with the addition
o reerrer and user-agent

lothlorien.ncsa.uiuc.edu - - [19/Sep/1995:15:19:07 -0500] "GET
/images/icon.gif HTTP/1.0" 200 1656 "http://hoohoo.ncsa.uiuc.edu/"
"NCSA_Mosaic/2.7b1 (X11;IRIX 5.3 IP22) libwww/2.12 modified"

An example rom a Microsot IIS Log ile ormat collecting user`s IP address, user name,
date, tie, serice name, serer name, serer IP address, time taken, bytes receied, bytes sent,
status code, \in32 status, request type and the resource being requested:

172.16.255.255,anonymous,03/20/98,23:58:11,MSFTPSVC,SALES1,192.168.114.201
,60,275,0,0,0,PASS,intro.htm

As you can see, dierent \eb serers oer slightly dierent data points and collect the
inormation in dierent ormats. Keeping track o this inormation or erer, request coming
to a \eb serer can quickly add up in terms o disk space. Because o this, \eb serers
oer the ability to both selectiely collect data and to rotate logs on a regularly occurring
basis. Selectiely collecting data is a good thing, and log rotation is simply a act o lie.

Consider 1able 2, describing the data aailable to extended logging. \hile it is nice to hae
some o this inormation, one must consider the question, \hat is the business alue o
knowing the \in32 Status o a request` More oten than not the answer to this question is,
\hat is a \in32 Status`-which is another way o saying that there is no business alue.
It is important to remember that logging does not come or ree and the more you collect,
TIe DIIIerenL Web TruIIIc DuLu Sources 1

the more you hae to process. \ith this in mind, here is an assessment o the business
necessity or each o the data points described in 1able 2.


Request Property Value Lxplanation
Resource Requested \es 1his is the actual document that the isitor is
requesting
Date \es 1he date o the request is necessary to put the
request into a temporal context
1ime \es 1he time o the request is necessary to put the
request into a temporal context
Client IP Address Maybe Oten used to determine uniqueness o isitor,
cookies are better but require additional set-up
Reerrer \es Knowing what \eb sites are linking to you and
sending you traic is a key component in
understanding reach and acquisition
User Agent Maybe I reporting end-users are deelopers responsible
or ensuring that \eb site is tailored or most
common browser then yes,` otherwise likely no`
Serice Name No Non-actionable, uninormatie inormation
Serer Name No Non-actionable, uninormatie inormation
Serer IP No Non-actionable, uninormatie inormation
Method No Non-actionable, uninormatie inormation
URI Query Maybe I your \eb site makes use o dynamically
generated pages and passes inormation to these
pages in the URL then yes,` otherwise, no`
l11P Status \es lor tracking errors and broken links.
\in32 Status No Non-actionable, uninormatie inormation
Bytes Sent Maybe I your analysis application is able to compare the
dierence between bytes sent and bytes receied
then yes`, otherwise no`
Bytes Receied Maybe I your analysis application is able to compare the
dierence between bytes sent and bytes receied
then yes,` otherwise, no`
1ime 1aken Maybe I your analysis application is able to report on the
aerage time taken or clients to download
resources, or een better the distribution o these
times, then yes,` otherwise, no`
Protocol Version No Non-actionable, uninormatie inormation
Cookie Maybe A much better strategy to determine the
uniqueness o a isitor is to use cookies ,rather
than IP address, but cookies require additional set-
up. I your serer supports cookies as a unique
user identiier ,UUID, then yes,` otherwise, no`
1able 3: Business necessity and value of data available via Web server log files.

CIupLer

In general, wheneer you are going to collect data it is adised to collect only the data you
are sure you will need to answer the speciic questions being asked. Many data analysts
subscribe to the notion that eery possible data point should be collected so that dierent
hypotheses can be tested ater the act. 1his is ine i you are conducting an experiment that
will be diicult or costly to reproduce ,such as sending people into space, but neer the case
when collecting \eb data. Changing log ile ormats is ery simple and can typically be
done in a matter o minutes. Reconiguring log ile analyzers can be slightly more
challenging but is not rocket science. Since the Internet is always on, i you add some data
points to your collection it should only be a matter o hours or perhaps days until you hae
enough releant data to study.

Remember that a undamental aspect o the continuous improement process, and thereore
\eb analytics, is to deine and measure. By careully deining the question you hope to
answer you will ery likely also deine the data that you need to collect. I you only collect
the data you need to answer any gien question you will sae yoursel the trouble o haing
to wade through meaningless data, thereby decreasing the time required to analyze and
increasing your likelihood or success.
Advantages of Web Server Log Files as a Data Source
\hen traic analysis applications were irst being deployed by online businesses, the
adantage o using \eb serer log ile data was that it was the only data source aailable
other than network data collection. \ith the emergence o client-side data collection online
businesses now hae a choice as to which data type they will use. Some adantages o using
\eb serer log iles as a data source include:
Ownership of Data
Since your \eb serer is creating the log ile as requests are being sered, the data is
collected and contained on your own machines. Regardless o the location o these serers,
unless your online business is sered rom a shared hosting enironment ,see disadantages,
below, your log iles can be stored on the same network sering your \eb pages, and so on.
\hile it is your responsibility to rotate, backup and maintain these iles, you theoretically
always hae access to the logged data and een i you switch reporting applications you will
likely be able to analyze historical data rom said log iles.
Flexibility of Data Collection: Types of Data
As preiously mentioned, your \eb serers can be instructed which data to collect and
which to ignore. I you hae determined that all you need to answer your undamental \eb
analytics questions are the reerring source, requested object and date,time o the request
then you can simply instruct your \eb serer to log only that inormation. Most \eb
serers een gie you the ability to log only certain types o requests, such as those or
l1ML documents but not or images or scripting iles. An adantage this lexibility
proides is the ability to keep track o non-l1ML pages downloaded ,such as PDl,
multimedia or executable iles, as well to track dierent types o serer errors ,500s, 404s,
and son on, and some serer-based redirects.
TIe DIIIerenL Web TruIIIc DuLu Sources

Ease-of-Implementation
Unlike page tagging there is no need to instrument eery page on your \eb site to take
adantage o a log ile solution. Since request logging is part o the \eb serer process, not
the pages or objects being sered, no additional setup or implementation is required to
create \eb serer log data.
Ability to Log Directly to a Database
Some \eb serers ,such as Microsot`s Internet Inormation Serer, allow you to log
requests directly to a database application. 1he adantage o doing this that i you hae
smart SQL deelopers you may be able to answer many o your \eb analytics questions
without needing an expensie analysis application. Since logging directly to a database can
be resource-intensie it is not recommended to use this method in high-olume request
situations. In these situations it is likely better to batch import \eb serer log data into the
database on regular interals.
Ability to Determine if Downloads Complete
One adantage \eb serer log iles hae is that they are able to report on the total o bytes
sent` when a request is made. Log-ile-based applications are able to take adantage o this
inormation and report on the percent o requests that successully downloaded. lrom this
inormation it is possible to determine roughly how oten isitors A, click the browser`s
stop` button or back up beore the page download is completed and B, successully
download non-l1ML objects such as executable iles and Adobe PDl documents. Many
companies in the business o proiding applications or content hae a ery real need to
know not only how many people begin to download a document but also how many o
those downloads complete.
Ability to Measure Traffic from Robots and Spiders
Because non-human traic is not usually excluded by deault, most sotware solutions using
\eb serer log iles as a data source also proide reporting on which robots and spiders are
isiting and with what requency. \hile it is extremely important to exclude this traic rom
regular reporting about the use o the \eb site, it can be ery useul in helping to determine
the eect o search engine optimization ,SLO, projects. 1he goal o SLO projects is to
attract more search engine-indexing agents and hae them traerse the site deeply and
broadly. laing this inormation can help the online business quantiy the eect o SLO,
especially when the work has been outsourced, and make decisions about uture SLO
eorts.
Disadvantages of Web Server Log Files as a Data
Source
An historical disadantage o \eb-serer log iles, and one that does continue to plague the
highest-olume \eb sites, is disk space. I you are collecting all aailable data, and your
q CIupLer

number o daily page requests is up in the millions, the necessary disk space to keep daily
logs can be quite large. In this instance the online business would likely need to hae a
structured program or moing logs on a regular basis-which can then impact the ease-o-
reporting. lortunately, most online businesses aren`t aced with this olume o traic and
since the cost o disk space and disk I,O has come down signiicantly in the past ew years,
storage is less o a disadantage.

Still, \eb-serer log iles as a \eb analytics data source do hae some signiicant
disadantages worth noting:
Proxy Caching
Beore the broadband reolution` began the most pressing issue aced on the Internet was
long, rustrating download times. In the home enironment dialing up to the Internet using
a modem, users would constantly, and justiiably, complain about how long it took to get
inormation. An early solution to this was caching-keeping oten requested inormation in
a location that was closer to the end-user to speed the deliery time. Proxy caching is a
technique employed by Internet Serice Proiders ,ISPs, to keep requently requested
content on their own machines which are ery close to the user, as opposed to the content`s
original site which can be ar away` ,on the Internet, ar away is both relatie and real,.
Proxy caching in essence takes a snapshot o a \eb page when it is requested the irst time
and keeps that image or a set period o time. I another user requests this page beore it
times out` then the page is sered rom the proxy serer to speed deliery. Proxy serers
are a great idea really, but the negatie impact on \eb serer log iles is that the request or
content neer actually comes through to the \eb serer. Proxy serers are less o an issue
or dynamically generated \eb pages, which typically are deliered with a no cache`
directie.
Browser Caching
Related to proxy caching is browser caching, or the ability or your \eb browser to store
requently or recently iewed inormation on your computer`s hard drie or speedy
retrieal. 1he most requent use o the browser cache is demonstrated by using the back`
and orward` buttons in one`s \eb browser. \hen an end-user isits a \eb page, they get
that page and the objects it is composed o ia the Internet ,whether the page and objects
are deliered rom the businesses \eb serer or a proxy cache is impossible to know,. As
the user clicks through a \eb site, each o these pages and their components are usually
stored locally on the user`s hard drie. \hen a user then clicks the browser`s back` button,
the page is quickly loaded rom the hard drie, saing the necessity or another request
through the Internet. 1he end result to the \eb serer log ile is lost inormation. I, when
the user backs through a site in the attempt to ind inormation, the requests or these pages
are neer een sent to the originating \eb serer, no record can be created and inormation
is lost. Unortunately, the use o the back` and orward` buttons is common among
Internet users and the knowledge o how isitors truly naigate is commonly inisible to the
\eb serer log ile.
TIe DIIIerenL Web TruIIIc DuLu Sources g

IP Address as a Unique Identifier
As mentioned aboe, one element that is always aailable to the \eb serer log ile is the IP
address o the client making the request or inormation. One would reasonably expect that
this IP address would then be a good identiier to help determine the uniqueness o a isitor
on a \eb site, right \rong. 1he same proxy serers mentioned aboe that speed
inormation along to end-users are also oten used to pass requests or inormation along to
\eb serers. 1he end result is that many dierent users in dierent geographic regions may
be identiied by the same IP address. 1he most requently asked question that arises rom
this issue is, \hy are 40 percent o my isitors rom Virginia` 1he answer is, Because
the AOL proxy serers are located in Virginia and they hae some 35 million users on the
Internet.` According to the AOL \ebmasters, AOL Members` requests or Internet
objects are usually handled by the AOL Proxy system. \hen a member requests multiple
documents or multiple URLs, each request may come rom a dierent proxy serer. Since
one proxy serer can hae multiple members going to one site, \ebmasters should not
make assumptions about the relationship between members and proxy serers when
designing their \eb site.` I you look closely at the AOL proxy and client IP addresses, you
will discoer that most o their 35 million users are connecting to \eb sites ia roughly 50
IP addresses. Multiply this problem by the number o multi-user portals working to build
market share currently and you will start to realize that IP addresses can be a dangerously
misleading indicator o the uniqueness o a \eb site isitor.

1hese three issues are the most serious disadantages o using a \eb-serer log ile as a data
source or your \eb analytics program. Some estimate that the inormation loss as a result
o proxy and browser caching approaches 40 percent or more ,meaning you would
undercount traic to your \eb site by 40 percent,. 1he inormation loss as a result o using
IP addresses to determine the uniqueness o a isitor is a unction o the olume o traic
your site receies rom proxy serers, which can be measured, but is requently quite high
depending on your business model.

1he irst and third issue can be oercome with careul and thoughtul implementation when
setting up your \eb serer and \eb analytics reporting application. 1he second issue, to
the best o the author`s knowledge, is impossible to oercome without page-by-page
instrumentation, discussed below under page tagging as a \eb analytics data source.

Some additional but slightly less serious disadantages o \eb serer log iles include:
Robots and Spiders and Crawlers, Oh My!
1raic rom search engines is something that most \eb managers strongly crae. laing
your \eb site appear at the top o the irst page o search results or a particular search term
is something that a company can spend thousands o dollars to achiee. 1he basic strategy
or achieing this success is to create search riendly` pages and submit them to the search
engines or indexing. It is this indexing process that can cause problems or \eb serer log
ile analyzers. 1he indexing applications, oten called crawlers` or spiders` or robots`
but generically reerred to as non-human user agents,` make requests to your \eb serer
or content to index. 1he issue is that, while these requests show up in \eb-serer log iles
as nearly identical to the same requests that normal human` isitors send, they are not
6 CIupLer

traic that most businesses want to measure, at least not as part o a \eb analytics program.
1hese user agents can be excluded rom the \eb serer log ile analysis ,see best practices
.` below, but i they are not, depending on the type o \eb site, robots and spiders can
signiicantly inlate traic and page request numbers.
Report Generation Time
Because \eb-serer log iles can grow quickly, especially when not instructed to limit the
data that they collect, it can also take signiicant time to parse these iles and generate
necessary reports to inorm the \eb analytics process. Some endors hae discoered ways
around long reporting times, but or many the wait to generate reports or months or years
worth o log iles can be days or weeks. Another common issue is the time it can take to
physically moe the log iles rom the machine they`re being created on to the machine
they`re being analyzed on.
Multiple Log Files from Multiple Servers
As traic to \eb sites grows it is common to sere requests rom more than one physical
computer. \eb traic load balancing deices are able to distribute requests in a ariety o
ways, the most common being the round robin,` but can also include logical distribution
based on the type o request or the actual processor load on the \eb serers at the time o
the request. 1he end result in a multi-serer enironment ,oten called a cluster` or serer
arm`, is that \eb-serer log iles or the exact same time can be housed on physically
dierent machines. Consider cases where \eb sites are mirrored geographically-that is,
multiple \eb serers sering the same inormation rom separate geographic locations-and
the problem increases. In order to get a clear picture o the traic to a \eb site at any gien
time you need to combine all o these resulting log iles into a single ile. Additionally,
depending on the strategy that a load balancer uses to distribute traic, a single isitor`s
click-stream can be spread across multiple \eb serers. I the online business hopes to
examine click-stream data they will deinitely need to stitch the multiple log iles together
and ensure that the clocks on each machine are exactly in-sync, i they are not, the traic
analysis application will not be able to reassemble the exact order o requests or inormation
rom the end-user. Depending on the exact goals o the \eb Analysis being perormed it
may be acceptable to only analyze data rom one or ew machines, but more oten than not
this strategy can lead to misleading conclusions and is not recommended. lortunately most
log ile analysis applications hae tools to manage multiple log ile situations but the solution
almost always comes at a cost-longer report generation time.
Access to Web Server Log Files
Some smaller online businesses, especially those just getting onto the Internet, contract with
ISPs who host many dierent \eb sites on a single machine through a single \eb serer.
In low-traic olume situations this is oten a cost eectie solution, one that makes good
oerall sense. 1he downside can be that many o these ISPs will not log traic data to
separate iles or each business, either because they are unable or unwilling to do so.
Occasionally these ISPs will orce the online business to use a \eb analytics package that
they proide, one that may not be adequate to answer the kinds o questions said business
has. 1he key question to ask your ISP i you are considering shared hosting is, low do I
TIe DIIIerenL Web TruIIIc DuLu Sources ,

access my \eb-serer log iles` \ou want to make sure that they can proide log iles or
your \eb site ,and your site only, ia l1P and that you are ree to analyze them with
sotware o your choosing.
Up Front Costs
1he inal disadantage oten associated with using a \eb-serer log-ile analyzer is the act
that you typically hae to purchase all o the sotware, hardware and expertise in adance.
1his diers rom the ASP model most requently associated with client-side data collection,
where you pay as you go.` 1his may or may not be a disadantage to your online business
but it is one that is requently cited.

\hile it may look like the disadantages o using \eb-serer log iles as a \eb analytics data
source heaily outweigh the adantages, keep the ollowing in mind. Many o the
disadantages can be oercome through careul and thoughtul implementation and your
online business likely already has much o what you need to take adantage o this data
source. It is likely that you hae a spare computer or two around to install whicheer
sotware package you choose and to store archied log iles. It is also likely that you hae
someone on your I1 sta that can help you install the necessary sotware and make changes
to your \eb serer so that you are able to collect the data you need. linally, it is extremely
likely that you already hae \eb serer log iles somewhere on your network that you can
use to set historical baselines and get more amiliar with the analysis package you choose
without haing to modiy your \eb pages in any way.
Best Practices When Implementing a Web Server Log
File Solution
Considering the risks and rewards associated with using a \eb-serer log-ile-based \eb
analytics solution, the author recommends the ollowing best practices` or implementing
such a solution. Most o these recommendations reole around data collection and iltering
and are relatiely easy to implement.
Configure Carefully
Lxtended log iles are able to capture a ariety o dierent data, including serer name,
serice name and serer IP address, which are likely o no use to your \eb analytics
program. Collecting eerything or eerything`s sake will do little more than slow down your
analysis. A better strategy is to careully consider the aailable data and decide which
inormation to collect and which to ignore. Modiying your serer coniguration ile to
exclude data is triial and will cut down on processing time and drie storage space. Consult
1able 3 or help in deciding which ields to include or ignore based on business alue.
Exclude Non-Human Traffic from Your Analysis
\eb-serer log iles cannot, by deault, dierentiate between human isitors and sotware
agents. \hile the goal o any \eb analytics program is to measure the actiity o human
S CIupLer

isitors, ailure to ilter out automated traic will artiicially inlate traic and page iew
numbers. Most applications hae an option to exclude based on the user agent` string that
should be exercised to ensure accurate data collection.
Plan for Multiple Log Files from Multiple Servers
I you incorporate a serer cluster or multi-serer enironment it is likely that recorded
traic is spread out among dierent log iles. Unless you are able to recombine this traic
you will end up with gaps in path or click-stream analysis and inaccuracy in page, isit and
isitor counts. Most log analyzer applications hae options or stitching together` multiple
iles and these eatures should be taken adantage o to ensure data accuracy.
Plan for Rotation, Storage and Backing Up
Most \eb serer log iles are created on a daily basis and gien a unique ile name. As time
passes these iles will likely consume too much drie space and need to be moed elsewhere
or long-term storage. A common problem that results is that i you then would like to look
back at dates contained in archied log iles you will hae to either moe those iles back
into a ormat that your log analyzer can access or deelop a strategy or analyzing archied
log iles.
Employ Cache and Proxy Busting Techniques
It is a good idea to attempt to implement some type o dynamic platorm on your \eb site,
i or no other reason than to increase the likelihood that your site`s \eb pages will not be
cached in proxy and browser caches. 1he l11P,1.1 protocol supports cache control and a
complete description o this protocol and its use can be ound at the \orld \ide \eb
Consortium site ,www.w3.org,Protocols,rc2616,rc2616-sec13.html,. 1he easiest way to
implement this is to use the l11P ML1A tag and setting:

<META HTTP-EQUIV="Expires" CONTENT="0">
<META HTTP-EQUIV="Pragma" CONTENT="no-cache">
<META HTTP-EQUIV="Cache-Control" CONTENT="no-cache">

\hile this method is not entirely perect, and it will cause your isitors some additional
headaches ,eery time they want to see a page it will hae to be downloaded again, it will
result in a more accurate measurement o the inormation requested by your isitors. 1his
recommendation should only be ollowed i a signiicant percentage o your Internet
audience is isiting using a broadband connection, thus decreasing the likelihood o
complaints about poor page load time.
Work With Your Application Vendor
Compared to client-side page tagging solutions, getting a best practices` implementation or
a \eb-serer log-ile-based solution really inoles careul setup and data alidation. Lach
o the current endors selling \eb-serer log-ile analysis solutions will hae speciic tips
and tricks to ensure the highest quality data collection and analysis. Don`t be araid to ask,
TIe DIIIerenL Web TruIIIc DuLu Sources q

or een orce, your solution proider to explain this to you. I they are committed to your
success with their application they`ll be more than happy to help you out.
Exclude Internal Traffic From the Analysis
By deault most log ile analyzers ao vot exclude internally generated traic to the \eb
serers, that traic generated by your deelopers and other employees making use o the
business \eb site. 1his traic should be excluded or segregated out o the inal analysis
because it is not necessarily representatie o your isitor`s interests. Depending on your
particular business model, internal traic can be a signiicant component o the traic to a
gien page or content area.

Most \eb analytics applications will allow you to exclude traic rom reporting and analysis
based on at least the IP address o the isitor. \hile IP address is a poor indicator o
uniqueness, it is oten useul or excluding groups o isitors as most corporate irewalls will
identiy internal users as haing coming through one or seeral unique addresses.
Examples of When to Use a Web-Server Log-File
Application
Proided you take the time to ensure that you are doing eerything you can to increase the
accuracy o data, there are a handul o scenarios when you would likely choose a \eb-
serer log-based solution. 1hese include:
When You Need to Incorporate Historical Data
It is extremely likely that your I1 sta has been collecting \eb serer log iles and storing
them somewhere on your network. 1his does not mean that these iles are clean` or that
the data is in the best ormat or your analysis, but the iles do almost certainly exist. Still, i
you want to start a \eb analytics program o and hae baseline data, or look back at the
eects o changes you hae already made on your site, historical log iles are about your only
option.
When You are Unable to Instrument Your Web Pages
1his will be discussed in the next chapter at great length but a requirement o client-side data
collection is adding additional code to each o the \eb pages you want to track ,hence the
term page tag`,. In some situations it is impossible to do this, in others, impractical. I
your I1 sta can proide a suitable explanation as to why` you cannot implement page
tagging, \eb serer log iles are the way to go. One caeat to this is be sure to ask your I1
sta exactly rb, a page tagging scenario would not work-the worst thing or a \eb
analytics program is haing to work through another department that is unable or unwilling
to help.
o CIupLer

When Ownership of Data is Extremely Important
One clear adantage o \eb serer log iles is that all o the necessary data to conduct an
analysis is yours and is contained in-house. Contrast this with page tagging where more
oten than not the data is collected into a remote data warehouse owned by the endor ,the
exception to this is described in the next chapter under the heading tags into logs`,. I it is
important to your organization to absolutely own the data underlying your \eb analytics
projects then \eb serer log ile are most likely the right data source or you.
When Privacy of Data is Extremely Important
\hile it is ery, ery unlikely, there is always the possibility that a malicious soul could
intercept the transmission o data rom a page-tagging solution and somehow determine the
identity o the person making the request and the object being requested. Most client-side
data collection technologies either encourage their users to not send any personally
identiiable inormation, or i they do-as may be the case in a commerce enironment-
they encourage their users to only pass obuscated user IDs. \eb-serer log iles
circument this issue by simply not transmitting data ,in most cases, but see exceptions,
caeats, and so on,` below,. 1ypically the urthest distance that log iles are transmitted or
analysis is rom one network resource to another without making use o the Internet itsel.
Exceptions, Caveats, and so on
1he inormation described regarding the use o \eb-serer log iles as a data source or
\eb analytics programs or the most part assumes the traditional sotware model, that is,
you own your log iles and you purchase or license an application to analyze those iles on
your own hardware, network, and so on. 1here is another model or analyzing \eb-serer
log iles, that o the application serice. \hile application serice proiders ,ASPs, are most
oten associated with client-side data collection, there are now a number o endors who are
willing and able to grab` your \eb serer logs, among other data sources, and process
them in their own enironments.

1wo o these endors worth mentioning briely are Reenue Sciences and IBM`s SurAid.
Both companies hae built a reputation on being able to combine multiple data sources and
connect them to data stored in \eb-serer log iles to answer complex business questions.
A clear adantage o this type o solution is that the need or hardware, technical support
and sotware licenses is outsourced, lowering the total cost o ownership signiicantly. A
disadantage o this type o solution is that the analysis and reports are typically limited to
what the endor can proide and the cost o integrating multiple data sources is potentially
high. More inormation about these endors and data integration is presented in Chapter 16.
Focus on Page Tags as a Data Source
As mentioned preiously, many companies proiding \eb analytics are now doing so using
a JaaScript page tag` to collect data instead o using the \eb-serer log ile. Because the
TIe DIIIerenL Web TruIIIc DuLu Sources 1

isitor`s \eb browser enables the data collection, this type o collection is oten reerred to
as ront-end` or client-side.` 1he Aberdeen Group reports that in 2002 the market share
o companies employing some type o ront-end` data collection technique grew rom just
more than 12 percent in 2000 to more than 15 percent in 2002. Oten endors proiding
page tags are reerred to as proiding outsourced \eb analytics,` which may or may not be
the case ,see tags into logs` below,. Aberdeen reports that more than 43 percent o the
endors in the \eb analytics market are oering solutions that incorporate both types o
data sources, page tags and \eb-serer log iles. Companies that proide client-side data
collection solutions include \ebSideStory, the \eb1rends diision o NetIQ, Coremetrics,
Omniture and a host o other, smaller companies.

1his market segment is growing quickly or two undamental reasons. lirst, in the case o
completely outsourced endors, the barrier to entry or an application serice proider ,ASP,
is much lower than that or a traditional sotware company. 1he ASP doesn`t hae to build
a sotware product and manage packaging and shipping-eerything is deliered
electronically. Second, companies are beginning to see the alue o outsourced data
collection as a way to lower up-ront costs while still proiding \eb analytics data. Rather
than haing to purchase sotware licenses, hardware, network space and human resources to
manage and maintain the application, the outsourced solution allows companies to pay as
you go.`

It is worth noting at this point the dierences in the use o outsourced` and client-side`
when discussing this particular data type. Oten endors and authors, this author included,
will use these terms interchangeably to describe data collected ia a JaaScript page tag.
\hile the use o the page tag is truly client-side data collection,` the use o a page tag does
not necessarily imply that the analytics program has been outsourced, there are more and
more endors proiding client-side data collection that can be run rom within the
businesses own I1 inrastructure ,see tags into logs` later in this chapter,. A truly
outsourced analytics program` is one in which the data is hosted and analyzed in another
location, usually the endor`s data center. Some endors ,Reenue Sciences, IBM SurAid,
some others, hae become adept at grabbing \eb serer log iles ia l1P and analyzing
them in-house, the deinition o outsourced analytics.` Conusing, yes. 1ruly important
No. All you really need to know in the context o JaaScript page tags is the dierence
between \eb serer log iles and client-side page tags as a data source, regardless o the
business model o the application proider.

lor the most part, companies proiding client-side solutions work like this:

1. \ou sign up or a serice, agreeing to pay some type o up-ront installation or set-
up ees and a monthly cost based on the total olume o traic and transactions to
your \eb site.
2. 1he endor proides you with a set o JaaScript tags` designed to detect
inormation about the isitor. 1hese tags need to be added to any o your \eb site
pages you wish to track.
3. \ou add the code to your site somehow, setting ariables in the code depending on
the exact endor solution.
4. \hen your isitors come to your site, they download your \eb pages, which now
include the page tag.`
CIupLer

5. \hen your pages are downloaded in the isitor`s browser, the page tags are
executed and inormation is sent to the endor`s data center, most oten using an
image request or a ery small image or another small ile, by appending a long
query string to the image request. 1his is the component oten reerred to as a
\eb bug.`
6. 1he data center receies the image request and parses the query string into the
endor`s particular data model.
. 1he image is sent back to the client and the transaction is complete.
8. \ou log into a secure online interace and iew the data in pre-deined reports, or,
depending on the endor, get customized reports deliered directly to you ia
email.

\hile this is a signiicant oersimpliication, undamentally that is how the process works.
1he dierences between endors proiding this type o solution are usually as ollows:
Java Script Variables
Some endors are proud o the act that they rely ery little on post-processing o data and
so their solutions are ery ast and scalable. Other endors are proud o the act that their
tags are ery small and require little or no modiication to unction properly. 1he number
and complexity o in-page JaaScript ariables is something that each endor in the
outsourced analytics space has resoled in a slightly dierent way and there is really no best
way` or this to be done. I the author has any guidance on this subject it would be that the
JaaScript and ariables be A, lexible, so they can be set independently, broken up on the
page or remoed entirely i they`re not being used and B, designed with page size
considerations in mind.
Amount of Data
Again, some endors proide ery lexible solutions that are able to gather many dierent
types o data, including multiple content groups and custom data types. Other endors
proide more rigid data models where only certain data types can be collected. Still other
endors accept the entire query string as a parameter, which can then be parsed and explored
ater the act.
Vendor Data Architecture
lere, some endors hae made a liing by proiding highly scalable but relatiely lat
architectures to store incoming data, oten in aggregate ormat. Other endors collect all
data into a relational database on a session-by-session basis. Still others store data in a
combination o the two, depending on the incoming data type.
Reporting
lor the most part, outsourced \eb analytics endors proide a static set o reports, grouped
logically ia a secure \eb-based application. Lach endor puts their own spin on the data
and organizes the inormation in their own particular way. Additionally, most endors
TIe DIIIerenL Web TruIIIc DuLu Sources

proide some type o custom reporting` to allow data to be combined in dierent ways as
to be more meaningul or your own business.
Price
\ithin the outsourced market it is likely that, depending on the olume o traic and
transactions you hope to measure and data you hope to get back, you will be able to ind
solutions ranging in price rom ree` to ery, ery expensie.` \hile oerall prices seem
to be coming down, the cost per million page iews model o pricing or client-side data
collection is still all oer the board.
Likely Longevity of the Vendor
1his dierence between endors is likely a unction o the downturn in the technology
market at the time this book was being written, while there was a general trend towards
consolidation in the analytics market in 2000 and 2001 that actiity has more recently
slowed. Lery endor will insist that they are doing better than eer` and that their own
long-term outlook is extremely bright but reality dictates that not eery business can be
successul and some endors currently proiding \eb analytics solutions may not be around
in two or three years.

\ith all these actors in mind, the general strategy when getting inoled with an outsourced
\eb analytics endor is to ind one that has reports you can read and understand, a tag you
can implement and a reputation you trust. It is ery easy to see i your competitors or
companies that you respect are using this type o data collection, you just go to a page on
their \eb site and do a browser View Source.` I the company is using client-side data
collection you will see their endor`s JaaScript somewhere on the page. Some common
strings you can search or to eriy this include:

Vendor Search for Lxample URL
\ebSideStory websidestory` www.backcountrystore.com
\eb1rends netiq` or webtrends` or data
collection serer`
www.webtrendslie.com
Coremetrics coremetrics` or webcriteria` or
unitag`
www.ictoriasecret.com
Omniture site catalyst` or omniture` www.ebay.com
IBM SurAid stats.suraid` suraid.dw.ibm.com
VisitorVille VisitorVille` or plgtric.js.php` www.isitorille.com
lireclick hints.netlame.cc` www.ireclick.com
1able 4: Commons strings used in Web page code to identify vendors supplying
page-tagging solutions.
Advantages of Page Tags as a Data Source
As mentioned in the preious section, \eb serer log iles hae long been the ae facto data
source or \eb analytics programs and traic analysis applications. 1his is changing,
q CIupLer

howeer, as the popularity o page tagging and client-side data collection continues to grow.
Some o the adantages o client-side data collection include:
Accuracy
Because data is collected directly rom the end user, not rom the \eb serer, client-side
data collection is typically much more accurate than \eb serer log iles. Most page tags
rely on cookies to determine the uniqueness o a isitor and hae cache-busting code that
ensures that data is collected regardless o where the page was sered. Client-side data
collection is also typically able to measure the use o the browsers back` and orward`
buttons because the code is executed eery time.
Accuracy, Revisited
As mentioned in the preious section, non-human user agents such as search engine
indexing spiders and crawlers can negatiely impact the accuracy o \eb-serer log-ile
analysis applications. lortunately, because vo.t o these user agent applications do not
actually render the \eb page once it has been downloaded, client-side data collection
applications are generally thought to be more accurate than \eb serer log iles. Put
another way, een though your \eb serer sees the request rom the search engine robot,
the robot neer builds the page, does not execute the JaaScript page tag and thereore the
inormation about the request is neer passed along to the data collection application.
Speed of Data Reporting
\hile \eb-serer log iles are arguably a less-obtrusie data collection technique, many
client-side technologies are able to report data in real time or nearly so. Since there is no
need or a log ile that needs to be parsed at a later date, client-side data collectors typically
parse the inormation into their respectie data structures as the data comes in. 1his real-
time parsing usually translates into inormation aailable ia the reporting mechanism a ew
seconds to a ew minutes ater the page iew is seen. \hile real-time data reporting can be
ery aluable in a handul o instances, see the section on About real time`` below to
determine how much o an adantage this may be to your actual business.
Flexibility of Data Collection: Variables
Since this technique depends entirely on a endor-deined model or data collection and
storage, client-side data collection typically allows or a greater range o ariables that can be
collected rom any gien page. Most client-side technologies require ariables deining
page name` and content group` but also allow or campaign ID,` isitor segment,`
product,` category,` brand,` price,` and so on, depending on the application and the
type o page containing the tag. \hile these ariables are ixed rom endor-to-endor, the
lexibility proided by the methodology allows online businesses to deine the type o data
they want to collect and then pick a endor that can accommodate their needs, rather than
simply making due with log iles or building complex workarounds.
TIe DIIIerenL Web TruIIIc DuLu Sources g

Lower Upfront Costs
Since nearly all client-side solutions are deliered ia the application serice proider model
,howeer see 1ags into logs` below,, the upront costs associated with implementing this
type o solution are lower. As mentioned earlier most endors make their customers pay
some type o setup ee based on the speciic solution, number o accounts or type o data
collected and then charge monthly or actual traic measured. 1here is no need to purchase
hardware or hire I1 sta to maintain the hardware and sotware-all o these costs are built
into the price or an ASP solution.
Disadvantages of Page Tags as a Data Source
\hile on the surace client-side data collection appears to be a more accurate, more
inormatie source or \eb analytics projects, there are a handul o disadantages, some o
which can be signiicant. 1he notable disadantages include:
Dependence on JavaScript and Cookies
All currently aailable page tags o any alue rely heaily on cookies and JaaScript.
1ypically, JaaScript is the mechanism to determine necessary inormation about the isitor
,what pages they are looking at, what browser they are isiting on, and so on, and the cookie
is the session-to-session storage mechanism. Additionally, cookies are used to determine the
uniqueness o the isitor. I a isitor has either o these technologies disabled, the quality o
inormation collected is diminished. Most page tags make use o some type o <NOSCRIPT>
code that passes part o the aailable inormation to the data center, but this is most oten
limited. Similarly, most page tags can detect the acceptance o cookies and will report back
on their acceptance or denial but in their absence will report little or nothing about the
isitor`s session and preious actiity on the \eb site.
Dependence on Cookies as a Unique Identifier
\hile this is only loosely a disadantage considering there really is not a better alternatie, it
is worth noting the problems associated with cookies as a unique isitor identiier. Page-tag
endors nearly always pitch their solutions as collecting inormation about human isitors`
when they`re really collecting inormation about \eb browsers. Client-side data collection
assumes the ollowing about isitors to make a one-to-one relationship between isitors and
\eb browsers:

Lery person uses only a single \eb browser to sur the Internet.
Lery \eb browser is used by only a single person to sur the Internet.
People rarely disable cookies when suring the Internet.
People rarely, i eer, clear their browser`s cookie cache.
People rarely, i eer, switch between browser types ,or example, Internet
Lxplorer to Netscape,
People inrequently upgrade their computers entirely.

6 CIupLer

As you can see, seeral o the assumptions listed here hae the potential to be incorrect rom
time to time. 1he most obious exceptions are or items one and two-many people sur
the Internet using one computer during the daytime and another eenings and weekends,
and in certain situations a large number o indiiduals may use the same
computer,browser,login combination to sur the Internet ,schools, public kiosks, and so
on,

Regarding item three, while there is a seemingly cyclic pattern in the news media proclaiming
the dangers o \eb cookies and their ability to inade your priacy,` serices that measure
the acceptance or denial o cookies put the global acceptance rate at more than 99 percent.
Data suggests that problems resulting rom items our through six are less signiicant but are
always a possibility. 1he author reers the reader to the section on cookies and their use in
\eb analytics` later in this chapter or additional inormation regarding cookies.
Adding Page Tags to Every Page
lundamental to page tagging is the need to add a JaaScript tag to eery \eb page you want
to track. 1his is not necessarily a huge disadantage depending on the endor you choose
and the type o \eb enironment you hae but is potentially a time-consuming problem.
1he best case is where you hae already built a header,ooter` structure or your entire site,
one containing a common header and ooter ile into which the page tag can simply be
inserted. 1he worst case is where you hae an entirely static \eb site with no opportunity
or serer-side includes, necessitating the manual inclusion o the page tag on erer, page you
want to track.
Complexity of Tag Implementation
Depending on the endor you choose you may be aced with a more complex
implementation than you`d hoped or. Some endors proide ery simple tags that are the
same regardless o the type o page you want to track. Other endors proide ery complex
tags that need to be modiied heaily and are dierent depending on the type o page you
want to track. It is not uncommon or page-tag implementation projects to take a matter o
hours, but also not unheard o or said projects to take weeks or months to successully
complete.
Performance Issues
No matter how small the JaaScript code is that needs to go on each tracked page it is neer
zero weight.` It is a simple act that anything you add to your \eb pages will increase the
total download time to your isitors. Regardless o how emphatically a endor claims that
our tags result in no additional delay in loading and rendering your \eb pages` they are not
being completely honest. \hile the execution time or a page tag may be ery short,
typically measured in milliseconds, the additional time to download the page oer a modem
connection may be noticeable. It should, howeer, be noted that in most instances this
additional wait to your isitors is oset by the aluable inormation gained by including the
page tag in the irst place, especially i you hae no \eb analytics program already in place
TIe DIIIerenL Web TruIIIc DuLu Sources ,

Measurement of Robots and Spiders
\hile we`e listed exclusion o non-human user agents as an adantage o client-side data
collection, there are o course exceptions to this rule. 1he most oten obsered exceptions
are perormance monitoring user agents, such as those rom Keynote Systems, Gomez and
Mercury Interactie. Because these types o user agents are speciically designed to measure
the download and render time o \eb pages they do execute scripts on pages including the
JaaScript that controls a page tag. Additionally, because these serices are designed to
repeatedly test response times or key pages on a \eb site, i their requests are not excluded
these requests can artiicially inlate traic and page actiity metrics. I you plan on using
serices like Keynote alongside a client-side \eb analytics solution, make sure that the
solution you choose allows you to exclude the traic coming rom this type o user agent.
Length of Time Data Can Be Stored
Unlike \eb serer log iles, which you are ree to keep and store or as long as you like,
some page-tag endors do not keep your data oreer. Because there is a cost associated
with disk space, and an een higher cost associated with database storage, some endors
limit the length o time that data is made aailable ater its initial collection. Note that not all
endors do this, but this is something you want to check into ery closely when choosing a
endor i you think that haing access to historical data is important.
Types of Data that Can Be Collected
Unlike \eb serer log iles, page tags are limited in their ability to report on non-l1ML
iews, such as those to downloadable ile types, error pages and redirects. Some client-side
collection solutions are able to track iews o dierent ile types ,such as PDl, zipped iles,
executables, or multimedia iles, but do so by tracking clicks as opposed to successul
downloads. Many client-side solutions are able to record inormation about error pages as
long as these pages allow the necessary JaaScript to be included. Some client-side solutions
can also be tricked to count certain types o redirects. Needless to say, none o these data
types are easily or automatically collected with any signiicant accuracy and this should be
considered when choosing a solution.
Inflexibility in Reporting Formats
One disadantage o outsourced \eb analytics solutions that the sotware endors oten cite
is the act that, or the most part, what you see is what you get` or analysis and reporting
rom outsourced solutions. Since ASPs run the same application or most, i not all o their
customers, the reporting interace is typically standardized or all customers. 1his can be
seen as a disadantage i you really know what you are doing and you only want a particular
subset o reports, want reports in a dierent ormat ,Lxcel, Adobe PDl, and so on, or
would like dierent language used in your reports. Since the sotware endors hae been
saying that this is a signiicant limitation or the last ew years, outsourced endors hae
more recently begun to incorporate a ariety o solutions to this problem in their
applications. Some examples include customized language in reports, alternate themes or
iews in reporting and completely customized reports deliered either on an ad hoc or
S CIupLer

scheduled basis, typically in Lxcel. It should be noted that any endor proiding a tags into
logs` solution will likely proide the same lexibility in reporting aorded to sotware-based
solutions while still making use o data collected directly rom the browser client.
The Sticky Question of Who Owns the Data
One o the biggest disadantages associated with outsourced solutions o any type is the
question o who owns the data i we decide to cancel the contract` \hile endors will
assure you that you own the data, most will charge you some additional ee to allow you to
export the historical data so that you can create a data backup just in case` you eer leae
the particular endor. According to Guy Creese o the Aberdeen Group, there appears to be
a trend o companies trying two or three dierent endors beore inally settling in with the
one that is the best it or their needs. I this is true, there will always be a question o how
do we keep track o historical data when we moe rom endor to endor` \hile some
outsourced solutions can acilitate the import o \eb-serer log-ile data, this author is not
aware o any serice that allows the import o data rom another outsourced solution.
Unortunately, most companies accommodate this problem by manually building Lxcel
spreadsheets o basic historical data just in case`-a tedious and time-consuming task.
Privacy Issues
Internet priacy is an extremely popular subject, especially as Internet iruses and identity
thet cases continue to be reported with alarming requency. Issues surrounding what types
o data are collected, how they are passed around and where they are stored are common.
Because o this nearly eery \eb analytics endor proiding an outsourced solution oers
some kind o priacy policy, recommend that their customers adopt some part o their
priacy policy, and een, in some cases, go so ar as employing a Chie Priacy Oicer
,CPO,. 1he core argument regarding priacy and \eb analytics is that you are obligated to
tell your isitors that you are using some kind o client-side data collection solution, describe
what is and what is not collected and proide your isitors the ability to opt-out` o this
data collection i they so choose. But as long as the online business works with their
outsourced solution to implement an appropriate priacy policy coering the client-side data
collection there is little to worry about, at least at the time that this book was being written.

Again, while it seems like the list o disadantages o outsourced \eb analytics solutions are
many and may outweigh the relatie adantages, one must keep in mind that none o the
aorementioned disadantages are insurmountable. Careul selection o an outsourced
endor and a thoughtul implementation that is relectie o your speciic business needs is
usually all it takes to negate most, i not all, o the issues listed aboe. It is recommended
that any company careully consider the ollowing best practices, especially those
surrounding the selection o an outsourced endor, beore committing any money to a \eb
analytics project i it has been determined that a client-side data collection is the
methodology o choice.
TIe DIIIerenL Web TruIIIc DuLu Sources q

Best Practices When Implementing a Page Tagging
Data Collection Solution
\hile the author beliees that page-tagging data collection is a ery useul, powerul and
accurate source o data or any \eb analytics program, there are a handul o best
practices` that should be ollowed when setting up this type o solution. Most o these
suggestions deal with implementation and endor selection.
Provide Human Readable Names
As you can imagine, and in the author`s experience, the more easy-to-read the names o
pages and content groups, the more likely people are to make use o the aailable data. 1he
dierence between .

pl_gp_12311_groc`


. and .

Store Category View: Product List 12311

. should be obious in terms o the ability to quickly understand what inormation was
coneyed to the isitor. Using the irst example ,pl_gp_ .`, there are without a doubt a
group o users in your organization that will understand exactly what page or content group
is being described. More oten than not this is a small group o I1 or technical people who
built the pages or the system. In the second example ,Store Category: .`, you can
reasonably assume that av, member o your organization who is amiliar with your \eb site
and business model should be able to quickly determine which page is being described.
light the temptation to take the easy way out and use simple to generate` names, spending
the extra time at implementation to proide simple to understand` names instead will pay
o in the long run.
Establish a Quality Assurance Program
Making sure that you hae a well-deined process or testing the deployment o page tag
scripts on new pages or sections o your \eb site is critical. lundamental to the way that
most page tagging technologies work is the rule i you don`t tell us about it we hae no way
to know.` Put another way, i you mess up the implementation and ail to collect data, likely
that data is oreer lost. 1o this end the author recommends working ery closely with your
application proider or other knowledgeable resource to implement and test any new code
being deployed prior to pushing the code to your lie serers.
Challenge Your Solution Provider
Many times in this author`s experience customers hae simply said just send us the code
and the manual and we`ll get it all up and running` only to come back at later and say we
neer elt like we got the ull alue o the application because we don`t think we
qo CIupLer

implemented it correctly.` Any endor you choose, no matter how small, should be more
than happy to help you understand how to best implement their application so that your
business makes use o all their bells and whistles. I the endor you choose is not able to
help you, ind another endor. Also, don`t assume that you will be able to igure it out
yourseles just to sae a little bit o time or money. I you are reading this book it is ery
likely that \eb analytics application integration is not your core competency. 1ake
adantage o outside resources who implement that particular technology or a liing and
you will signiicantly increase your chances o getting more out o the solution in the long
run.
Conduct Performance Testing
In all cases, using a page tagging solution will increase the physical time it takes to download
your pages somewhat, but ollowing this best practice` will let you quantiy this increase.
1hings to watch are total increase in kilobytes ,most endors will round down when
describing the total weight o their scripts, and total increase in response time, especially or
the image or object that is being requested rom the endor`s data collection center. An
important caeat to this is that download time should not be the absolute indicator o the
eect on the isitor experience. Seeral endors proide combinations o scripts, some that
load eery time to beat the cache and others that are highly cacheable to improe oerall
perormance. \hile you should monitor the eect o download time it is also important to
double-check this against some type o usability testing, asking the subjects i they perceie a
dierence between pages containing the script and those without.
Exclude Internal Traffic from the Analysis
By deault most analytics applications ao vot exclude internally generated traic to the \eb
serers, that traic generated by your deelopers and other employees making use o the
business \eb site or whateer reason. 1his traic should be excluded or segregated out o
the inal analysis because it is not necessarily representatie o your isitor`s interests.
Depending on your particular business model, internal traic can be a signiicant component
o the traic to a gien page or content area.

Most \eb analytics applications will allow you to exclude traic rom reporting and analysis
based on at least the IP address o the isitor. \hile IP address is a poor indicator o
uniqueness, it is oten useul or excluding groups o isitors as most corporate irewalls will
identiy internal users as haing coming through one or ew unique addresses.
Do Your Homework Regarding Vendor Selection
Because most page tagging solutions are oered as an outsourced serice, the single most
important aspect o your success with an application will be the company that supports it. It
is critical to ensure that you trust the endor team that will be responsible or your success.
1he author highly recommends that at minimum you do the ollowing:

1. \hile still in the sales cycle, ask to meet or talk to the team that will support you
ater you sign the contract, as this is almost always a dierent group than you talk
to in the sales process.
TIe DIIIerenL Web TruIIIc DuLu Sources q1


2. Ask or customer reerences and actually call those reerences. Ask these people
things like, low long hae you been with this endor` \hat was the
implementation like` low do you eel about the support they proide`
Describe a situation where you didn`t eel like the endor responded as quickly as
you would hae liked.` \ho is your primary point o contact at the company` I
the reerence really likes their primary point o contact, ask the endor i that
person can work with you ask well. And keep in mind that tbe revaor bava.etect. tbe.e
referevce. to .a, tbe be.t tbivg. o..ibte abovt tbev.

3. lunt down other customers that the endor does not proide as reerences and
call these people. Considering that the endor will only proide reerences who
gie glowing recommendations and that you can pretty easily igure out whom else
is using their technology ,consult the Our Customers` page on the endor`s \eb
site, it will beneit you to spend a little time trying to run down a non-reerence
customer. I you are able to ind a reerence that is in your line o business you will
likely get more releant and honest answers to your questions. \hen you identiy
the company, simply write to webmaster` the company and clearly explain what
you are trying to accomplish and could they please pass your contact inormation
along to the appropriate resource. Include all o your contact inormation so the
recipient knows you are serious and not just spamming them.

4. Ask the endor what kinds o support and training they proide, how oten it is
aailable and how much it will cost. Any good endor will be happy to sell you
training and support ater you hae already signed a contract. Strong endors will
also proide ree training and support, at least on a limited basis, to improe
customer retention and satisaction. I training and support is only aailable at an
additional charge, ask what they recommend or proide to most other customers
and then make them include this at no additional charge in the contract as part o
the negotiation.

As you can see, because page tags were designed to collect true \eb analytics data there are
ewer recommendations here about how to ensure the data is good and more suggestions to
help you leerage the solution you choose to your ongoing success.
Examples of When to Use a Page-Tagging Data
Collection Solution
1he ollowing are examples o when you`d likely choose an outsourced \eb analytics
solution:
When Information Technology Staff Are Too Busy
1he most common reason that companies switch rom an in-house to an outsourced
solution is a lack o resources to manage, maintain and update an internal solution. \hether
the necessary sta is oerworked, underpaid or simply disinterested in helping, attempting to
make the most o an in-house solution without internal support is nearly always an exercise
q CIupLer

in utility. In contrast, outsourced solutions typically require ery little internal support on
an ongoing basis, usually only at the onset o a project where pages are being tagged or when
data collection is being modiied to suit speciic measurement goals. Nearly all outsourced
solutions eliminate the need or I1 to be inoled in the reporting process, thus speeding the
inormation deliery to the end-user`s hands.
When Traffic Volume is Very High
In high-traic olume situations there is a tradeo between haing to pay more or an
outsourced solution that can aggregate data rom multiple machines and report data back
quickly, or paying less but waiting long periods o time to get reports. Depending on the
size and number o log iles being generated by high olume sites, \eb-serer log-ile
analysis solutions can take hours, days or weeks to report back on inormation that can be
generated in real time` by outsourced solutions ,but again, see the section on real-time`
reporting, below,.
When Needs are Diverse
Because page-tagging solutions are more lexible in terms o the inormation they can
collect, client-side data collection is typically preerred in situations where companies would
like to tie traic data to online purchasing data, perorm complex isitor segmentation, or
rapidly change naming conentions to accommodate dierent measurement needs. 1he
ability to pass in customer-deined ariables typically imparts additional measurement and
reporting power that can be leeraged to answer more interesting questions than, simply,
low much traic did the \eb site get last week`
When Accuracy is Very Important
Page-tagging applications are ery accurate by deault in terms o measuring unique
indiiduals, busting proxy caches and excluding non-human traic rom measurement and
reporting. All o these things can be accomplished using a \eb serer log based solution i
great care is exercised when implementing and the solution is careully maintained. Lacking
this attention to detail, client-side data collection is the preerred method or accurate data
collection.
Tags into Logs
A more recent entry into the battle between \eb serer log iles and client-side page tags is
the tags into logs` solution. 1he rationale behind tags into logs` is that while page tagging
is recognized as a powerul and accurate method or collecting data, some businesses preer
the reporting lexibility o an in-house, sotware solution. 1ags into logs` gies the
business both-the \eb site is tagged as i the reporting were entirely outsourced except the
data collection is collected internally through a mechanism that simply writes the request to a
ile. 1his ile can then be parsed and analyzed using the same type o sotware solution one
would use to analyze a traditional \eb serer log ile.

TIe DIIIerenL Web TruIIIc DuLu Sources q

1he clear adantages o tags into logs` are accuracy, ownership o data and lexibility o
reporting. 1his build a better log ile` concept eectiely neutralizes many o the
disadantages o a \eb-serer log-based solution. No more concerns about determining the
uniqueness o a isitor, proxy or browser caches, no more concern oer robots and spiders.
Also, this type o solution consolidates actiity across multiple \eb serers, eliminating the
need to stitch logs rom multiple serers together prior to analysis and reporting.

1he disadantages o this type o solution are similar to those o any in-house solution.
Unless you outsource the analysis application you will still need to hae your I1 sta install,
monitor and maintain a sotware application, you still need to store and rotate log iles, and
you will still usually need to wait hours, days or weeks or reports to be generated. Also,
unless the particular solution you choose stores the inormation collected in a standard log
ile ormat ,such as NCSA combined, you are still stuck with the problem o who owns the
data`-een i you own and store the data on your serers, i it is in an incompatible ormat
it is unctionally lost i you change solutions ,consider Beta ideotapes and 8-track
audiotapes,.
About Real-Time Data Collection and Reporting
A marketing point that some outsourced solution proiders go to great length to emphasize
is the alue o real-time` data reporting-the ability to see isiting traic almost
immediately ater it occurs. \hile this sounds great, there are only a ew instances where an
online business would actually need true real-time` reporting.
When the Online Business is a Content Provider
I the online business is publishing content to create adertising iews, engender loyalty and
create site stickiness, there is something to be said or knowing the eects o content
immediately ater this content has been posted. 1he best example o this is the news portal
content editor who wants to maximize the eect o a new news story. le or she would post
a new story, watch the traic lowing to that story, and continue to eature that story
prominently until the percentage o traic iewing that story slips below a pre-deined
threshold based on the ratio o traic to the story compared to oerall traic olumes.
Where the Business is an Online Retailer
Being able to see the real-time or near real-time response to product placement on key pages
throughout a \eb site, including product conersion rates and reenue generated, can help
smart merchandisers maximize the alue o product placement. An example o this would
be the online lorist testing three dierent lower arrangements just beore Mother`s day,
closely watching the conersion rates or each in real time, rotating out the arrangement with
the lowest conersion rate eery thirty minutes until the most popular product spread is
achieed, based on conersion rates that start high and stay high.
qq CIupLer

When You are Serious About Improvement
Because there are no minimums or maximums to the amount o time changes should be
measured when using a strong \eb analytics methodology, smart business managers can use
real-time reporting to test the eects o changes to multi-step processes rapidly. Rather than
run the risk o testing changes made to a process or a ew days only to discoer that the
result is higher abandonment or lower conersion, real-time reporting can help identiy this
trend quickly so that these changes can be backed out, thus minimizing any detrimental
eect.

Other than these three examples, real-time reporting is usually a nice eature that ew people
actually take adantage o, much like a sotware manual or online help system. Again, beore
you choose a endor simply because they proide real-time` reporting make sure that this is
something you actually need, something you can take adantage o.
Cookies and Their Use in Web Analytics Applications
At this point in !eb .vat,tic. Dev,.tifiea you hae read about cookies and their use in
analytics applications many times. Cookies hae eoled into an important and controersial
technology on the Internet-some demonize their use as an inasion o priacy, others
praise their ability to simpliy and personalize the browsing experience. \e`e already
discussed some o the assumptions and limitations inherent to cookies as a tracking
mechanism. Regardless o their limitations and how people eel about their use, cookies are
critical to \eb analytics applications and their ability to accurately track the uniqueness o a
isitor.

Because isitors in most cases are not willing to log in to a \eb site simply to allow the site
to track them accurately, and since it has been shown repeatedly that IP addresses are
inadequate to determine the uniqueness o a isitor, cookies are here to stay. \hile isitors
are ree to set, change and delete their cookies, especially using new priacy tools aailable in
\eb browsers such as Internet Lxplorer 6.0 and the growth o P3P priacy standards, data
suggests that ew people actually take adantage o this. I this is true, cookies will continue
to be an eectie tool to track isitor actiity on a \eb site.

One o the major concerns or priacy experts is the dierence between irst- and third-
party cookies. lirst-party cookies are those cookies sered directly rom one`s \eb site,
sered directly rom the \eb site`s own domain. 1hird-party cookies are those cookies
sered rom another domain ,such as DoubleClick or \ebSideStory,. Lmerging priacy
standards are dierent or each type o cookie and are generally stricter or third-party
cookies. 1he issue that most priacy adocates cite is the third-party cookie`s ability to
obsere isitor behaior across multiple domains-i I am tracked with the DoubleClick
cookie I can be seen` on any site tracked using DoubleClick`s tracking technology. \hile
this may be technically true, in the author`s experience endors proiding outsourced
solutions that depend on third-party cookies are etrevet, conscientious about not creating
isibility across sites. Some go so ar as to identiy their cookies as being part o a combined
domain using the sub-domain attribute-they set a cookie to ehg-bcstore.hitbox.com`
instead o just www.hitbox.com.` \hile there is really no dierence technically speaking,
TIe DIIIerenL Web TruIIIc DuLu Sources qg

and both URLs are rom a third-party domain ,hitbox.com`,, this combined assignment
would, in theory, make it more diicult or someone other than litBox to access the
inormation contained in these cookies.

lirst-party cookies skirt this issue by being written directly rom the business` own domain.
Only applications and script running within that domain hae access to the cookie, a priacy
standard built into all \eb browsers. \hile the use o irst-party cookies protects the isitor
slightly more, some perceie the act that their use necessitates running sotware in-house as
a disadantage. Clearly there would be an adantage gien to an outsourced analytics
proider who was able to write irst-party cookies, thus increasing priacy ,or at least the
perception o priacy to the isitor or customer,.

lundamentally, as long as the cookies you use to track isitors to your \eb site do not
contain any personally identiiable inormation and your endor is not using third-party
cookies to track isitors across multiple customer \eb sites, cookies should be considered a
sae and eectie tool to enable your analytics program. Usually the kinds o Internet users
that complain about the use o cookies by a \eb site are either unsophisticated or oerly
sophisticated. 1he unsophisticated isitor has heard that cookies are bad` and that they are
an inasion o priacy and has taken this at ace alue. \hen their personal irewall pops up
a warning about your cookies they become alarmed and more oten than not ire o an
email telling you to keep your hands o my computer.` 1he oerly-sophisticated isitor is
usually an I1 proessional or security expert, someone who should know better but has little
else to do than stir up trouble or legitimate businesses. 1his type is more likely to actually
create problems or you in that they may post the act that you are using cookies on
newsgroups or een report your use to the media in an eort to gain personal attention.

I you are using cookies, and it is strongly recommended that you do so, you do leae your
business exposed to the aorementioned users. 1o protect against this exposure it is also
strongly recommended that you work directly with your analytics proider to crat a priacy
policy that describes eactt, how you use cookies and what you use them or. \hile nobody
actually reads priacy policies, it is important to hae a document that you can point to i
someone gies you trouble and say, \ell, we`re sorry you eel that way but our priacy
policy clearly states how and why we use cookies.` In the author`s experience, proiding the
policy is correct and up-to-date, this generally helps to minimize the allout and encourages
the problem user to moe onto more important things like how to get a date to the prom,
how to clean the lint rom their belly-buttons, and so on

Consult your analytics endor or more inormation about how cookies are used within their
application and their suggestions or modiications to your priacy policy in this regard.

q6 CIupLer q

CHAPTIR q
WIB ANALYTICS
TIRMINOLOGY


We measure this information based on hits in our traffic analysis program which is
different from the visits that we get from our advertising engine which cannot be
compared to the visitors described by our panel measurement service and are also
not quite the same as the page views that are cited in the recent articles about the
state of our particular market niche. Anyway, here are the numbers.
Anonymous Marketing Person


A good subtitle or this chapter would be, \hy the same words mean dierent things to
dierent people and what you can reasonably do about it!` Because \eb analytics is still
ery much an emerging discipline, dierent endors hae been ree to deine words as they
saw it with little or no consideration or common usage or standardization. 1his lack o
attention to detail has resulted in common questions like:

\hat the heck is a hit` and what do hits` tell me about traic to my site
\hat is the relationship between a isitor, a isit and a iew \hat is the dierent
between a page iew,` a total iew` and a iew`
\hat is a unique` and why wouldn`t eery isitor be a unique isitor`
\hy do some endors measure some statistics by iews` and others by isits`
and others by isitors` \hich endor is right!

1his type o rustration goes on and on and on, especially when the group charged with
deeloping \eb analytics programs within an organization is then tasked with presenting
their indings to non-\eb analytics people. 1his would not be so bad i it were as simple as
educating the outsiders` about a new language, one that was well understood, such as
mathematics or Spanish, you simply translate the term once or twice and they`re likely to
igure it out. 1he problem with \eb analytics is that you end up giing people the
\eb1rends,NetIQ deinition o page iew,` which is dierent than the litBox deinition
o page iew,` neither o which are talking about a hit` but hit` is what we call a page iew`
here internally.` \ou may hae already experienced this type o rustration.

1his chapter will proide standard deinitions or the most commonly used \eb analytics
terms, outlining what the words .bovta mean, regardless o how endors hae used them.
Web unuIyLIcs TermInoIogy q,

A Small Group of Mostly Useless Terms
1he irst group includes mostly useless terms that hae managed to make it into the
common parlance o \eb analytics. 1erms like hit` and click-through` only sere to
conuse people, making them think they`re measuring something that they are not.
Hit
lit` is a particularly bad word, people talk at length about the number o hits` that their
\eb site receied in such-and-such a timerame. 1he common usage o hit` appears to
approximate page iew`-the number o times a \eb page is actually seen by a isitor.
Compare this usage with the deinition proided by \eb1rends,NetIQ on their \eb site:

(A hit is) an action on a Web site such as when a user views a
page or downloads a file.

At irst glance the reader is tempted to agree that a hit` and a page iew` are analogous.
1he deinition states that a hit is an action, such as a user iew,ing, a page,` does it not
1his is not the problem, the problem is the second part o the deinition, or downloads a
ile.`

Consider the makeup o a \eb page, \eb pages are simply an l1ML description o the
relationship between text and objects. 1he result o the assembly o objects around text is a
rendered page.` Lach o the objects` that make up this page` need to be downloaded
along with the l1ML description o the page and herein lies the problem. Most o the hit`
numbers that are reported by \eb serer log ile analyzers include l1ML pages ava
downloadable ile types such as .`GIl,` .`JPG` and .`CSS.` 1his means that or any gien
page` that is downloaded there can be dozens or hundreds o hits` required to completely
render the page. 1alk about artiicially inlating the numbers!

Another way to think about hits is using the pyramid model o \eb analytics data presented
later in this chapter. 1he pyramid describes the utility o data types in inorming \eb
analytics projects, rom least to greatest alue in terms o unique isitors to a \eb site.
lits,` i this data is een aailable, are the bottom o the pyramid.
Click-Through
Click-through,` or click-through rate,` is used to describe the number o people who
click on a particular thing -a banner ad, a link, a keyword. Again, \eb1rends,NetIQ
deines click-through rate` as:

(The) percentage of users who click on a viewed advertisement. A
good indication of an ads effectiveness.

\hile it is good to measure rates and percentages, as opposed to absolute numbers, using
the percentage o clicks on a particular link as a measure o eectieness can be misleading.
Consider the number o banner ads that are now cleerly disguised as content on \eb sites.
qS CIupLer q

Consider the number o times you hae clicked on a link that looked useul, only to discoer
that link was taking you to a site designed to sell you something you don`t need. Consider
the number o times you hae clicked on a search result, only to immediately back up when
the response was slow.

A common experience on the Internet is the click, oops, back up` method o browsing.
1he user clicks, accidentally or not, on a link but immediately thinks, I didn`t mean to do
that` and uses the browser`s back button. In most instances the click` part o click, oops,
back up` is counted as a part o the click-through rate.` I people change their minds 5
percent o the time, your click-through rate would be inlated by 5 percent, i people change
their mind 20 percent o the time, common when site response is slow, your click-through
rate would be inlated by 20 percent.

A better measure than click-through` is response rate, the measure o the number o
isitors that actually arried at the site ater clicking on the link. 1he response rates or a
particular ad or link can help you better understand one o the most useul measurements in
\eb analytics, the actiity conersion rate.

In short, don`t use hits` or click-through` when talking about \eb analytics i you can
help it. I someone else says, \e had 200,000 hits to our \eb Site last month,` ask them i
they really mean hits or perhaps they`re talking about page iews, or isits or unique isitors.
I they tell you, Our click-through-rate rom Oerture is 95 percent,` ask them what
percentage o those clicks actually gets to their \eb site and how many o those respondents
actually do something important. I they aren`t sure, please gie them a copy o this book.
A Slightly Larger Group of Useful But Confusing
Terms
1he next group includes useul and necessary terms to \eb analytics that are oten
conusing because dierent endors deine them in dierent ways. 1his group includes core
\eb analytics terms like page iew,` isit,` isitor,` unique` and reerrer.`
Page View
Page iews ,or iews`, are the next highest leel o inormation in the pyramid model o
\eb analytics data. Page iews are actual isitor iews or requests or inormation,
regardless o the number o images or objects necessary to construct said page. 1he
ollowing issues contribute to conusion regarding the deinition o a page iew:

1he question of what is a page. Most systems are able to measure l1ML
pages with ease, either rom a request logged in a \eb serer log ile or ia a page
tag. Problems may arise when considering dynamically generated pages. It is ery
common or business \eb sites to make use o a dynamic system to generate a
large number o pages` rom a small number o templates and a database. In this
instance, the deinition o page can become blurred. I a single Actie Serer Pages
template can deine 100 dierent product iews, do we measure 100 iews o the
Web unuIyLIcs TermInoIogy qq

template as a page iewed 100 times, or each dierent product iew as a page
iewed one time Both strategies or measurement would contribute the correct
number to a sum total o all page iews or the site, but only the latter would
proide the leel o detail that most online business managers need.
1ypes of documents included in the definition of a page. Again, in general
l1ML pages are measured without problem. Problems can arise when dierent
systems are used or content presentation, such as Cold lusion, Actie Serer
Pages or PLRL,CGI. \hat i some o your ASP pages are designed to present
content but others are or data processing Do you count all ASP pages as page
iews` \hat i your catalog is presented using PLRL,CGI but your \eb
analytics application does not recognize .cgi` as a alid page name extension In
this regard, clear deinition o which document types should be considered page
iews` is necessary.
Repeat views (resets or reloads). It is unortunate but some \eb analytics
application proiders choose to deal with reloaded page iews as dierent rom the
original page iew. One endor counts any page that is reloaded or rereshed
within iteen seconds o the original page loading as a reload` and segregates the
counting o the reload rom the page iew.` \hile it is perhaps interesting to
know this, in most cases this inormation proides little additional alue and only
seres to obuscate the deinition o a page iew.
Irames. Lasily one o the most conounding issues in \eb analytics and the
deinition o a page` is the ability or \eb designers to use rames and ramesets.
Briely, a rameset is a description o how other pages should be laid out in a
browser to represent a page` to the end user. \hile rames proide some ery
minor adantages to \eb deelopers ,static naigation elements, the ability to
proide a common header and ooter that aren`t constantly reloaded,, end users are
oten conused by ramed pages.

1he problem with rames in the context o \eb analytics is the question o what
constitutes a page` Is it just the content rame, or is it all rames considered
together I you are using a \eb serer log analyzer, rames can present serious
problems when counting page iews -should you count the rameset description
and the non-content rames I not, what is your strategy or excluding ramesets
and non-content rames rom the analysis Client-side data collection serices hae
an easier time with rames since it is generally easier to deine what is counted and
what is excluded, but een then you cannot be 100 percent certain what the user
was truly seeing when conducting an analysis.

\ith all o this in mind, the deinition o a page iew should be as ollows:

. age rier i. covvtea ritb tbe .vcce..fvt toaaivg of av, document containing content tbat ra.
reqve.tea b, a !eb .ite ri.itor, regarate.. of tbe vecbavi.v of aetirer, or vvvber ava freqvevc, ritb rbicb
.aia covtevt i. reqve.tea.

Note that this deinition is ery inclusie -content does not necessarily only denote text or
l1ML. I you hae a \eb page that loads a llash ile but you would like an accurate count
o how many times that llash ile was loaded you can use page iews ,llash can create
dierent problems, howeer, dealt with below under \hat about llash`,.
go CIupLer q


Using this deinition it should be clear that i your \eb site is built using rames you should
only be counting the actual content the isitor is requesting, not the header, ooter or
naigation rame content. It should also be clear that no deinition o page iew should
depend on whether the page was loaded once or many times in rapid succession.
Visit
Visits are the next highest leel o data alue in the pyramid model o \eb analytics data. A
isit, sometimes reerred to as a session` or user session,` is what a isitor` is engaged in
while generating page iews` on one`s \eb site. 1he most conusing aspects o isit are
synonymy and duration.

Synonyms or isit are session` and user session`-all three terms mean the same thing.
1he authors preer the term isit simply because it is conenient when describing the
pyramid model, that is, vviqve ri.itor. ri.it \eb sites generating age rier..

1he duration o a isit is deined as any period o actiity, as measured by page iews or
requests or pages, separated by 30 minutes o inactiity. 1his is why some endors reer to
isits as haing a 30-minute timeout.` 1hirty minutes o inactiity was originally an
arbitrary number but has become the ae facto standard. 1he ollowing examples attempt to
illustrate how the 30-minute timeout should be applied when calculating isits:

A isitor clicks to a \eb site at 1:00 PM and surs at random or 20 consecutie
minutes beore closing their browser and going oline. One visit.

A isitor clicks to a \eb site at 1:00 PM and starts putting items in a shopping cart.
At 1:15 PM the telephone rings and or 31 minutes the isitor does not click any
links on the \eb site. At 1:46 PM the isitor ends their phone call and completes
their purchase transaction by 1:5 PM. 1wo visits-the irst rom 1:00 PM to 1:15
PM, the second rom 1:46 PM until the transaction is completed at 1:5 PM.

A isitor clicks to a \eb site at 1:00 PM and randomly surs or 45 minutes. At
1:45 PM the isitor notices a link to another site which, when they click, opens a
new \eb browser ,leaing the existing site open,. 1he isitor surs the new site or
31 minutes, at which time they close the newly opened browser and continue to
sur the original site. 1wo visits-een though the browser containing the original
site was let open, i there is no actiity on that site the isit will hae timed out at
2:16 PM ,1:45 PM plus 31 minutes on the new site,.

A isitor clicks to a \eb site at 1:00 PM and randomly surs or three hours
without any break in actiity. At 3:00 PM the isitor makes a sandwich but is back
suring at 3:29 PM, at which time they sur or 30 more minutes beore closing the
browser and logging o. One visit-the longest period o inactiity was 29
minutes, one minute shy o the 30-minute timeout.

Web unuIyLIcs TermInoIogy g1

1he deinition o a isit should be as ollows:

. ri.it i. covvtea rbev a vviqve ri.itor create. actirit, ov !eb age-rbicb are beivg vea.vrea ria a age
rier .tati.tic-regarate.. of tbe avratiov of tbi. actirit, a. tovg a. tbe erioa of ivactirit, betreev age rier.
aoe. vot eteva be,ova tbirt, vivvte..

Note that the key to the deinition o a isit is that the pages the isitor is browsing are those
being measured as part o the page iew statistic. 1his is coered by the third example
aboe, where the isitor was browsing page on a dierent \eb site, but this could easily
extend to pages on your own site not being tracked by your \eb analytics package.
Visitor or Unique Visitor
Visitors, commonly called unique isitors,` are the next highest leel o data alue in the
pyramid model o \eb analytics data. A unique isitor is the person isiting the \eb site,
generating page iews. luman beings are isitors, search engine indexing agents and
perormance monitoring applications are not.

1he undamental problem with unique isitors is the method used to determine uniqueness.
As preiously discussed, strategies or determining uniqueness o a isitor include IP
addresses and cookies. 1he section in Chapter 3 on \eb-serer log iles discusses the
disadantages o using IP addresses to determine uniqueness at length. 1he section in
Chapter 3 on client-side data collection discusses the disadantages o using cookies to
determine uniqueness. \hile cookies are not perect or determining the uniqueness o a
isitor, they are the preerred method or identiying anonymous isitors.

1he deinition o a unique isitor should be as ollows:

. vviqve ri.itor i. covvtea rbev a bvvav beivg v.e. a !eb bror.er to ri.it a !eb .ite, regarate.. of tbe
vvvber of age. ri.itea or tbe avratiov of tbe ri.it. . ri.itor cav be vviqve for aifferevt erioa. of tive, ava
tbe ivairiavatit, of a ri.itor i. aefivea, referabt,, b, tbe re.evce of a coo/ie .torea iv tbe ri.itor`. !eb
bror.er.

Note the dierentiation o uniqueness or dierent periods o time. \eb analytics
applications oten proide distinct measurement or daily, weekly and monthly unique
isitors. 1his concept necessitates an addendum to the deinition o a unique isitor.

. aait, vviqve ri.itor .bovta be covvtea tbe fir.t, ava ovt, tbe fir.t, tive a vviqve ri.itor ri.it. a !eb .ite
ov av, girev aa,. . ree/t, vviqve ri.itor .bovta be covvtea tbe fir.t, ava ovt, tbe fir.t, tive a vviqve
ri.itor ri.it. a !eb .ite iv a girev ree/. . vovtbt, vviqve ri.itor .bovta be covvtea tbe fir.t, ava ovt, tbe
fir.t, tive a vviqve ri.itor ri.it. a !eb .ite iv a girev vovtb.

1his deinition can be simpliied somewhat when considering any arbitrary timerame:

. vviqve ri.itor for av, arbitrar, tivefrave .bovta be covvtea ove tive ava ove tive ovt, ov tbeir fir.t
ri.it betreev tbe .tart aate ava tbe eva aate.

g CIupLer q

1his way, i a person isits a \eb site or the irst time in a month, they should be counted
as a daily unique isitor on the day o the isit, a weekly unique isitor in the calendar week
o the isit, and a monthly unique isitor in that calendar month. I they return the next day,
the isitor would only be counted as a daily unique isitor or that day, unless the day in
question is in a dierent calendar week or month than the preious day.
Referrer
1he term reerrer` is used to describe the source o traic to a \eb site. Most oten
reerrers are reported as reerring URL` or reerring domain.` 1he ormer is the complete
URL describing the \eb page containing the link to one`s \eb site. 1he latter is only the
domain portion o this URL ,the amazon.com` portion o
http:,,www.amazon.combookID~12345,. 1he reerrer is inormation contained in the
l11P header when the isitor is passed to the \eb site being tracked, inormation that is
aailable to both \eb-serer log iles and client-side page tags.

Common problems with reerrers include what to report when there is no reerrer and how
reerrers are grouped. Oten, when there is no reerrer, \eb analytics applications will
report that the link came rom a browser aorite` or bookmark` or that the isitor typed
the URL or the \eb site directly into the browser. Occasionally, the absence o a reerring
URL means that the click was passed through some technology that stripped the reerring
URL, most non-l1ML based email applications and some banner-sering network sotware
are known to do this.

Any groupings o isits that do not contain reerring inormation make assumptions. One
assumption, i this traic is reerred to as Bookmarks or Directly Reerred URLs,` is that
people oten bookmark \eb pages on the site being tracked, or perhaps that marketing
actiity is eectie-people know the URL and type it in directly to the browser. 1he
problem with these assumptions is that they are assumptions-one has no way o accurately
knowing which answer is true. Because o this, it would be better to describe isits that do
not contain reerring source inormation as haing no or unknown reerrer.`



Iigure 2: Lxample referring URLs driving traffic to the author's Web site,
www.webanalyticsdemystified.com.
Web unuIyLIcs TermInoIogy g


In terms o grouping URLs, \eb analytics applications hae a tendency to consider traic
rom the top search engines as traic rom search engines` and group these results, at least
when reporting on reerring domains. 1he problem with this is that not all traic rom
these sites is simply search results. 1his traic could be rom banner ads, paid search
inclusion, partnerships, ailiate programs or any number o non-search actiities. 1he
lumping o search engines together results in a loss in granularity rom these reerring
domains and should be aoided i possible.

1he deinition o a reerrer should be as ollows:

. referrer to av, !eb .ite .bovta be av vvaifferevtiatea ava covtete vviforv re.ovrce tocator ;|R)
ae.cribivg tbe eact age ov tbe referrivg !eb .ite tbat covtaivea tbe tiv/ to tbe .ite iv qve.tiov.

\hen reporting on reerring URLs, all necessary components to reconstruct the reerring
page should be included ,that is, the URI and query string,. \hen reporting on reerring
domains, domains should be reported indiidually and without grouping or categorization.
Any requests that do not contain reerring source inormation should either be reported as
unknown reerrer` or not displayed in a reerrer report at all. Because reerring source
inormation is critical to measuring reach and acquisition ,Chapters 11 and 12, this
inormation should be reported as accurately and without bias as possible.
A Medium-Sized Group of Truly Useful Terms
1his inal group includes terms that are extremely important to \eb analytics measurement,
but are oten either oerlooked or are just now coming into common usage. 1his group
includes terms like conersion rate,` abandonment rate,` attrition,` loyalty,` recency`
and requency.`
Conversion Rate
Conersion rate is likely the most used and well understood o all these terms. \hereas in
the past most traic analysis ocused on hits or page iews, nearly all adocates o \eb
analytics encourage online businesses to measure and improe their conersion rate,s,.
Simply put:

. covrer.iov rate i. tbe vvvber of covteter.` airiaea b, tbe vvvber of .tarter.` for av, ovtive actirit, tbat
i. vore tbav ove togicat .te iv tevgtb.

As you can see, this deinition is deoid o units o measurement and any notes about where
an actiity begins and where it ends. 1he deinition is deoid o units because conersion
rate can be measured using iews, isits or isitors ,although you should not use iews unless
no other metric is aailable, and een then it is not a good idea,. 1he deinition does not
suggest that conersion rate needs to be measured rom the home page, or any page, as there
are likely vav, dierent conersion rates one can measure on any \eb site.

gq CIupLer q

Conersion rates will be discussed at length in Chapter 13, describing the need or tools to
measure multi-step processes online.
Abandonment Rate
Related to conersion rate, the abandonment rate is a measurement within any multi-step
process describing the number o units that don`t make it rom step n` to step n - 1.`
Simply put:

1be abavaovvevt rate for av, .te iv a vvtti.te roce.. i. ove vivv. tbe vvvber of vvit. tbat va/e it to
.te v - 1 airiaea b, tbo.e at .te v ;1 -;;v-1),v)).

Again, this number can be measured using whateer metrics are aailable to you at the time
,iews, isits, isitors,. \ou should see that or any multi-step process there are one ewer
abandonment rates than there are steps in the process, e.g., a ie step process will hae our
abandonment rates ,ligure 3,.

\ou may hae seen the term leakage` used in a similar context as abandonment. Leakage is
a term that other authors use to describe the abandonment or loss experienced prior to a
isitor actually engaging in a conersion actiity. \hile there is a case to be made or
dierentiating leakage and abandonment, this separation is artiicial and both terms are
describing the same thing measured at dierent points in a process.



Iigure 3: Lxample of how abandonment and conversion are often measured using
some type of conversion funnel or process measurement tool.
Attrition
Once you hae successully been able to measure how many isitors conert ater coming to
your \eb site, you will want to keep track o the number o these people that continue to
come back and conert again. I your particular business model depends on acquiring
isitors who conert and haing them return to conert repeatedly, attrition is the measure
o the number o repeaters that you lose. Put another way:

Web unuIyLIcs TermInoIogy gg

.ttritiov i. a vea.vrevevt of tbe vvvber of eote ,ov bare beev abte to .vcce..fvtt, covrert ovce bvt are
vvabte to retaiv to covrert agaiv.

lere you can see that we apply a unit, people, since regardless o how you measure
conersion and abandonment, it should be recognized that it is human beings that are doing
the conerting and abandoning.

1wo important considerations when measuring attrition is A, do you expect people to return
and conert a second time and B, how oten do you expect this I you are measuring the
conersion rate or people downloading a white paper or application, one would expect your
attrition rate to be nearly 100 percent since they ery likely do not need to download that
same item again. Conersely, i you are selling products oer the Internet one would hope
that any isitor you could successully conert into a customer will return to purchase again
and again. I the latter is true, being able to measure the aerage requency o purchase will
help you better measure your attrition rate, as you don`t want to measure attrition oer too
short or too long a period o time.
Loyalty, Frequency and Recency
I it is a goal to hae people isit your \eb site more than once, which is nearly always the
objectie, it is a good idea to measure loyalty, requency and recency. Lach o these terms
help the \eb site owner quantiy how well they are doing at retaining traic.

Loyalty is the measure o the number o times a isitor can be expected to return. Simply
put:

o,att, i. a vea.vre of tbe vvvber of ri.it. av, ri.itor i. ti/et, to va/e orer tbeir tifetive a. a ri.itor. t
.bovta be vea.vrea a. tbe rar vvvber of ri.it. att ri.itor. bare vaae .ivce vea.vrevevt ra. ivitiatea, ava
tbe vvvber of ri.it. .bovta be aeavticatea.

One should be able to quickly see that loyalty is a moing target, a measurement whose
characteristics will change oer time. I you measure loyalty the irst week you turn your
\eb site on you are likely to be sorely disappointed, as most isitors will hae only isited
once. But i you measure loyalty oer time you will be able to use this inormation to build a
gross proile o your isitors.

Loyalty should be reported as the number o raw isits per isitor, or example, 100 isitors
made 3 isits, 8 isitors made 4 isits,` and so on. 1hese numbers should be de-
duplicated-i a person were to isit or the third time she should be counted in the 3
isits` column until she isits or her ourth time, at which time the 3 isits` column should
be decremented and the 4 isits` column increased. lailure to do this will artiicially
increase the number o isitors to low isit groups and skew the proile one is able to build.

lrequency is the measure o the relatie amount o time between isits, and is oten reported
in categories such as once a day,` once a month,` and so on, although reporting the actual
number o aerage days between isits and the distribution o return isitors in each group is
more useul. lrequency is deined much like loyalty:

g6 CIupLer q

reqvevc, i. a vea.vre of tbe actirit, a ri.itor geverate. ov ,ovr !eb .ite iv terv. of arerage tive betreev
retvrv ri.it.. t i. ti/et, vea.vrea iv togicat grov. or a. ai.creet vvvber. of aa,. betreev ri.it.. Regarate..
of re.evtatiov, tbe aata .bovta be aeavticatea.

Recency is related to loyalty and requency, but is usually reported in the context o online
purchasing as the number o days since the last purchase. It can, howeer, be extended to
non-commerce actiity. Simply put:

Recevc, i. tbe vvvber of aa,. .ivce tbe ta.t ri.it ;or vrcba.e) ava .bovta be reortea a. tbe vvvber of
ri.itor. rbo retvrvea after v aa,..

Again, the reader should recognize that recency is a moing target and will eole oer time
depending on the makeup and actiity o the audience.
The Pyramid Model of Web Analytics Data
\ithin \eb analytics there are many dierent sources and types o data. 1hus ar we`e
described a number o dierent data sources, drilling down into speciic data sources or
traic analysis data, namely \eb-serer log iles and client-side data collection. 1here are
also dierent types` o data inherent to each o these data sources, and these data types
proide dierent kinds o inormation.

Speciic to traic analysis, there is a pyramid` model o data, describing a continuum o
data in terms o both olume and alue. At the bottom o the pyramid are hits`-data that
has already been described as both oluminous and mostly useless. At the top o the
pyramid are identiied unique isitors`-data that is likely diicult to collect in large
numbers but is potentially ery aluable to one`s \eb analytics program. ligure 4 describes
the pyramid model, rom bottom to top.


Web unuIyLIcs TermInoIogy g,


Iigure 4: 1he pyramid model of Web analytics data.
Hits
lits are useless as a \eb analytics data type or the most part, data most commonly used by
Network Administrators to determine the load on a single \eb serer within a serer arm.
It would not be unusual or a single day`s analysis o a small \eb site to report that
100,000,000 hits were receied today` which unortunately tells you nothing about actual
human actiity.

lere is some more about what \eb1rends,NetIQ has to say about hits:

Each file requested by a visitor registers as hit. There can be several
hits on each page. While the volume of hits reflects the amount of server
traffic, it is not an accurate reflection of the number of pages viewed.

\ithout a doubt, the term hit` is a remnant rom the days when \eb serer log iles were
nothing more than a tool to monitor the actiity o the serer application, a term which
should no longer be used by business and marketing proessionals.
Page Views
Moing up in the data alue continuum we hae page iews. Page iews are the oundation
or any \eb analytics program. 1he measurement o oerall actiity on a \eb site is
gS CIupLer q

measured in iews, as is the popularity o any gien page or section o a site. Keeping in
mind the deinition o a page iew gien preiously, iews are key to helping measure reach
and acquisition ,discussed in Chapters 11 and 12,.

Page iews are not a good basis or measuring rates, especially conersion and abandonment
rates. 1he problem with using iews or these measurements is that a iew is recorded eery
time a page loads. Consider the ollowing scenarios describing simple three-step actiities:

1. 1en isitors iew page one 10 times, ie moe on to page two and iew page two
ie times, one moes on to page three and iews that page once.
2. 1en isitors iew page one 20 times, reloading the page twice on aerage, ie
moe on to page two and iew page two ie times, one moes on to page three
and iews that page three times.

In the irst example, based on the number o page iews, the step one-to-step-three
conersion rate is 10 percent, with 50 percent abandonment between step one and step two.
In the second example, een though the same number o isitors moe through the process,
the step-one-to-step-three conersion rate is 15 percent ,3,20 ~ 0.15, with 5 percent
abandonment between step one and step two. Although the two conersion rates only
dier by 5 percent, one can imagine how these dierences could be urther aggraated with
larger numbers.

Views should also not be used to determine the number o actual isits or isitors coming to
a \eb site. lor this inormation one needs to moe urther up the pyramid.
Visits
Visits are the irst leel o measuring actiity roughly in terms o human behaior. \hile it is
impossible to learn much about interaction with a \eb site rom page iews other than the
popularity o indiidual pages and content, isits allow us to stitch together actual patterns o
usage. Nearly all \eb Analysis applications report naigation paths in terms o isits. Any
tool that acilitates the measurement o a multi-step process should do so using isits ,see
Chapter 6,.

Visits also proide us inormation about where isitors enter and exit the \eb site, as well as
the olume o traic rom any reerring source.
Unique Visitors
Unique isitors are near the top o the pyramid, proiding details regarding the number o
people coming to the \eb site in a gien timerame. \hile this metric is somewhat limited
by the accuracy o collection, it is better than any type o data lower on the pyramid or
determining the number o human beings isiting a \eb site, especially i one has no way o
uniquely identiying isitors-that is, actually being able to identiy each isitor who comes
to your site ,see below,.

Web unuIyLIcs TermInoIogy gq

Visitors are good or helping understand the global distribution o one`s \eb traic,
retention metrics such as loyalty, requency and recency, whether someone has been to the
\eb site preiously and the technographic breakdown o browsers used to access one`s \eb
site.
Unique Identified Visitors
One thing that a unique isitor measurement is not good or is determining exactly who is
isiting the \eb site. lor this one needs to hae some way to uniquely identiy a isitor in
such a way that this identiication can be tied back to a database o names, and so on. 1his
identiier is oten reerred to as the unique user identiier` or UUID. Depending on how
your \eb analytics package supports the UUID, one can use this data type to drill ery
deeply into one`s understanding o the isitor. 1he UUID can enable one-to-one marketing
and other holy grails` o \eb analytics.

Not all \eb analytics packages support the UUID, and those that do may require a great
deal o setup to take adantage o the data. 1hat being said, depending on your oerall goals
or your \eb site, inding a endor that proides easy access and reporting based on the
UUID may be critical to the success o your \eb analytics program.
6o CIupLer

CHAPTIR g
CONTINT ORGANIZATION
TOOLS


People won't use your Web site if they can't find their way around it.
Steve Krug in Don't Make Me 1hink


Lery \eb site has some kind o organization associated with it, either explicit or implied.
In the best case, this organization is relectie o the site`s inormation architecture or
naigation schema. An example would be a site that has product inormation,` company
inormation` and inestor inormation.` 1heoretically eery page on the \eb site would
it cleanly into one o those categories, perhaps with the exception o the home page.
Perhaps een the \eb designers, when assembling the site, created three directories
relectie o this schema:

/product_information
/company_information
/investor_information
homepage.html

I this were the case, when one is examining a report describing the actiity to speciic pages
,1op Pages` or Most Requested Pages`, there would be enough descriptie inormation
to let you know what page and what part o the site was being accessed. 1his is ine, in
theory, but problems arise when one considers the ollowing conounding inluences:

1. Poorly architected ,or heaily appended, \eb sites
2. Dynamically generated \eb sites
3. \eb sites deried rom content management platorms

In the irst case, imagine the same situation aboe except the pages and directories are now:

/p2002
/ci
/invtinfo
default.html

\hen analyzing content reports, you may be inormed enough about the \eb site to know
that p2002` was in act product inormation,` that ci` was company inormation` and
intino` was inestor inormation` but it is equally likely that you will hae no idea. \ou
also may or may not be aware that deault.html` is a common name or a home page. As
you can see, this nomenclature, while not insurmountable, can cloud the process o reading
\eb analytics data, orcing a translation step that oten times slows the analytics process.
ConLenL OrgunIzuLIon TooIs 61




Iigure S: Sample content groups and page names from the author's Web site,
www.webanalyticsdemystified.com.

In the second case, imagine a dynamically generated \eb site that was built around a single
page. 1he template presents dierent inormation to the isitor depending on the alues
passed into the template ia the URL query string. Lxamples o this could be:

page.asp?pid=1234&cid=8&vid=41A
page.asp?pid=1234&cid=3&vid=41A
page.asp?pid=1234&cid=8&vid=42B

In order to translate this inormation meaningully seeral things need to happen. 1here
needs to be some way to determine exactly what inormation is being generated and map
that back to a meaningul content grouping. lor the sake o this discussion imagine that all
three pages display inormation about a pair o shoes but that the irst URL proides
inormation about aailability, the second URL inormation about pricing and the third URL
compares these shoes to another pair. In this the case, reasonable content groups or these
three URLs may be:

/shoes/availability
/shoes/pricing
/shoes/compare

1he second list is clearly easier to read than the irst, at least in common terms. 1he same
kinds o problems arise in \eb sites rendered rom content management platorms. 1his
demonstrates a key requirement or the content measurement tools that your \eb analytics
application proides:

.v, robv.t ava v.efvt !eb .vat,tic. aticatiov ritt roriae tbe toot. vece..ar, to trav.tate veavivgte.. or
covfv.ivg covtevt ae.critiov. ivto veavivgfvt ava v.efvt covtevt grovivg.. 1be.e covtevt grovivg. .bovta
be fteibte, aefivea b, tbe ovtive bv.ive.. ba.ea ov tbeir veea. ava reqvirevevt., vot b, tbe tivitatiov.
ivo.ea b, tbe aticatiov.
6 CIupLer


1he ability to deine content groups manually,` rather than being stuck with either the
groups proided by your \ebmaster or a lat, meaningless structure, should be required or
the ollowing reason: 1he easier the inormation is to read and share, the greater the
likelihood that the inormation will be acted upon. Imagine that eery time you generate a
content usage report, one that proides critical inormation about how isitors interact with
the inormation or products that you proide, sharing this inormation is met with replies o,
\hat does this line mean` and, Can you explain this report` because people don`t know
how to translate the inormation into meaningul terms.

\hen people don`t understand the inormation they`re gien, the tendency is to ignore or
deer. By requiring reports that are readable and logically organized this common hurdle can
be oercome, increasing the likelihood o success or your \eb analytics program.
Metrics Content Organization Tools Should Provide
1he minimum requirement or the kinds o metrics that can be measured using content
organization tools is that you should be able to see how many page iews were generated or
any content or sub-content or whateer timerame you are ocusing on. Page iews are the
deault measurement or popularity, so more popular or interesting content will usually
receie a greater number o page iews. Beyond page iews, it would be nice to hae the
ollowing inormation or any content group on your \eb site:

Number o isits
Number o unique isitors, broken down by at least day and month ,an option or
custom timerame is desirable as well,
Aerage time spent iewing this content group
Some type o content usage analysis, such as what content groups led isitors to
this particular group, and what content groups they were likely to isit ater this
group

Occasionally it is also ery useul to be able to assign a single page or group o pages to more
than one content group. Imagine an online store that has products normally aailable but
also makes some products aailable as eatured products.` 1he \eb analytics team may
want to ask the question how oten are these products iewed, and how oten are they
iewed as part o the eatured products section` Being able to assign these products to
their normal group ,,products`, but to at.o assign them to a eatured products group
,,eatured products`,, when they are iewed as such, would proide the answer to that
question.
Examples of Different Content Organization
Strategies
1here are two ery common content organization strategies and one that is less common,
but arguably more powerul. 1hese are commonly reerred to as directory structure,`
ConLenL OrgunIzuLIon TooIs 6

inormation architecture` and business objectie` models o organization. 1he ormer is
generally more I1 or \ebmaster-centric, relecting the actual distribution o documents on
your \eb serer. 1he latter is ery business-centric, relecting a natural language expression
o your goals and reasons or haing a \eb site and how each \eb document contributes to
those expressed goals.
Directory Structure Content Organization
I you were to simply install your traic analysis application and begin collecting data, either
by parsing your \eb-serer log iles or by dropping deault code on your pages, you would
ery likely be looking at your data organized by directory structure. As companies irst begin
to establish \eb analytics programs internally, a directory structure-based organization is
oten employed, i or no other reason than it is usually easy to set up. Lery \eb site has a
directory structure o some kind-some ery deep, some ery lat-but the nature o
l1ML dictates some type o content structure by deault. Some examples:

www.csd.toshiba.com/cgi-bin/tais/su/su_sc_home.jsp-the document
su_sc_home.jsp` is organized into a hierarchical structure cgi-bin,tais,su,`
which at least supericially suggests that the document is dynamic in nature ,the
cgi-bin` is common to dynamically generated pages.,
www.cnn.com/2003/US/0J/J8/sproject.irq.us.protests/index.html-the
document index.html` is organized into a hierarchical structure
2003,US,01,18,sproject.irq.us.protests` which appears to relect the date this
document was published and its general topic.
www.apple.com/switch/questions-the deault document is organized into a
hierarchical structure switch,questions,` which is also a relection o the
inormation architecture on this site.
www.broadvision.com/One1oOne/SessionMgr/home_page.jsp-the
deault document home_page.jsp` is organized into an application structure
One1oOne,SessionMgr,` which is common to nearly eery page` on this \eb
site.

Nearly eery \eb site has one o these our types o organization. 1he irst and last
examples presented aboe ,rom the 1oshiba and BroadVision \eb sites, present a
common problem associated with this type o content organization. In dynamic
enironments content structures are oten ery lat` owing to the act that a single or small
number o scripts are designed to delier a large number o dierent pieces o content rom
a database or data structure.
Information Architecture Content Organization
A \eb site`s inormation architecture describes the relationship between content and
naigation. Louis Roseneld, in a keynote presentation at the American Society or
Inormation Science and 1echnology summit, said:

6q CIupLer

Inormation architecture inoles the design o organization, labeling, naigation
and searching systems to help people ind and manage inormation more
successully.`

\ou can see the inormation architecture or any \eb site simply by examining the
naigation system presented on any gien page. A ew examples:

www.circuitcity.com -the primary naigation along the top o the Circuit City
home page includes Llectronics,` Games,` Moies` and Music.` \e can
easily iner that Circuit City has made a decision to organize their content around
broad product lines ,ligure 6,.
www.ebay.com -the primary naigation or eBay users includes Browse,`
Search,` Sell,` My eBay` and access to the eBay Community,` as well as
seeral secondary areas ,pay,` register,` and so on, ,ligure ,.
www.hotels.com -the top line naigation or lotels.com isitors includes
lotels,` Suites & Vacation Rentals,` Vacation Packages,` Deals & Specials`
and Destinations & Interests` ,ligure 8,.



Iigure 6: Lxample of top-line navigation on Circuit City Web site
(www.circuitcity.com).



Iigure 7: Lxample of top-line navigation on eBay site (www.eBay.com).



Iigure 8: Lxample of top-line navigation on Hotels.com site (www.hotels.com).

1he goal o inormation architecture is to assist isitors in more quickly inding the
inormation that they hae come to the \eb site seeking or that they may be interested in.
1his same goal is easily extended to the concept o content organization within your \eb
analytics package.

ConLenL OrgunIzuLIon TooIs 6g

Using the Circuit City \eb site as an example, i you were to click rom the home page to
Computers, Printers and PDAs,` select Desktop Computers` rom the list and click a link
to any o the desktop computers aailable you would theoretically be in an inormation
group deined as:

Home > Computers Printers and PDAs > Desktop Computers

1his inormation group can easily be translated into a content group to help Circuit City
understand the low o isitor traic to pages in exactly the same way they hae chosen to
present those pages. Len though content on the site is dynamically rendered, and so is
entirely lat, containing no inormation about Computers` or Desktop Computers` in the
URL proper, most \eb analytics application proide tools to translate the inormatie part
o a URL into a logical and useul content group. In this example, i Circuit City were to opt
or an inormation architecture-based content organization strategy, we would hope that
their \eb analytics application would support the creation o a content group that is
roughly:

/Home/Computers Printers and PDAs/Desktop Computers/Product Views

lrom this, Circuit City would be able to see how many isitors and iews each leel in this
hierarchy ad rom top to bottom, how much time isitors spent at each leel, and so on.

As online businesses` understanding o \eb analytics matures, the inormation architecture
iew o content becomes the most common type o content organization employed. 1his
iew is much more useul than the directory structure iew in that it better ties \eb analytics
data back to the \eb site using business-releant language and structures.
Business Objective Content Organization
A content organization strategy that is emerging in popularity is that o business objectie`
organization. Very ew companies hae \eb sites just or the un o haing one. Most, i
not all, companies build \eb sites and \eb-based applications to accomplish speciic goals
and objecties. At a high leel, these business objecties are typically airly straightorward.
lor example:

1. Sell products and,or serices
2. Gather qualiied leads or marketing
3. Decrease internal costs by sericing customers ia less expensie channels
4. Attract loyal and requent isitors ,to increase adertising reenue,
5. lacilitate the isitors search or inormation

As you can see, each o these objecties is really make more money` ,in the case o items 1,
2 and 4, or make money by saing money` ,in the case o items 3 and 5,. Lach o these
high-leel goals can be described in abstract terms by speciic online actiities in which
isitors can engage. Lxamples o such actiities, or item 1, selling more products and,or
serices, are:

1. lind products or serices
66 CIupLer

2. Research and compare products or serices
3. Request more inormation about products or serices
4. Purchase products or serices

Now consider each o these actiities in more concrete terms, thinking o them as a
container or \eb pages. Lach will likely hae an exclusie, or nearly so, set o \eb pages
that deines each actiity. lor item 1, ind products or serices, you would include the pages
that hae lists o products aailable or purchase online, perhaps including your home page.
lor item 2, research and compare, you would hae pages with speciic inormation about
said products or serices. lor items 3 and 4 there are likely a small number o pages and
orms that acilitate the isitors` request or inormation and purchase processes.

1he concepts presented here regarding a business objectie organization or your \eb site
are closely related to the practice o using personas` to ensure that your \eb site is built
with a speciic set o users in mind. Personas, irst presented by Alan Cooper in 1be vvate.
.re Rvvvivg 1be ..,tvv in 1999, hae become more and more popular among high-end
design irms. \hen incorporated into application design processes, personas are a powerul
tool to help ocus design eorts and ensure that application low is consistent with primary
user needs. 1he relationship between business objectie content organization and personas
is simply this: I you use personas to design your \eb site or application you should be
absolutely sure that the goals o your key persona,s, are in line with your business objecties.
I this is not the case then either A, you hae misidentiied key personas or B, you hae
misidentiied your business objecties. Additional inormation about the relationship
between personas and business objecties is presented in Chapter 16.

Consider the example o BackcountryStore.com, which exists online primarily to sell high-
quality outdoor gear or the serious enthusiast, but is also interested in improing the
customer experience through search eatures, online communities and education.
BackcountryStore.com has used litBox Lnterprise to organize their content groupings
around six major business objecties:

1. Selling products online
2. lacilitate searching or the products they sell
3. Introduce BackcountryStore.com to new isitors
4. Grow their online community
5. Lducate their customers and potential customers
6. Gather leads or their online marketing group

Lery page on the \eb site can be categorized as belonging to one o these six main
objecties, allowing management to easily see which pages and sub-content groups
contribute most to the success o each business objectie.


ConLenL OrgunIzuLIon TooIs 6,


Iigure 9: Business objective content groupings for BackcountryStore.com.

I the owners o BackcountryStore.com want to know how likely men`s ski jackets` are to
contribute to the oerall goal o selling products online they can simply drill-down into the
Sell Products` content group until they ind Men`s Ski Jackets`.

/Sell Products/Mens Clothing/Mens Winter Clothing/Mens Ski
Jackets



Iigure J0: Drill-down to Men's Ski Jackets in Sell Products on
BackcountryStore.com.

As you can see in ligure 10, or the timerame under examination 220 daily unique isitors
generated 888 page iews o Men`s Ski Jackets, nearly our iews per unique isitor. In this
type o content organization iew, the owners o BackcountryStore.com can easily see which
product lines are most popular-and thus most likely to be purchased through their \eb
site.

Orgaviivg covtevt arovva a bv.ive.. ob;ectire voaet attor. ,ov to botb vea.vre tbe attractiov of age. ov
,ovr !eb .ite a. art of togicat grov. and .ee bor ri.itor.` v.e of tbo.e age. covtribvte. to ,ovr oreratt
bv.ive.. .vcce...

1his model immediately inorms eeryone responsible or \eb site success about the
number o isitors who were in the pool o candidates to purchase products, generate leads,
and so on. Ater strategic changes are made these numbers can be compared to earlier
numbers rom the same content group to quickly measure the eect o these changes. 1his
model allows managers to determine the success or ailure o changes to speciic actiities
online, proiding easily understood data to answer questions such as, Do more people
moe rom research products` to purchase products` actiities i a dierent color or layout
is used on product description pages` and, \hich are our most popular products or
product lines`

As more and more businesses accept that their \eb site is simply a channel to support
existing business goals, the business objectie model or content organization will become
more widely used. 1o take adantage o this model one must truly iew the \eb site as an
6S CIupLer

integral-and integrated-part o the oerall business and be prepared to deal with it as
such. Many businesses that are serious about succeeding online see the alue in the business
objectie model or content organization and are already implementing it successully.
Risks Associated With Content Organization
1he major risks associated with content organization are a lack o lexibility on the part o
your reporting application and diering opinions about how content should be organized
within your organization. In terms o lack o lexibility, it is not uncommon or \eb
analytics applications to impose rules on how content groups are named or ormed. \eb-
serer log-ile analysis applications are oten limited in their abilities to dynamically name
content groups, oten requiring some kind o translation ile or table to map inormation
deried rom URLs to clean, readable content groups. Client-side data collection
applications are typically more lexible in terms o deining and setting up content groups
but oten do not allow you to go back and rename content groups ater they hae been
named and data has been collected ,although there are exceptions,. Additionally, outsourced
applications oten require a single page to belong only to a single content group ,again, there
are exceptions,.

None o these risks are showstoppers, nor should they be treated as such. In the irst case,
using a \eb-serer log-ile analysis application, the analytics and,or inormation technology
team will hae to pay careul attention to the relationship between the \eb site`s directory
structure and how internal users would like to see pages mapped into content groupings. I
the groups responsible or each do not communicate clearly with each other, or i the \eb
site directory structure is signiicantly changed, there is always the possibility that content
mappings will be wrong or simply dropped rom an analysis.

In the second case-where once content groups are named and data has been collected
applications do not allow you to go back and change their names or relationships-this
usually only poses a problem i you redesign your \eb site or inormation architecture.
Consider the situation where you hae organized your content data collection around your
inormation architecture. I your business decides to change the inormation architecture,
and you are unable to go back and rename content groups or historical data, you will be
orced to decide between renaming the content groupings with the new architecture or
maintain the content grouping as is.`

1he third case-where a single page is only allowed to belong to a single content group-
leads directly into the second major risk in content organization, diering opinions about
how content should be organized. Imagine three sets o stakeholders trying to decide what
content organization strategy should be applied to the \eb site. 1he inormation
technology team will likely preer a model based on the \eb site`s directory structure
because the data makes sense to them. 1he marketing team may preer a model relectie o
the inormation architecture on the site, which is dierent than the directory structure, so
that they are able to quickly assess isitor interest in each major category o inormation the
company presents online. 1he executie sta will hopeully preer a business objectie`
model o organization so that they are able to quickly determine how the \eb site
contributes to oerall business success. In this situation, i your \eb analytics package only
ConLenL OrgunIzuLIon TooIs 6q

supports a single content structure or your site, who decides which model is applied Most
likely the executie team will get their business objectie` model, with the other groups
being asked to make due.`

I the \eb analytics package your organization uses supports multiple content groups per
page, either through inherent technology or the ability to track multiple accounts or proiles
per page, these kinds o problems can be aoided. A major actor in the success o any \eb
analytics program is the ability to analyze data in a context that you are amiliar with. I you
are orced to constantly translate data rom a ormat you are unamiliar with, compounded
by the act that \eb analytics in itsel is not easy, you are likely to be less ocused or less
interested in the inormation contained therein and so less likely to act.

Web Analytics Tip: Why Are Content Groupings Important?
Many customers I`e had the pleasure o working with in the past hae treated content
grouping as an aterthought when implementing their analytics reporting. laced with the
choice between taking the time to think out how their content should be organized and
implementing around that, and simply using the deault content implementation, many
clients hae opted or the latter. \hile this does let them rapidly implement their analytics
solution and start collecting data, it is ery common that they will come back to me later and
complain that the data is not organized ery well and is hard to use.

lmmm, really

Being the consummate proessional I always ight the temptation to say, I told you so,` and
giggle-nobody wants to hear their consultant talk smack. I usually proide them a gentle
reminder that we`d discussed this at the time o implementation and that they made a tactical
decision to choose rapidity oer detail. 1hen we go back and ormulate a plan to re-
implement using either the inormation architecture or business objectie content groupings
described in this chapter.

1he lesson here is to not attempt to cut corners when implementing your analytics solution,
especially regarding your content structure. Most o the work inoled is mental-sitting
down as a team, thinking about the content you hae and mapping pages to the content
groupings that you`d like to see. 1echnically, translating this content mapping can be more
or less diicult, depending on the application enironment you hae, how your pages are
published, and so on, but once the map is created all that is let is busywork. 1he time and
money you spend haing someone in your organization implement useul and well thought
out content groups will be returned many times when the data turns out to be easy to use
and easy to understand.





,o CIupLer 6

CHAPTIR 6
PROCISS MIASLRIMINT
TOOLS


If you want a great site, you've got to test. 1esting reminds you that not everyone
thinks the way you do, knows what you know, uses the Web the way you do.
Steve Krug on Several true things about testing in Don't Make Me 1hink


As briely described in the section on business objectie content organization, most
business-class \eb sites are simply a set o actiities that users hopeully engage in to
contribute to business success. Since more oten than not these actiities are ordered,
multiple-step processes, the \eb analytics package that you choose should proide tools to
acilitate process ,or scenario, measurement. 1hese tools are reerred to in a ariety o ways
by endors-scenario analysis, conersion unnels, process analysis, unnel analysis or all-
out reporting-but regardless o the name, process analysis tools should allow you to
measure step-by-step abandonment and oerall conersion rates or any multi-step process.

1he alue in measuring and understanding user interaction with your actiities and processes
is simply this:

1be vo.t effectire .trateg, for v.ivg !eb .vat,tic. aticatiov. ava tbe aata tbe, roriae to trvt, ivrore
tbe to or bottov tive. for ,ovr ovtive bv.ive.. i. va/ivg v.e of aata roriaea b, roce.. vea.vrevevt toot..
Data roriaea b, roce.. vea.vrevevt toot., iv tavaev ritb .vcce..fvt aticatiov of tbe covtivvov.
ivrorevevt roce.., cav bet ,ov aravaticatt, ivcrea.e ,ovr covrer.iov rate.. vcrea.ivg ,ovr /e, covrer.iov
rate. i. tbe ea.ie.t atb to va/ivg vore vove, ov tbe vtervet.

Consider the most common multiple-step process that people ocus on online, the purchase
process. Most purchase processes begin with a shopping cart and low into a checkout
process that requires the isitor to proide a ariety o inormation. In most cases this
process is three or our steps long, including the shopping cart, but longer purchase
processes are not unheard o. 1ypically a \eb site will be able to attract a large number o
isitors to iew products and many o those isitors will place an item in the shopping cart.
1he unortunate reality is that signiicantly ewer isitors will then click the checkout`
button and ewer still will complete the actual checkout process.

\hile most \eb analytics packages will proide inormation about the number o isitors
that place an item in the shopping cart and the number o isitors who complete the process,
process analysis tools take multi-step analysis a step urther. Process analysis tools should
enable the measurement o loss between any two steps and the total loss rom start to end.
Knowing which steps hae the highest abandonment will help your \eb design team know
Process MeusuremenL TooIs ,1

where to ocus eort. 1he ability to measure the eect o changes to any step by obsering
changes in the abandonment rate at that step helps to alidate the alue o changes. 1he
reduction o loss at any step in a multi-step process will hae a net positie eect on the
oerall conersion rate or the process unless loss is introduced at other steps in the process
at the same time.



Iigure JJ: An example of how one vendor helps their customers visualize the
abandonment and conversion through an online activity.
An Important Note about Process and Conversion
Rates
1wo important things to keep in mind are that A, the purchase process is by no means the
only important multi-step process worth measuring and B, there are vav, dierent
conersion rates that can be measured on any \eb site. Regarding the irst point, some
example actiities that can be measured with an eye towards oerall improement include:

lorms used or lead generation
Online purchase processes
Search unctions
Naigation to support materials or inormation
Downloads

Basically any actiity that inoles isitors iewing inormation on at least two successie
pages can be measured as a process. 1hat being said, the most aluable processes you
should measure are those that collect some item o alue rom your isitors-those items
that beneit your business.

, CIupLer 6

Regarding the second point, any \eb site has a number o dierent conersion rates. One
rate is the number o isitors who isit the home page and then purchase. Another is the
number o isitors who iew a product or serice description page and then place the item in
a shopping cart. Another rate would be the number o isitors who click checkout` and
then complete the purchase process. \et another would be the number o isitors who
successully ind the customer support content they are looking or. 1he challenge o the
business owner is iguring out which ones make the most sense to measure.

A complicating actor when calculating conersion rates is the disposition o your isitors at
the time o their isit. I you calculate your site` conersion rate as the number o total
unique isitors diided by the number o actions ,orders, leads, and so on, you are making
the assumption that all o your unique isitors are potential conerts. In many instances this
is not the case-customers who hae preiously purchased may simply be looking or
technical support, isitors who hae preiously submitted personal inormation may be
simply double checking something they read, your competitors may be doing competitie
analysis on your \eb site, and so on. \hile it is impossible to truly know an anonymous
isitor`s intentions, you need to keep this complicating actor in mind and attempt to
minimize the eect when calculating conersion rates.

linally, conersion rates are numbers that are ery requently bantered about, oten
incorrectly. Many studies hae been published detailing aerage conersion rates or online
businesses and oten companies try and compare their rate to these published rates.
Problems associated with making this type o comparison include:

Method o measurement-i the published rates were collected rom \eb-serer
log-ile data but your business uses page-tagging technology the data is similar, but
not same.
Deinition o actiity-i the published rates are the measure o isitors to the
home page who then successully purchase but you are only able to measure
isitors who place an item in a shopping cart who purchase, the conersion rates
are apples and oranges and not alid or comparison.
Statistic used to calculate rates-i the published rates are based on the data
generated rom page iews but you are able to collect data based on isits or unique
isitors, the data is not alid or comparison. 1he author`s suspicion is that many
groups publishing reports comparing conersion rates simply ask businesses or
said rates, not the statistics used to calculate the rates.

In general it is best to simply ensure that you A, hae an accurate method or measuring
abandonment and conersion in multiple-step actiities and B, compare your \eb site only
to preious data collected on your \eb site using the continuous improement process
described in Chapter 1 o this book.
Metrics Provided by Process Measurement Tools
Regardless o the application you choose to acilitate your \eb analytics program, the
process measurement tools the application proides should allow you to do the ollowing:

Process MeusuremenL TooIs ,

Deine multiple-step processes in terms o steps in the process
Assign \eb pages to a gien step or steps
1rack the low o isitors rom step to step, measuring abandonment between
steps
1rack conersion, that is, inormation about isitors who complete a process as
designed`
Learn as much as possible about actiity at each step, including aerage time spent
on the step, pages leading to the step, pages leading rom the step and links that
isitors use to leae the process entirely

At a bare minimum, or any step in the process you should be able to see:

Number o isits or unique isitors, but isits is preerred
Number o page iews to pages in the step, both as an aggregate and indiidually
Aerage time spent as an aggregate or the step
Links most likely to be clicked leading to abandonment rom the step

1hese data are enough to create a ery complete understanding o actiity through any
online unnel or process. Some nice to hae` eatures would be:

Side-by-side process isualization tools, or the ability to compare two days, weeks
or months graphically without haing to open two applications or browser
windows.
Notes, allowing or easy sharing o inormation regarding changes that had been
made to the process regarding when and why any modiications were made.
Measurements o isits, page iews ava unique isitors, allowing greater lexibility
or comparison to legacy processes or measuring process abandonment and
conersion. laing these data also allows you to calculate the number o times the
aerage unique isitor is likely to engage in the process ,isits diided by unique
isitors ~ isits per isitor per timerame, see Chapter 12,.

As long as the application you choose proides or the bare minimum measurements you
will be set to measure and improe the low through any multi-step process your \eb site
may hae.
Example Uses for Process Measurement Tools
As described aboe, process measurement tools can be used in a number o common
situations online to better understand where isitors are vot successul. I you think you hae
an actiity on your \eb site that is a good candidate or measurement using these types o
tools, but you are not sure, ask yoursel this question:

f cav get vore ri.itor. to covtete tbi. actirit,, ritt be vore .vcce..fvt ovtive.

I the answer to the question is yes` then the actiity should deinitely be measured. 1he
most common online actiities that businesses are measuring include:
,q CIupLer 6


Purchase processes, to understand where potential customers drop out o the
process and monitor changes in abandonment and conersion rates,
Registration processes, especially those that are many steps long, to start to
understand what questions you may be asking potential registrants that are driing
them away,
Demonstrations o products or serices that are many steps long, to ensure that
prospects are seeing the entire length o the demo and not just getting part o the
picture,
On-site search tools, especially those designed to lead isitors either to product
purchases or support content and inormation, to ensure that isitors are
successully able to use your search tools,
1he use o any online application that you may hae built speciically or customer
use to ensure that customers are getting ull alue out o the tools you hae built.

One o the best examples o how process management tools are eectiely used is
demonstrated by BackcountryStore.com and its eorts to improe their site-wide conersion
rate. In early 2002, the checkout process on BackcountryStore.com included a required
registration step or eery isitor who wanted to make a purchase. \hen anyone clicked the
Checkout Now` button they were not taken to the irst step in the checkout process, where
they would proide shipping and billing inormation. Rather they were taken to a page ery
similar to the one presented in ligure 12.



Iigure J2: Required registration step on BackcountryStore.com from early 2002.

\hile requiring registration is not always a bad idea, when BackcountryStore.com applied
process measurement tools it was able to see signiicant abandonment at the registration step
,ligure 13,.


Process MeusuremenL TooIs ,g


Iigure J3: Process measurement tools employed by BackcountryStore.com.

1o make a long story short, BackcountryStore.com made the decision to remoe this step
and proide customers a dierent path to register and log in i they were inclined ,ligure
14,. 1he end result o making this change, combined with a small handul o usability
improements that were also identiied using \eb analytics tools, resulted in a 20 ercevt
ivcrea.e iv tbe .iteriae covrer.iov rate. \hat is more interesting is that, during
BackcountryStore.com`s twice-yearly sales, the lit in conersion rate is estimated to go v b,
a. vvcb a. 0 ercevt rbev covarea to recbavge vvvber..



Iigure J4: New method for logging in as a registered user on BackcountryStore.com.

\hile not eery online business employing process measurement tools will be as successul
as BackcountryStore.com, you can be assured that some will be much, much more successul
increasing conersion rates, depending on the scope o the problem the isitors are
experiencing.


,6 CIupLer 6

Web Analytics Tip: When to Require Registration
Regarding the BackcountryStore.com example, clients will oten ask me whether I
recommend required registration in online purchase processes. 1he catalyst behind requiring
registration appears to be a combination o actors:

Marketing people want to hae the registration inormation or a ariety o reasons ,opt-in
mailing, database marketing, and so on,.
1he commerce application used to drie the purchase process requires this inormation.
Amazon.com does it so it must be the right thing to do.

Consider these actors in order-o course marketing people want to hae this inormation,
marketing people tire for ivforvatiov tbat ritt tet tbev /vor tbeir cv.tover. better. Unortunately the
potential customer has learned to be wary o marketers thanks to a constant deluge o mail
spam, phone spam and email spam, as well as the recent emergence o identity thet. 1he
unny thing is that the potential customer will likely end up giing you most, i not all, o the
inormation you are trying to elicit in the course o the checkout process i you just let them
get there.

Concerning commerce applications that hae hard and ast rules about how the purchase
process works-sotware is code and code can be changed. Rather than run the risk o
alienating some percentage o people who are actiely engaged in the process o trying to
gie you money, and so help you grow and maintain your business, you should seriously
consider modiying the code that dries your purchase process, or implementing new code
that proides you more lexibility. Successul online businesses verer let themseles be held
hostage by code or the programmers who create code.

Regarding the Amazon.com does it` phenomenon, authors and analysts like me are guilty
o perpetuating this problem to some extent. Because Amazon.com has been so
phenomenally successul on the Internet, eeryone has a tendency to look to them and see
what they are doing. 1he problem is that Amazon.com has spent what are likely millions o
dollars perecting processes that are the best or tbeir cv.tover., not necessarily yours.
Amazon.com has created a one-stop-shop or all kinds o purchases and thusly has
deeloped an audience that is likely to return and purchase quite requently. It is this
requency o purchase that dries the registration requirement online, not the desire to hae
inormation.

In general i you are debating whether to require registration or not you should ask
yourseles the ollowing questions:

low oten do our customers make online purchases I the answer to this
question is hmmm, we don`t know` or less than three times a year` then ao vot
require registration.
Can we collect the same inormation in the course o the purchase process and
then automatically register the customer I the answer to this question is yes`
then ao vot require registration. I the answer is no, we cannot` you should let the
answer to question 1 drie your decision.
Process MeusuremenL TooIs ,,

Can we oer our customers any monetary incentie to register in the purchase
process, such as ree shipping or a discount on the total order I the answer is
no` to this question then ao vot require registration.

In summary, nine times out o ten when someone asks me i I recommend registration in
purchase processes I answer no`, ortunately, in nearly all instances when clients hae
ollowed my adice and remoed registration the results hae been positie.

Risks Associated with Process Measurement Tools
1he biggest risks associated with process measurement tools are A, the tools you hae
chosen are not able to proide accurate, useul data and B, een though you hae the data
you are not entirely sure what it is telling you. In the case o the ormer, your business .bovta
vot inest in a \eb analytics package that proides process measurement .otet, based on
page iew` metrics. 1here are multiple inherent dangers in this type o process
measurement, most notably the eect o reloads and multiple page iews and,
undamentally, i a endor is telling you that page iews are a ine way to measure process
abandonment` they are also telling you we hae no idea what we`re talking about.`

Process measurement statistics should always be based on isits` or unique isitors,` with
isits being the preerred metric. Visits are preerable to unique isitors because unique
statistics are always tied to a timerame, such as a day, and i a isitor attempts to repeat the
process more than one time in a day this inormation is potentially lost. Visits are the single
most accurate statistic or measuring process abandonment because the isit captures
inormation about both multiple isits in a single day,week,month and also the isit has a
timeout that lets us understand how oten a isitor is attempting a process. \ou are
encouraged to reisit Chapter 4 or more on the dierences between a isit and a isitor.

Regarding not knowing what the data is telling you, this is the \eb analytics catch-22.`
\hile you can be assured that i you are vot cottectivg \eb analytics data regarding actiities
on your \eb site then you hae no chance o knowing how or where improements can be
made, the author has no way to ensure that i you are collecting this type o data that it will
be easy to use. Process measurement tools are designed to help you generate baseline data
regarding isitor success while they moe through designed actiities. 1his data will easily
identiy the steps in an actiity that isitors hae the most trouble with but not necessarily
tell you rb, these isitors hae trouble.

lortunately you hae already read about the continuous improement methodology in
Chapter 2 o this book. By ollowing a measure change measure` strategy you will be
able to identiy the high abandonment steps in a multiple-step process, hypothesize about
possible reasons or abandonment ,Visitors don`t want to be asked their mother`s maiden
name,` Our next button appears below the old and many isitors do not een see it`,,
make changes based on well thought-out hypotheses and then test the eects o those
changes. By ollowing this strategy you will ery likely be able to iteratiely identiy the
problems isitors are haing and remoe them rom your process.

,S CIupLer 6

1wo bits o adice:

1ry and let go o pre-conceied notions o how your isitors behae.` Most oten
these assumptions are alse and will oten limit your ability to hone in on the
changes that will yield the biggest gains.
Don`t expect miracles. Nobody has 0 percent abandonment rom step-to-step and
a 100 percent conersion rate, nobody. 1he work you are doing using process
measurement tools is attempting to vivivie to.. not vaivie ftor. It is a subtle but
important dierence.
About A/B Testing
One o the most requent uses or process measurement tools is to conduct A,B tests to
determine whether changes actually hae a positie and measurable eect. I you make a
small number o changes to pages or content in a known multiple-step conersion path that
you can send some percentage o your traic through, you can measure the dierence on
the changed, or B,` path, compared to the original process. 1his strategy is oten aored
because you are able to maintain your current conersion rate or at least part o your traic,
just in case the changes are detrimental. I you are planning on running A,B tests on your
critical conersion processes, here are a ew tips that can help ensure that you get
meaningul, measurable results:
Only Change One Variable at a Time
\eb designers oten orget that i you change more than a single ariable in an analysis like
this you will be unable to determine whether the changes are due to change 1` or change
2` or the combination o changes 1 and 2.` Apply the scientiic method in these instances
and make sure that you can isolate the source o any measured dierences. Please see
Chapter 2 and the section on continuous improement or additional guidance about making
and testing changes.
Understand the Process for Diverting Traffic
A common mistake or those running the test is that they do not ully understand the
mechanism or dierting traic and so are not getting accurate counts o traic olume.
1here are many strategies or dierting traic ,application layer, load balancers, scripts on
pages, but a common problem with A,B testing is that these are not well understood. 1he
essential goal o the traic diersion mechanism is to be able to redirect a known percentage
o isitors through this modiied process. Ideally, the percentage o traic redirected is
easily changed without haing to radically modiy pages.
Determine Accurate Measures of Volume
I you are able to get a isitors per page` count rom your \eb analytics package it is highly
recommended that you do so or the irst step in the test. laing this number allows you to
ensure that you are actually getting the percentage o traic moing through the unnel that
Process MeusuremenL TooIs ,q

you expect. I you are lowing 50,50 through A` and B` you should see near equal
numbers o isitors to the irst page in the process, running the test until the numbers hae
conerged to almost exactly 50 percent through each path.
Analyze Carefully
1his is ery important! lopeully you hae robust process measurement tools at your
disposal and you are able to measure abandonment at each step in your multiple-step
process. I you see a dierence in the conersion rate or the B` test you need to ensure
that this dierence can be attributed to the .te tbat ,ov cbavgea in the process. I you are
getting small dierences rom many dierent steps then you are not testing anything, all you
are seeing is ariation in the quality o isitors lowing through the A` and B` tests
,noise`,. \ou need to run your test until all o the change can be attributed to the exact
step that you modiied in the process, or until you are conident there is no change.
Run a Null Test
Once you hae set eerything up and are ready to test, take the time to low 50,50 traic
through the exact same pages. Doing so will let you eriy that you get the same conersion
and abandonment rates and that the measurement tools are correctly set up. I you are not
getting nearly exactly the same rates ,within 5 percent, or both tests then something is
wrong and any data you get rom an actual test will be inalid. I this kind o problem is
happening you should check that A, you are sending people into the tests eactt, the same
way ,so you are not pre-selecting participants and sending more qualiied people down one
path or the other, and B, you are sending enough people into the test. \ou will see
dierences in rates until you hae a reasonable sample. Depending on the olume o traic
your \eb site receies, attaining a reasonable sample can take some time. Beliee that this is
time well spent and run a proper null test prior to making changes, i you do you will be
more conident in the resulting data.
Run Your Test Until You Are Sure the Results Are Real
A common mistake in A,B testing is not running the test or long enough to determine
actual dierences. Oten \eb designers will take the time to set up an A,B test only to run
it or just a ew hours because they see dierences and declare B the winner!` 1his is
dangerous because until you hae a representatie sample you cannot be entirely sure that
B` really does increase or decrease your conersion rate. It is not uncommon or a rate to
be higher or lower at the onset o a test and then hae that number drit back towards the
original rate as the sample grows larger. lor inormation on what constitutes a statistically
releant sample` consult the college statistics textbook you hae kept on your bookshel all
these years or exactly this reason.
Consider Segmenting Test Subjects
Depending on the segmentation tools that your application proides, segmenting test
subjects will allow you to monitor the actiity o test subjects when they return to your \eb
site. \hile these isitors may behae no dierently than non-test subjects, you won`t know
So CIupLer 6

unless you tag` them. A reasonable question that could be answered is what percentage o
isitors that went through the B` test returned to the \eb site within three months`
laing this inormation, and being able to compare it to your site-wide loyalty metrics, will
let you better understand the long-term eects o the A,B test.

I you ollow these simple guidelines you will be able to take adantage o one o the more
complex applications o the continuous improement process. 1he upside o A,B testing is
that i your changes are detrimental you will not alienate att o the isitors moing through
the process as you would i you simply changed the process or eeryone and then watched
the eect. 1he downside is that A,B tests are not always easy to set up properly and do
require some additional thought during the analysis phase.

\eb analytics endors are moing towards proiding tools that support A,B testing such as
multiple-timerame iews and side-by-side unnels or iewing data. \hile these
isualization tools are great, keep in mind that they are only that, isualization tools and not
a substitute or careul design and set-up as described aboe. Any \eb analytics application
will allow you to conduct A,B tests, een i the application does not support process
measurement tools, although the testing becomes much trickier.
VIsILor SegmenLuLIon TooIs S1

CHAPTIR ,
VISITOR SIGMINTATION
TOOLS


Retailers, on average, generated $7 in revenue for every dollar invested in
acquisition and $2S for every dollar spent on retention, nearly double the previous
year's averages of $4 and $J3, respectively.
Boston Consulting Group report on Retailing Online: Coming of Age


Lery online business has many dierent isitor alue groups, or isitor segments, making
use o their \eb site eery day. Some isitors are simply browsing the \eb site, trying to
orm an opinion o what the business does. Others hae actiely engaged with the \eb site
by proiding some type o personal inormation. Still others are committed to the \eb site,
haing made a purchase or downloaded a document. It is a mistake to lump all isitor alue
groups together and use that inormation to try and understand how to best sere the groups
proiding the highest alue-thus the need or isitor segmentation tools.

Visitor segmentation tools are a general term or some technology that allows \eb analytics
teams to say, Based on some behaior, I want to track this isitor as a member o a speciic
segment or group o segments.` Some \eb analytics application endors proide this
unctionality by allowing you to place isitors in a segment at the time they perorm a
speciic actiity. Others allow you to mine data already collected and place isitors in a
segment based on past actiity or isit data.

Regardless o the strategy used by the \eb analytics package you employ it is important to
hae the ability to segment isitors. Visitor segmentation allows you to answer important
questions like:

Do dierent segments o isitors behae dierently in terms o the pages or
products that they iew
Do dierent segments o isitors spend more time on your \eb site or return
more or less requently than others
Are dierent segments o isitors conerting at higher or lower rates than others
low long does it take, on aerage, to moe a isitor rom a lower alue group to a
higher-alue group and can that time be inluenced by making changes to the \eb
site

1he undamental use or isitor segmentation tools is to better understand how your most
aluable isitors interact with your \eb site. 1his inormation will allow you to ensure that
S CIupLer ;

any changes you make will not negatiely impact these isitors. \ou can also use this type o
inormation to gie preerence to content that higher-alue groups are interested in,
hopeully attracting more members o these higher-alue groups.
Metrics Provided by Visitor Segmentation Tools
In a perect world \eb analytics endors would collect data on all aailable metrics or
isitor segments, proiding complete coerage or all segments as well as the larger group o
isitors as a whole. Because this is not a perect world, likely you will be required to settle
or ewer metrics than you would like. In act, many \eb analytics endors do not support
isitor segmentation at all. \ou don`t necessarily need to aoid these endors but the author
does recommend questioning them ery careully about their uture plans to support
segmentation prior to signing any contract.

\ith that in mind, the ollowing are metrics that are useul in understanding the actiity o
segmented groups o isitors:

1raic statistics such as page iews and isits
Daily and monthly unique isitors
Pages and contents iewed
1ime spent on site and content distributions
Reerring source inormation ,i present,
Campaign source inormation ,i present,
Process conersion inormation ,starters, conerters, i applicable,
Commerce inormation ,i applicable,
Segment inormation, such as segment conersion and new s. returning members
o the segment

Lach o these measurements contributes to building a more detailed picture o how dierent
alue groups interact with your \eb site. Knowing which pages and content dierent
isitor segments iew most oten will help you better target members-and potential
members-o that group. Campaign and reerring source inormation will let you know
how best to reach members o a group. Process measurement inormation will let you
identiy which segments are likely to begin an actiity but not complete it, and which
segments are most likely to complete business-critical actiities.

Again, no endor known to the author at the time o writing supplies all o the data points
listed aboe. 1hese recommendations are a guide to help you make better decisions when
choosing an analytics package and are proided in hopes that eentually all endors will
support robust isitor segmentation tools.
VIsILor SegmenLuLIon TooIs S

Examples of Uses for Visitor Segmentation Tools
1he two best uses or isitor segmentation tools are identiying higher-alue isitor
preerences and understanding isitor low oer time rom lower- to higher-alue groups.
Some examples o each type o usage include:

Segmenting isitors into browsers` and buyers` based on whether they hae
made a past purchase. Doing so would allow you to identiy pages and content
preerred by the buyers` ,more aluable, group, enabling you to draw greater
attention to those pages and content. By highlighting aluable pages and
simpliying the naigation to these pages you eectiely improe the oerall
customer experience.
Segmenting isitors into interested` and committed` based on the total number
o pages they hae iewed during preious isits to the \eb site. Doing so would
again allow you to identiy content deemed more important by the committed`
group.
Measuring the time to conersion rom lower-alue groups such as unregistered
isitors` to higher-alue groups, such as registered users` and customers` in
order to orm more clear expectations o how long it takes to truly acquire a
customer.`
1racking reerring domains and,or campaigns that bring higher-alue isitors like
customers` to the \eb site. Doing so will help you identiy online business
partners more likely to contribute to your ongoing success.
Mining preiously collected data to identiy the reerring sources or isitors who
browsed a speciic combination o pages in order to again identiy best online
business partners.

1he list goes on and on. Visitor segmentation tools will play an increasingly important role
in improing online marketing in the uture.

Perhaps a more concrete example o isitor segmentation in action, especially segment
conersion, is how BackcountryStore.com went about determining whether isitors preer
registering prior to purchasing or purchasing anonymously. Similar to the example gien in
Chapter 6, BackcountryStore.com was concerned that its required registration prior to
checkout was driing some isitors to abandon their shopping carts. 1o test this,
BackcountryStore.com created a handul o isitor segments including unregistered,`
registered` and customer` ,ligure 15,.
Sq CIupLer ;




Iigure JS: Visitor segments being tracked by BackcountryStore.com.

\heneer a isitor used the anonymous checkout` option on BackcountryStore.com they
were put in the unregistered` segment. \hen they registered they were segmented as such.
1hen when either segment completed a purchase they were segment conerted` to the
customer` segment. 1hanks to segment conersion, BackcountryStore.com was able to see
a strong isitor preerence or anonymous checkout, additional data that supported
remoing required registration, as described in Chapter 6 ,ligure 16,.



Iigure J6: Visitor segment conversion observed by BackcountryStore.com to support
a streamlined checkout process.

Although you can see that registered isitors were making purchases aster, as eidenced by a
lower number o aerage isits beore conersion and a shorter aerage amount o time
spent beore conersion, more than 10 times as many anonymous isitors made purchases
,Unregistered ~ Customer`, compared to registered isitors ,Registered ~ Customer`,.
In the end the data pointed out that isitors preerred anonymous checkout and
BackcountryStore.com implemented this to great success ,see Lxamples` in Chapter 6,.
Risks Associated with Visitor Segmentation Tools
1he major risk associated with isitor segmentation is that segmentation tools currently
proided by the major \eb analytics endors are limited in their abilities. Some endors
only allow or moe orward` segmentation and are unable to create segments by mining
preiously collected data. Other endors only allow or data mining segmentation and are
limited in their ability to permanently` identiy a isitor as a member o a group. Still other
endors do not support isitor segmentation at all.

VIsILor SegmenLuLIon TooIs Sg

An associated risk is the number and type o measurements aailable in isitor segmentation
toolsets. \hile some endors proide limited numbers o statistics associated with isitor
segments, none known to this author proide truly covtete sets o statistics. 1he limitations
imposed are oten due to olume or type o data, or to presentation limitations. Regardless,
truly useul \eb analytics packages in the uture should support lexible deinition o
segments and complete statistics packages or those segments.

Web Analytics Tip: Use Segments to Collect Demographic
Data
Many o the customers that I work with hae some type o registration process on their \eb
site. Oten these registration processes ask moderately benign questions, such as, \hat
year were you born` and, \hich gender are you`-questions with answers that can be
used to segment isitors based on demographics. \heneer you can determine the year
someone was born, their gender, their marital status or household income you should
consider translating this inormation and tracking these isitors as large demographic
aggregates.

Imagine that you ask or year o birth and gender. lrom this inormation you can create
seen demographic segments -two or gender and ie or age brackets.

Gender
Male
lemale

Age Group
Under 18
18 to 25
26 to 31
31 to 40
Oer 40

1hen, depending on the answers a isitor proided on the registration orm, you would add
that isitor to either segment 1 or 2 and one o the segments 3 through . As an example, i
I indicated that I was a male born in 190 I would be added to segments 1 and 6.

Doing so will allow you to do seeral things. lirst and oremost you`ll be able to determine
through sampling how your \eb audience breaks down along demographic lines, by
counting page iews, isits and unique isitors. Second, depending on which metrics your
\eb analytics application proides, you`ll be able to see what pages and content each
demographic group inds most interesting. 1his inormation can be leeraged a number o
dierent ways-by your marketing group to better speak to your most important
demographic, by your adertising sales group to help sell adertising space to adertisers
seeking speciic demographics, by your product design group to help ensure that you`e
created the right personas or your usability and design testing.

1he nice thing about this type o segmentation is that it is done in aggregate and so proides
anonymity to your isitors. Nobody has to worry about indiiduals being identiied by age


S6 CIupLer ;



or gender because-hopeully-you can get large numbers o isitors segmented using this
strategy. A potential pitall and something that should be watched ery closely, howeer, is
whether you`re able to segment enough isitors using this strategy to proide a statistically
releant population. Consult your college statistics book or more inormation about what
constitutes statistically releant,` but be cautious o making business decisions based on
small samples.



CumpuIgn AnuIysIs TooIs S,

CHAPTIR S
CAMPAIGN ANALYSIS
TOOLS


Potential customers make a decision to visit, and in doing so, are broadcasting their
needs, which they look to the Web site to meet.
Bryan Lisenberg in Converting Search Lngine 1raffic


Building a \eb site without adertising and marketing the products or serices you proide
is like building a house that no one eer lies in. \ou certainly could do it, by why would
you Assuming you agree with that statement you will immediately see the need or tools to
help measure the eectieness o drawing people to your \eb site.

Likely you are using common marketing techniques to let people know that you are out
there and hae the products, serices or inormation that they`re looking or. \ou may be
purchasing pay-per-click adertising, such as that oered by Oerture or Google in their
keyword buying programs. Perhaps you are sending email to opted-in or unknown
recipients, perhaps you are buying banner ads on large portals sites. Perhaps you hae a
business partnership with a portal or a company like your own and you are trying to create
synergy through common interests. Regardless o how you are trying to make people aware
o your online business you need to track the quality o each message and medium.

Campaign analysis tools are designed to help you measure two things at minimum: response
and conersion. Response is the number o isitors, called respondents` in this context,
who actually click on the link you hae embedded in your adertising. Conersion is the
number o responders that complete a speciic actiity ater they arrie at your \eb site.
Being able to measure response and conersion is key to designing campaigns that
contribute positiely to your online business. \ithout these metrics you are only guessing
that your marketing eorts are working.



Iigure J7: Lxample of how one vendor helps their customers measure response,
conversion, conversion rates and campaign values for pay-per-click advertising
online.
SS CIupLer 8


One important thing to keep in mind: I you are not measuring some type o conersion
actiity or your campaign respondents, and you are not simply a portal designed to sell
adertising and little more, you are likely wasting your time adertising online. \ou should
vot be spending money or taking time establishing marketing channels that simply bring
people to your \eb site-you should be bringing them to your \eb site with a speciic goal
in mind ,sering \eb pages is not a goal, it is a means to an end,. Make sure that you hae
speciic goals in mind or the isitors you are trying to reach-purchase something, submit
inormation, download a document, and so on-beore you try and attract isitors to your
\eb site.

1his opinion contrasts that o those online marketers who insist that bringing isitors to
your site, regardless o their likelihood to complete a conersion actiity, is helping you build
and establish brand identity.` \hile this is likely true, that the more people who see your
adertising the more well known you become, ery ew businesses hae lourished simply
or being well known.` 1here is nothing wrong with building brand identity as long as you
are also working to create a more personal relationship with isitors at the same time,
learning more about the isitor or proiding them something that they desire.
Metrics Provided by Campaign Analysis Tools
1he undamental metrics that you need rom a campaign analysis tool are the number o
respondents and number o respondents conerted or any gien campaign. lrom these,
proiding you know roughly how many times the ad or message was iewed, you can
calculate the three most important metrics in online adertising:

Response rate ,respondents diided by total impressions sered,
1otal respondents
Conersion rate

Consider the discussion on conersion rates in Chapter 6. \ell-designed campaigns are
simply a process like any other, only one that is tied to a speciic reerring source. Metrics
or measuring campaign success are exactly the same, except you are likely measuring
abandonment or many pages as a group rather than discreet steps.

In addition to response and conersion there a number o useul unctions and metrics when
measuring campaigns. Additional unctionality includes:

1he ability to measure multiple, sequential goals oer a period o time, such as a
campaign designed to attract isitors and hae them register ,conersion actiity 1,
then later purchase ,conersion actiity 2,. 1his type o inormation is useul or
tracking multiple abandonment points i you hae complex conersion goals.
1he ability to organize campaigns into lexible groups. More oten than not
businesses are running hundreds o campaigns at any gien time. Gien this, it is
powerul to be able to group these campaigns using lexible criteria such as type
o campaign,` site,s, campaigns are running on,` actie dates,` message,`
placement,` creatie,` and so on. laing the ability to aggregate and roll-up the
CumpuIgn AnuIysIs TooIs Sq

response and conersion or lexible groupings allows online marketers to easily see
where their money is best spent.
1he ability to easily create campaigns or create them on the ly.` Some \eb
analytics packages require a campaign setup that, while not particularly diicult, is
cumbersome, especially when running hundreds or thousands o campaigns. Being
able to dynamically create campaigns and then modiy their properties ater the act
simpliies the process o collecting data.
1he ability to assign lexible costs and alues to campaigns in order to build return
on inestment ,ROI, models or campaigns. \hile being able to calculate response
and conersion rates are the minimum or campaign analysis, new tools are
emerging that allow online marketers to input cost inormation or campaigns
,total cost o campaign,` cost per click` or cost per acquisition,` and so on,, as
well as the real alue o a conersion ,lietime alue o a customer, alue o
products purchased, and so on,. 1his inormation enables \eb analytics
applications to calculate ROI, helping business people to better predict when a
campaign will pay or itsel, as well as the likely positie alue o a campaign.

Useul metrics or measuring campaign analysis ,in addition to the unctionality described
aboe, include:

Campaign respondents
Campaign conersions and conersion rate
Campaign entry pages ,assuming multiple entry points,
Campaign goal pages ,assuming multiple goal pages,
Paths to conersion
Reerring domains ,assuming the campaign is running on more than one \eb site,
Commerce statistics ,including products purchased` and more. 1his topic is
discussed in Chapter 9, Commerce Measurement 1ools.,

Again, keep in mind that the minimum unctionality or campaign analysis is the ability to
measure the number o people who click on a link and successully arrie at your \eb site,
as well as the number o those people who complete an actiity that has some alue to your
online business. I the \eb analytics package that you hae supports this, you are in
business.
An Important Note about Response Rates
A ery common problem that businesses hae trouble understanding is the drop-o`
between clicks and response in online campaigning. 1he basic scenario is this: \our pay-
per-click ,PPC, adertising engine is telling you that a particular campaign has recorded
1,000 unique clicks` or the month. 1hey report this to you because you pay them some
amount or each click generated. 1he problem arises when you then consult your campaign
analysis tools in which you hae coded and measured the response to your PPC adertising
and you notice that you only had 800 respondents or the same timerame. \hat happened
to the other 200 respondents

qo CIupLer 8

\hat happened, most likely, is that people dropped o` between the time it took to click
and the time it took to get to your \eb site, that is they clicked the link but neer arried.
\hy would people do that, you ask 1he single most common reason or this type o
abandonment is slow response.

I you are purchasing keywords on Google, which is notoriously ast to respond, but your
\eb site is bloated` with code and images, isitors are ery likely clicking your link and
then backing up when your \eb site does not immediately load in their browser. \hile it is
ery important or online businesses to look good and hae sexy` \eb sites, online
business owners need to recognize that your potential isitors want the inormation they are
looking or right away. See the discussion under \eb Serer Perormance Data, in Chapter
2, or more inormation.

I isitors are searching or inormation on a ery responsie search engine they are haing a
particular type o experience. Imagine now that they search or a keyword that you and your
competitors hae purchased, and your PPC ad is one o many in a list on the search results
page. I the isitor clicks on your link ater haing been moing quickly through search
results pages and has to wait more than a ew seconds or your \eb site to respond it is ery
likely will hit the back` button and click on one o your competitor`s links.

Now imagine the same scenario on a 28.8 modem connection.

In general when you are paying anyone to help you generate isiting traic it is important to
keep the ollowing best practices` in mind:

Be mindul o page weight in terms o total kilobytes and load time. I you hae to
hae an image rich` page consider using a deliery optimization serice such as
Akamai or Speedera to speed the deliery o images.
Design landing pages that are targeting the type o isitors you are trying to attract
with the particular campaign you are running. I you are attempting to pull traic
rom a large portal, such as America Online, be sure that the language and look and
eel o the landing page relects the demographic you are likely to attract.
Make sure the message on the landing page closely matches the message in your
adertising. Don`t run a banner ad promising ree serice or a month` and then
vot hae a big, obious link on the landing page describing how they can get this
ree serice. Visitors hae little patience or haing to search or inormation they
think should be readily aailable.
Similarly, be sure that the action item is completely obious to the respondent as
soon as they arrie. I your ad stated ree shipping with any purchase` then the
landing page should hae language describing the ree shipping oer and a big,
bold button that tells them to start shopping now!`
I possible, hae as ew links as necessary on the landing page to make the
respondent comortable, as well as to push them towards the conersion actiity.
laing too many links on your landing pages simply proides increased
opportunity or distraction. O course you should gie isitors a link to your
normal` \eb site but don`t push them towards it.
CumpuIgn AnuIysIs TooIs q1

Example Uses for Campaign Analysis Tools
Besides the obious use o quantiying how each and eery one o your online campaigns
contributes to your oerall business success, some companies hae also used campaign
analysis tools to measure internal usability. Imagine the scenario where you hae three
dierent links rom you home page that lead to a particular product you are trying to sell. A
commonly asked question is, \hich o these three links leads to the most purchases`

In most \eb analytics packages known to this author it would be triial to code each o the
links with a unique campaign ID ,usually by appending the campaign ID to the URL`s query
string,. 1hen A, eery time a isitor clicked one o the tagged` links, that link would be
counted as a response` to the speciic link`s campaign and B, when isitors completed the
purchase process the link would be credited with the conersion. 1hen, at the end o the
day, it would be simple to assess the contribution each link made to the oerall business
success.

Another important thing to keep in mind regarding campaign analysis is that there is no
reason to limit your measurement to only online campaigns. lor many successul businesses
o reasonable size, online adertising is only part o the total marketing program. I you also
run teleision spots, radio ads, print ads or create and distribute brochures you can make a
reasonable attempt at quantiying the eect each hae at driing isitors to your \eb site by
creating unique, branded landing pages.

Imagine, or example, that a multi-channel retailer like Circuit City had a unique landing page
at http:,,www.circuitcity.com,t. 1his page redirected anyone typing that URL into their
browser directly to the teleision section o their \eb site and told the isitor that not only
did Circuit City hae the greatest selection o 1Vs on the Internet but they also oered ree
shipping and also no interest inancing until January 2005. 1his is a nearly perect example
o a landing page and how to measure the eect o oline campaigning.

Now imagine that Circuit City was running ads in the Sunday circular in your local paper,
telling you that you could purchase rom the best selection o teleisions online, hae your
new 1V deliered to your ront door and get no-interest inancing until January 2005, simply
by going to your computer and typing in http:,,www.circuitcity.com,t.` I you were to do
that, and they were to associate any isitor that came to that page with a unique campaign
ID, Circuit City would easily be able to quantiy the eect the ad in the Sunday circular had
on their business. ,Please note that this is a made-up example, and does not relect any
current content or special that Circuit City may or may not hae on their site.,

One important caeat to this is that, in general, people are lazy. \hile 10,000 isitors may
hae been added to the oline teleision` campaign because they ollowed instructions and
included the ,t` portion o the URL, it is likely that many others simply went to
www.circuitcity.com and looked around or teleisions in the naigation system. 1his is ine,
nobody can change human nature, so it is simply important to remember that the response
you measure using this strategy is likely lower than the total number o responses and
conersions you can attribute to the campaign. 1his does not discount the alue o
measuring oline campaigns at all-the lazy isitor` actor is probably een across all
q CIupLer 8

similar campaigns and thereore can be actored out a. tovg a. you are careul when designing
these landing page URLs.

Designing landing page URLs careully or oline campaigns simply means that
www.circuitcity.com,t or t.circuitcity.com is better than www.
circuitcity.com,teleision_ad_2004 or www.circuitcity.com,departments.jspc~1&b~g&
department~1eleisions. \ou want to design unique landing page URLs that are short,
sweet and easy to remember. Consider adding ,t` or ,radio` or ,august` to the end o
your domain URL and using redirect scripts on those pages to add your campaign ID and
pass the isitor along. \ou should also consider mentioning to your audience that they need
to use this special URL` to take ull adantage o your oer to try and create a little more
urgency to remember the anity URL.

Another example o the use o campaign analysis tools is the work that this author does in
buying keywords on pay-per-click engines like Google and Oerture. Because the author is
working rom a budget to spread the word about demystiying \eb analytics, he needs to be
careul to spend money only on words that successully attract book buyers. One o the
tools used to keep track o this are the campaign analysis tools proided by his \eb analytics
proider.



Iigure J8: Lxample of how the author's pay-per-click terms show up in Google
AdWords search results.

In ligure 18 we can see what would appear in the right-hand column on Google i you were
to search or web analytics demystiied` or some deriatie thereo. 1he click-through
URLs in the Google engine are tagged with a name-alue pair that identiies that click as
haing come rom the keyword buy at Google. Because o this identiier and the \eb
analytics package that the author uses, as soon as isitors start responding to this keyword
campaign he can begin to see the response in real-time.



Iigure J9: Responses, conversions, conversion rate and campaign value for a sample
of keywords purchased at Google by this book's author.

CumpuIgn AnuIysIs TooIs q

As you can see in ligure 19, the author is able to easily determine both the rate at which he
can expect people clicking on the paid link to make purchases and also the actual alue o
the campaign, to date, or each indiidual keyword purchased. 1his inormation can be
easily compared to the cost inormation proided by Google to ensure that each keyword
purchased is paying or itsel and helping keep !eb .vat,tic. Dev,.tifiea a proitable enture.

In addition to this granular inormation it is important to the author to keep track o how
the entire Google pay-per-click` campaign is perorming relatie to preious timerames.
Remember, \eb analytics is not about numbers but how those numbers are changing ,see
Chapter 15 or more inormation on key perormance indicators,.



Iigure 20: Campaign performance comparing two different timeframes.

As you can see in ligure 20, the news is mostly good or the author regarding the Google
PPC campaign, as all o the raw response and conersion metrics are increasing while the
campaign conersion rate stays the same.
Risks Associated with Campaign Analysis Tools
1he major risks associated with campaign analysis tools are generally improper setup and
campaign IDs oerriding each other. In the ormer case, user error is surprisingly common
when setting up campaign IDs or tracking, as most \eb analytics packages currently
aailable proide some interace or tool to create campaign IDs and a airly regimented way
to track said campaigns. It is likely that i one ails to properly deine a campaign or tell the
analytics package what to look or the data may be lost ,although \eb-serer log-ile analysis
applications minimize this risk,. 1he only real solution or this problem is to be careul
when setting up campaigns and campaign tracking.

Regarding the latter risk, campaigns can oerride each other i you are tracking many
dierent campaigns at one time. I a isitor gets an email rom you and clicks the link you
will count that isitor as a respondent to the email campaign. Imagine then that this isitor
leaes your \eb site without completing the conersion actiity but a ew days later sees a
banner ad, is reminded that they intended to complete that actiity and clicks the banner ad
to return to your \eb site. At this point you hae counted the same isitor as a respondent
twice, assuming you are tracking both the email and banner ad campaigns, which you should
qq CIupLer 8

be. I the isitor then completes the conersion actiity, it is likely that the conersion will
be attributed to the banner ad campaign, not the email campaign. \hile making the
assumption that the last campaign responded to should be gien credit or the conersion is
not a huge problem, it does increase the risk o drawing aulty conclusions.

A reasonable solution to this problem would be to credit att the campaigns the isitor
responded to within a reasonable amount o time rom the conersion. I your \eb
analytics package supports this option you should consider using it i you are interested in
knowing the contribution that each campaign makes to the likelihood a isitor will conert.
But i you are more interested in knowing only which message, method or medium is most
likely to actually get the isitor to complete the action, only crediting the conersion to the
ta.t campaign the isitor responded makes sense.

Web Analytics Tip: Track Every Campaign You Run, Carefully
As the technology sector picks up again, as it appears to be doing in the irst hal o 2004, it
is worthwhile to remind online businesses not to inest in online marketing with the same
blind exuberance that was exhibited in the late 1990s. \hile there is surely temptation to
inest heaily in customer acquisition, the smartest course is to inest careully in marketing
campaigns that demonstrate positie returns in both the short and long term.

1o ensure that eery campaign is tracked through to conersion, the best course o action is
to encourage your marketing and marketing analytics groups to work together and make sure
that they understand how to use the campaign analysis tools proided by your \eb analytics
application. Making sure that eery campaign is coded correctly and goal pages are properly
set will help ensure that the eect o each campaign can be accurately quantiied.

Additionally, many clients I hae worked with expressed the desire to track both internal and
external campaigns-the ormer being elements designed to drie traic towards a particular
goal, such as internal banner ads, merchandising oers or speciic applications on the site
itsel. \hile the aorementioned recommendation is to track eerything, one must be
cautious and ensure that your analytics package supports both internal and external tracking.
I your application only has a single campaign set and is unable to dierentiate internally and
externally drien responses, tracking internal campaigns can cause data loss or external
campaigns at the point o conersion.

Be sure when planning your campaigns that you consult with your analytics proider and ask
them or their recommendations regarding campaign setup and tracking internal campaigns.
1hey will likely hae speciic recommendations and be able to guide you towards the most
eicient and accurate implementation.



Commerce MeusuremenL TooIs qg

CHAPTIR q
COMMIRCI
MIASLRIMINT TOOLS


In 2002, US online retail grew to $76 billion. 1his represents a 48 percent increase
over 200J. In 2003, online sales will grow by 26 percent over 2002, to $96 billion.
Online sales will account for 4.S percent of total retail sales in 2003, up 3.6 percent
from 2002.
1he State of Retailing Online 6.0 from Shop.org/Iorrester Research


Although most online commerce platorms proide reporting o some kind, rom a \eb
analytics perspectie there are a number reasons to employ additional commerce
measurement tools-primarily or the ability to tie commerce actiity directly back to the
Internet, that is, the reerring sources that sent isitors to the \eb site in the irst place.
\hile any good commerce platorm, regardless o its price and unctionality, will proide
inormation about which products, categories and brands you are selling, ew are able to
directly tie sales back to campaign actiity and een ewer can connect to non-campaign
reerring sources. A good set o commerce tools proided by your \eb analytics endor
should allow you to do this with ease.

A secondary reason to employ commerce measurement tools within your \eb analytics
application is to acilitate ease-o-use and ease-o-understanding. I all o the data about
inbound campaigns, isiting traic and online purchases are aailable in one place the data is
likely more actionable. Imagine how much easier it is to hae all o this aluable data in one
place and not hae to moe between applications, setting date ranges and parameters in
each, only then to hae to moe that data to a third-party application or summary reporting.
\ou can deinitely do it, but why spend the time i you can aoid it

Robust commerce measurement tools within \eb analytics applications should proide
unctionality to allow you to easily measure key processes such as shopping cart and
checkout, isualize product abandonment, generate browse-to-buy ratios and ilter
customers based on common segments ,new s. returning, irst isit s. many preious isits,
and so on,. 1he greater the additional v.efvt unctionality a \eb analytics commerce toolset
can proide, the greater the likelihood that the inormation proided can help you increase
online sales.

lunctionally, commerce analysis tools should proide the who, what, when, where and
how` or your online purchasers:

q6 CIupLer q

\ho` is making the purchase, granular to the person`s name i your application
and priacy policy support this
\hat` did they purchase
\hen` did they make the purchase
\here` did they come rom
low` did they purchase

1he answers to these questions allow the online business to make signiicantly better
decisions regarding marketing and merchandising.

linally, imagine the case where you hae built your own commerce application rom the
ground up, a situation common to many smaller online businesses. In this case it is ery
likely that you only took the time to include the bare minimum o data reporting capabilities,
just what you needed to do order ulillment and inentory management. In this instance,
implementing a \eb analytics package that proides commerce tools can hae signiicant
beneits, oten at a ery reasonable cost. By simply identiying or tagging your order
conirmation page to collect transaction data it is ery likely that you will greatly increase
your knowledge o the customer with little or no additional programming on your part.
Metrics Provided by Commerce Measurement Tools
As described aboe, your online commerce application likely proides a great deal o
inormation about total products sold and the customers who purchased them, inormation
that is important to your inentory and ulillment groups. 1he metrics that are most useul
to \eb analytics users are those dealing with merchandising, sources o customer traic and
the online purchase process.
Merchandising
Metrics that are most aluable when measuring the eects o your online merchandising
include:

Products, categories and brands sold, especially i this data is aailable by placement
or position so you can measure lit ,that is, eatured product` s. category listing`
s. search results`,
Product conersion rates ,browse-to-buy ratio, shopping cart conersion,
Internal searches leading to purchase, helping you understand better how your
customers think about the products you sell and obsere the eects o any cross-
selling actiities you engage in ,people searching or books are purchasing both
books and shoes`,
Links leading to purchases and amount o reenue attributable to each. \hile it is
not always clear how to attribute alue to a link on a page, endor attempts to do
so hae been met with great enthusiasm rom merchandisers and thus any
reasonable, well-explained algorithm or making this assignment is worth looking
at.
Commerce MeusuremenL TooIs q,


Being able to determine the eect o dierent merchandising strategies allows you to
optimize your product merchandising eorts. \hile there are relatiely ew unique ways to
present products to online shoppers, aailable data suggests that dierent products and
product categories hae dramatically dierent purchase rates depending on how they are
presented. \ith this in mind, successul online businesses work to understand the best
method or presenting products to optimize the look-to-book ratio.
Referring Sources
Sources o customer traic can be paid or unpaid. Metrics most aluable in helping to
quantiy the eects o reerring traic include:

Reerring domain and URL
Search engine and search engine keyword or phrase
Campaign measurement tools, as described in Chapter 8. It is especially important
to be able to tie paid campaigns to real product purchases as well as the number o
customers attracted, the number o orders placed and total reenue rom the
campaign. Ideally you are able to directly tie the campaign metrics preiously
described to all customer order actiity generated ia the campaign.
Lietime alue o a customer rom each type o reerring source. It is ery useul to
know which o the dierent types o reerring sources are sending you the most
aluable traic oer time. 1he idea behind lietime alue o a customer by
source` is to keep track o all recurring purchases rom a customer and credit them
back to their original source. 1he alue o this is deliered oer time because,
gien this data, you will be able to not only determine which sources proide the
greatest short-term gains but also which sources are sending you the most qualiied
traic oer time.

Search engine keywords are especially useul because this inormation proides insight into
what customers were thinking about just beore they ound your \eb site. Much like
measuring the merchandising eects o internal searches, understanding how your best
isitors` ind your \eb site will help you optimize your site or these kinds o isitors.
Online Purchase Process
1he online purchase process is simply a specialized process and is measured much like the
processes described in Chapter 6. 1he key conersion rates and statistics useul to measure
the online purchase process include:

Shopping cart and checkout conersion rates. 1he shopping cart conersion rate is
the ratio o completed shopping carts to cart starts` ,isitors adding at least one
item to the cart,. 1he checkout conersion rate is the ratio o completed shopping
carts to the number o times a isitor clicked the checkout` button.
Product, category and brand cart and checkout conersion rates. lor any product
on you oer you should be able to see how many times this product was added to a
qS CIupLer q

shopping cart, checked out and ultimately purchased. It is desirable to be able to
see this inormation or both category and brand, as well. A key metric is the ratio
o times an item,category,brand is carted to the number o times that
item,category,brand is actually purchased.
Oerall conersion metrics speciic to the commerce process, including key
aerages ,reenue per customer, reenue per isitor, items purchased per customer,
items per order,
Browse-to-buy ratios or products, categories and brands. 1he ability to see how
oten an item or group is looked at and quickly compare that to the number o
times those items or groups are purchased proides aluable insight into which
products and groups are currently in aor` and should be considered or
additional merchandising eorts.

Making these kinds o measurements or your purchase process will help you identiy the
areas where isitors struggle. Additionally, by measuring product, category and brand
conersion rates you will be able to better merchandise on key pages by eaturing items that
are more likely to sell. Making simple but smart decisions like these are important to
improing oerall sales online, keeping in mind the our Ps`-product, placement, price
and promotion. 1hese rates and ideas are discussed at length in Chapter 13.
Sessionizing/Identifying Information
In addition to merchandising, reerring sources and online purchase processes, many
companies may be able to derie alue rom tracking customers based on sessions. Some
endors proide the ability to segment based on indiidual isitor actiities, including
purchase history. 1his inormation potentially proides answers such as, Customers who
hae made less than >100 in lietime purchases are interested in these pages,` Customers
who hae purchased product A` and product B` come back to the \eb site with x`
requency` and, Gie me a list o all o the names and email addresses or isitors who hae
put product C` and product D` into their shopping carts but did not complete the
purchase.`

\hile the answers to the irst two examples are interesting it can be less clear how to make
these types o inormation actionable, or the larger analytics audience, attempting to obtain
this type o inormation may simply result in increased costs without signiicant gains.

1he third example, howeer, is potentially ery useul to any \eb analytics program, being
able to tie speciic online actiities back to known indiiduals makes or much improed
online marketing. \hile there is potential to anger isitors who hae not speciically
requested inormation and you may run aoul o priacy guidelines, being able to market
directly to measured isitor interests is ery likely a large part o the uture o \eb analytics.

Ideally, or all o the merchandising and reerring source metrics listed aboe, your \eb
analytics application will proide access to identiying inormation or all customers. Len
more useul is the ability to obtain this inormation or non-purchasers, isitors who put
items in the shopping cart but do not purchase or people who look at a particular group o
content but neer put items in the cart. \hether this inormation is proided within the
Commerce MeusuremenL TooIs qq

application ia an interace or ia a customer ID export or integration into your data
warehouse is just a matter o preerence, as both require some leel o setup.

I you think your online business would be able to take adantage o this type o inormation
it is worth asking your application proider how they are able to delier this inormation.
1he easiest case should be that you proide a unique ID or each isitor you wish to track
and the application allows you to use this to build releant iews o the data. Some
applications take this a step urther and gie you the ability to import or integrate your own
customer database to query inormation inline.` Again, where the inormation is made
aailable is a matter o preerence, as long as the inormation is aailable somewhere.
Example Uses of Commerce Analysis Tools
1he most practical application o commerce analysis tools within a \eb analytics application
is the ability to tie purchasing actiity back to reerring sources. Some o the most
signiicant wins in terms o return on inestment into \eb analytics occur when online
businesses use these types o tools to quantiy the eects o pay-per-click campaigning or
business relationships.

1he popularity o search engine keyword buying and the pay-per-click purchasing strategy
has prompted many online businesses to begin bidding on words with wild abandon. \hile
companies report great success tied to this relatiely new channel or online marketing,
others are discoering that, with alarming requency, clicks are not translating into
conersions, or i they are, not high-alue conersions. 1he campaign analysis tools
described in Chapter 8 are excellent or helping you to ensure that you are achieing a
minimum conersion rate or any campaign that you are paying or. Commerce analysis
tools are good at getting deeper into the data and helping you to ensure that the campaign is
selling the right` products and that the campaign itsel has a good aerage order alue.

Consider an example company bidding or a popular, generic keyword and haing to pay
>1.00 per click. Now imagine that they get 1,000 clicks rom that word eery day or a total
cost o >1,000 per day or >,000 per week. Most companies preer to inest in marketing
that will return 10 times, meaning that this company would need to make >0,000 per week
rom this campaign. I they measure their conersion rate rom this word to be 3 percent,
again not unheard o, then eery day they would be conerting roughly 30 respondents to
customers. 1his keyword campaign would need to be selling products with a minimum
aerage order alue ,AOV, o >333.00 to be hitting their return target. So:

>10,000 daily target , 30 customers conerted ~ >333.33

So now imagine that the company is selling trading cards and their AOV is only >4.00 per
order drien rom the campaign.

>4.00 AOV rom campaign 30 customers conerted ~ >120.00

Lery day this campaign costs the company >880.00 ,or has an ROI o 0.12-or eery
dollar spent the campaign returns >0.12,.
1oo CIupLer q


\hen shown the math, most marketing managers would agree that this campaign is a dud
and immediately pull the keyword rom spending rotation, howeer many marketing
managers neer see this type o inormation. Len in this modern age companies are
spending money without any tools to quantiy the results, eectiely wasting money. More
inormation about quantiying the alue and return or online campaigns is presented in
Chapters 12 and 13.

Regarding the use o commerce analysis tools to quantiy the quality o business
relationships, consider the explosion o \eb sites and the you link to me I`ll link to you`
schemes that occurred in the late 1990s. Any number o \eb sites may hae approached
you with promises o large olumes o high-quality traic that they could send to ,ovr \eb
site, or a ee. In the absence o any \eb analytics tools this type o relationship is a
mistake-how do you know you are getting any o this traic and that respondents are
engaged in actiities that proide your business alue 1he deployment o commerce
analysis tools allows you to take relationships like this one step urther and quantiy the alue
you get rom your traic partner.`

An example o quantiying the alue o a relationship between business partners can be seen
at BackcountryStore.com which has relationships with a number o dierent \eb sites to
drie traic their way. Using the \eb analytics package it has standardized on,
BackcountryStore.com is able to keep track o who is sending traic that conerts into
customers and drill down into the products those isitors are purchasing.



Iigure 2J: Commerce data showing the distribution of orders, revenue, and so on,
from referring traffic partners. Of note is the browse-to-buy ratio (B/B), which is
good for comparing the quality of traffic sent by each partner.

As you can see in ligure 21, BackcountryStore.com is able to learn a great deal about traic
sent to it rom each online business partner, regardless o the type o partner.
BackcountryStore.com is also able to drill down or any business partner and see exactly
what products that partner is helping to sell ,ligure 22,.


Commerce MeusuremenL TooIs 1o1


Iigure 22: Drill-down into the products that BizRate.com helped
BackcountryStore.com sell in this example.

BackcountryStore.com can now take this inormation and mine it to learn which products
each online partner is most likely to help sell. laing this inormation will allow them to
reine the relationship they hae with each company and work towards optimizing these
relationships. Many smart marketing and commerce managers hae used this kind o
inormation to sae their companies money, millions o dollars in some instances, simply by
applying the right metrics to the problem. \hile you will not always endear yoursel to the
traic partner, especially i you need to go back to them and end the relationship, this should
be o no concern. 1he goal o eery online business working to be successul should be to
make and sae money.
Risks Associated with Commerce Analysis Tools
1he major risks associated with commerce analysis tools are expectations and accuracy.
Oten when managers hear that new tools are being implemented to track something they
expect that these tools will track eerything.` \hile it would be nice to be able to purchase
one o the aailable \eb analytics packages and hae it collect and report on eery
measurement your business needs, this is, unortunately, unrealistic. O the aailable
packages known to the author, some are ery good at collecting inormation about indiidual
customers but less good at summary reporting, others are ery good at presenting
inormation but only present a limited subset o what would be truly useul. 1he challenge
to the commerce manager is to select the application that proides the best it or their needs
vor and answers the questions that people hae already been asking.

It is extremely common or companies, when selecting a \eb analytics endor, to create an
extensie request or proposal ,RlP, that lists eery metric and eature the company needs,
will need, may need or has eer heard o. \hen you are trying to select an analytics package
the best adice you can get is to ight these temptations and not create a list so long and
complex that no one endor can hope to satisy your requirements. RlPs like this do not
help endors and in most cases simply draw out the decision-making process, preenting
anyone rom getting any analytic work accomplished. By properly setting expectations about
what commerce analysis tools will proide you will sae yoursel, and your co-workers, time
and headaches.

Accuracy is a sticky subject or \eb analytics practitioners, one that becomes een stickier
when you are talking about commerce analysis. Consider the inormation proided about
accuracy in Chapter 3. Lach \eb analytics data source has the potential to be more or less
1o CIupLer q

accurate than the other in regards to the number o page iews, isits and isitors to your
\eb site. 1he same complication extends to commerce analysis but is urther aggraated by
the simple act that eote eect av, covverce or rerevve reortivg to be 100 ercevt accvrate. \hen
customers are asked, low accurate is accurate enough or \eb traic data` most indicate
that somewhere between 90 percent and 95 percent is pretty good. Ask the same question
about commerce data and the answer is almost always 100 percent.

\hile 100 percent accuracy in commerce data is highly desired it is, unortunately, an
unreasonable expectation. Because the Internet is not perect, and was not designed to be
so, some inormation will occasionally be lost. \hile it may be a small percentage o traic
and commerce data, een a single transaction lost will preent in-house numbers rom
matching those rom a commerce analysis application. Rather than iewing this as a
signiicant problem and picking up the phone to berate your application endor, consider
the ollowing points about commerce analysis applications:

Commerce analysis tools that are embedded in \eb analytics applications are
designed to help you understand where your best customers come rom, what they
respond to and where they hae diiculty in your online purchasing process. 1hey
are not designed to be a replacement or the existing analytics and reporting tools
that are part o your commerce engine.
Len 95 percent accuracy is enough to make good use o commerce analysis tools.
I you collect all but 5 percent o your transactions you will still be able to see the
trends that you are looking or. Again, these trends are the who, what, when,
where and how` or your online purchasers.
\ou should hold any endor you choose responsible or data collection within 95
to 98 percent accuracy but not call them eery time these numbers slip a bit. 1he
best practices` recommendation is to compare the numbers your commerce
analysis tools report to your actual numbers once a week. I you see a downward
,that is, less accurate, moement or more than two weeks, or o more than 2 or 3
percentage points in a single week, then consider picking up the phone. Look at
both number o orders and reenue, as sometimes a great deal o reenue
dierence is due to a single lost order and this situation does not warrant an angry
call to your \eb analytics application endor.

I you keep these simple guidelines in mind wheneer you look at your commerce analysis
data you are almost assured to hae a better experience and be able to use the data as
designed` to great beneit or your organization.

Web Analytics Tip: Tie Internal Searches to Purchases
One o the great adantages o haing a suite o commerce tools tied to your \eb analytics
package is the ability to measure how speciic eents or actiities drie isitors to purchase.
One o the most common eents I recommend that my clients track is internal searches.
1he careul use o a session cookie can allow you to track the search words and phrases that


Commerce MeusuremenL TooIs 1o



isitors are entering into your internal search engine that result in a purchase in that same
isit. 1he basic setup looks like this:

1. \our customer uses the internal search engine.
2. \ou tell your \eb analytics application about the search term, ideally using a
custom ariable.
3. At the same time you write this search term to a session or isit cookie, one set to
expire at the same time the isit ends.
4. I the isitor successully completes a purchase in the same isit, you include the
search term as part o the inormation you pass to your commerce toolset.

1his strategy, o course, depends on your commerce toolset haing the ability to accept
custom order-leel ariables that can be tied to the products actually purchased. I you hae
that ability, ollowing this strategy can create isibility into which search terms sell which
products and how much each internal search is worth. Some ariations on this theme
include tracking att o the search terms entered during a isit, perhaps as a comma-separated
list, or keeping track o search terms across multiple isits and crediting the search term
when the purchase is inally made.

Clients I hae worked with hae used this strategy to calculate browse-to-buy ratios or
indiidual internal search terms, inormation that can be leeraged in a number o dierent
ways. I commonly recommend that \eb business owners keep track o internal search
terms to understand the isitor mindset-isitor intent is ery diicult to determine but
search-term analysis can proide aluable clues to what is on someone`s mind when they
isit your \eb site. 1racking search terms to purchase takes this a step urther and helps the
business owner igure out which searches are the most aluable inancially. \ith this
inormation you can begin to moe rom understanding the intent o the aerage isitor to
the intent o the aerage cv.tover.





1oq CIupLer 1o

CHAPTIR 1o
AN INTROLCTION TO THI
CLSTOMIR LIII CYCLI


People come to your site, leave footprints and move on. But those footprints are
merely an indication that they were there and tell you nothing about the people who
made those marks. Unless you know who are casual callers and who are loyal
devotees, you cannot tell if your promotion and conversion efforts are working to
your benefit or not.
Sterne and Cutler in L-Metrics for the New Lconomy


At this point you may be thinking, Great, I understand what \eb analytics can do or me
and the tools that will acilitate my goals but I still can`t make heads or tails o the actual data
we`e collected.` Or you may be thinking, I paid good money or this book and the author
has not taught me jack-squat that I can take to my boss.` \hile the inormation presented
in the preious chapters are an important background on the subject, eerything rom this
point orward is where the rubber hits the road-practical \eb analytics.

1he ramework or our discussion about which metrics you should be measuring and what
they mean to dierent business types is the customer lie cycle-reach, acquisition,
conersion and retention. Any online business manager who has taken the time to read this
book has a business that needs to identiy high-quality sources o potential isitors, attract
them to their \eb site using a ariety o marketing programs, entice them to perorm some
actiity or actiities that are o alue to the business and then keep them coming back to
perorm these actiities.

Some readers may now be thinking, But that`s not what I`m trying to do. I don`t hae any
conersion actiities on my \eb site,` or, \e really only depend on getting someone to the
\eb site one time.` In the irst case, all \eb sites hae conersion actiities o some kind,
een yours. In the latter case, you should still understand reach, acquisition and conersion
een i you only want to conert isitors one time. Lery business spending money on
marketing online can take adantage o deeloping a \eb analytics program around the
customer lie-cycle ramework. 1he reasons or this are two-old:

1he ramework proides a business-centric model or understanding your \eb site and
relating your successes to others. As more and more online businesses become critical
components o larger multi-channel organizations it becomes more important to hae a
common language throughout. Data suggests that more companies are working to integrate
promotion and marketing in both on-and oline channels. By establishing the same broad
An nLroducLIon Lo LIe CusLomer IIe CycIe 1og

categories or success ,reach, acquisition, and so on, managers rom dierent groups tasked
with the same goals will be able to report back regarding success in the same terms.
\our \eb analytics program will be entirely more successul i you apply a ramework to it,
and the customer lie cycle is better than most. Many \eb managers in the past hae
attempted to simply collect data and then relate it back to the business. 1his model resulted
in many reports that stated, \e hae 20 percent more hits this month than last` to which
most people responded, Great! \hat does that tell us` 1he customer lie cycle as a
ramework allows \eb managers to generate reports that let the company know that Our
online marketing actiities hae helped us to acquire 20 percent more isitors, month-oer-
month, and these isitors are completing our key actiities at a rate o 4 percent.` 1he
dierence is subtle but powerul.

So assuming the reader is at least intrigued about applying this ramework to their
measurement, the ollowing deinitions o reach, acquisition, conersion and retention will
clariy the types o metrics we`ll be looking at starting in Chapter 11.
Reach
Reach, in adertising terms, is the likelihood or potential that you will be able to gain a
prospectie isitor`s attention. Depending on the ehicle you use or adertising, this can be
quantiied in a number o dierent ways:

1he number o people who see banner impressions sered on a \eb site
1he number o people who search or a keyword that you bid on
1he number o people who see an article written about your company or products
1he number o subscribers to a newsletter you sponsor or adertise in
1he number o readers who subscribe to a newspaper or magazine you adertise in
1he number o driers who see a billboard adertising your business or products
1he number o iewers who watch a commercial you run on teleision
1he number o recipients who receie a piece o direct mail you send

As you can see, each metric is both easy ava impossible to accurately quantiy. 1hey are easy
to quantiy in that any banner network can tell you how many banner impressions were
sered or search terms searched or, any magazine or newsletter can tell you how many
readers they hae and any marketing manager can tell you how many alid addresses your
direct mailing was shipped to. But they are impossible to accurately quantiy in that you
hae no way to know exactly how many people actvatt, reaa your message.

lerein lies the diiculty in measuring reach within your \eb analytics program. lew
applications hae the ability to automatically measure impressions sered across multiple
marketing channels. \hile \eb-serer log-ile analyzers are oten able to measure the
number o impressions sered or banners, images and media iles, this same ability does not
extend to oline channels. Page tagging applications, at least those known to the author, do
not hae the same ability but oten are able to measure impressions within some types o
email and rich banner adertising. Regardless o what your application cav measure, it most
1o6 CIupLer 1o

assuredly cannot measure the number o people who actually read and think about your
marketing message.

Because o this, reach is oten tied closely to acquisition, while we cannot measure the
number o people who read a message we can iner this alue rom the percentage o people
you are able to acquire. Practically speaking, most o the measurements o reach discussed
in this book are those tied to the actual acquisition o isitors.
Acquisition
I reach is a measure o the likelihood to gain a prospectie isitor`s attention, acquisition is
the measurement how successul you are in driing them to action, the number o people
who click on a link or type a URL into their browser and arrie successully at your site.
\hile some may argue that acquisition is better measured in terms o a isitor arriing ava at
least starting some actiity o alue, it is the author`s opinion that A, any isitor who arries
at your site ater responding to some type o adertising has the potential to complete an
actiity o alue and B, sometimes simple metrics are more powerul and in most instances it
is much simpler to measure the act o arrial or response.

Statistics used to measure acquisition are primarily ocused on the source o traic-what
reerring domains, search engines, search keywords, and so on-are bringing isitors to you.
Campaign analysis tools extend this measurement and allow or the qualiication o
isitors-and in the process bleeds oer into the measurement o conersion.
Conversion
Conersion and actiities leading to conersion are the reason that you hae your \eb site,
the successul completion o speciic actiities by isitors that somehow contributes
positiely to your online business. Conersion can be measured a number o dierent ways
or a ariety o dierent actiities, many o which were discussed in Chapter 6. As a
reminder, a successul conersion does not hae to be an online purchase, it can be:

Starting a download o a document or application
Submitting inormation-lead generation
Locating inormation, such as a support document or lAQ
Naigating rom more general to more speciic inormation
Spending a deined amount o time or iewing a speciic number o pages
Viewing key pages, such as a pricing page or serice agreement

1he list goes on and on. Depending on the multiple goal page lexibility proided by your
\eb analytics application you can deelop a ery complex model or understanding how
changes to your \eb site and marketing program hae an impact on your isitor`s ability to
complete goals.

An nLroducLIon Lo LIe CusLomer IIe CycIe 1o,

Another important piece o inormation to remember rom Chapter 6 is that there are many
dierent conersion rates` or any \eb site. \hile it is ery important to measure
conersion rates it is equally important to not get too hung up on one rate or the other,
especially when it comes to comparison. As preiously discussed, research irms hae put
out studies talking about conersion rates online` and conersion rates by business type or
category.` \hile this inormation is interesting, especially when you are able to know the
conersion rate or a particular \eb site that you can then go and isit, one should be
careul to not put too much alue on reported numbers. \ith ew exceptions these
numbers are deried rom dierent types o data and so are, in this author`s opinion, not
alid or comparison. One recently published report rom the lorrester Group`s Shop.ORG
project cites a slight increase in oerall conersion rates or multi-channel retailers rom 2001
to 2002 ,rom 3.1 percent to 3.2 percent,. 1his increase is much larger or \eb-based
companies, with an aerage increase rom 3.2 percent to 5.8 percent. \hile this report does
deine their measurement o conersion to be site isitors diided by total orders` it does
not tell us how site isitors are calculated ,which data source was used or the analysis, and
so creates potential apples and oranges` comparisons o these rates to your own data.

\hen it comes to conersion rates and comparing yoursel to others, you should do three
things:

Deelop your \eb analytics program in such a way that you are able to measure
three or our key` conersion rates and watch these religiously. 1hese will dier
depending on your business model but eery business model has three or our key
rates that should be watched and added to your own regular key perormance
indication program.
Do not compare your rate to published rates, een i they are or your direct
competitors. Because, by irtue o haing read this book, you will hae a ery well-
deeloped understanding o conersion rates and how to measure them you will
likely get little alue out o comparing yoursel to other \eb sites. Additionally,
the audience you reach and acquire is not necessarily that o any other site-you
should make changes to your site that will improe the customer experience or
,ovr customers.
Use conersion rates to drie action in your \eb analytics program. Conersion
rates are an easy-to-understand metric telling you i changes you hae made are
successul or not. I your rate goes up, they were successul, i your rate goes
down, they were not and you should back out the changes. Obiously \eb
analytics is more complex than this, but i you learn to lie and die by your
conersion rate you are doing most o the hard work.

Additionally, guard your conersion rate as i it were one o your trade secrets. Don`t brag
about how great your rate is, don`t tell the press or analysts unless you are extremely sure
they will protect your identity when reporting rates, and by all means don`t tell your
competitors. All anyone needs to know a tremendous amount about the success o your
business is an estimate o your monthly traic-which is aailable or purchase through
serices like ComScore-and your conersion rate. Roll in your aerage order alue and any
competitor can know how much reenue you are bringing in eery month.
1oS CIupLer 1o

Retention
Len ater you hae spent the time and money necessary to reach, acquire and conert
isitors you still need to work to keep them coming back. It is ery easy to ind studies
talking about how much more aluable repeat customers are than irst time customers, how
they spend more money, how they tell their riends about you, and so on. 1he hard thing is
to actually get customers to repeat, perhaps because there is ery little loyalty on the Internet.
I you are selling something this week, next week a competitor may hae it cheaper. I you
are proiding a serice, someone else may be able to delier it more quickly. Unless you are
selling a brand or proiding a serice that nobody else can delier you will always hae to
deal with churn. Len i you hae the luxury o selling your own items you may still hae to
work to compete with your channel-or rather, work to not compete with your channel.

Retention is the measurement o the actiities o your repeat isitors, whether they are back
or support inormation, back to purchase again or just back to do additional research.
Being able to segregate these isitors such that you can better understand their behaiors and
habits will allow you to better respond to their needs and market additional products and
serices to them. Visitor segmentation tools are critical to understanding dierent classes o
repeat isitors.
A Note about the Average Visitor
1hroughout the ollowing chapters o !eb .vat,tic. Dev,.tifiea the concept o the aerage`
isitor or the aerage` isit will be used and this abstract concept merits some discussion.
1he act that most \eb analytics applications, at least in their deault implementations, treat
\eb isitors as anonymous entities oten creates issues or marketing olks who are
constantly striing to understand the isitor better.` Unortunately it is diicult to
understand something that is essentially unknowable because o its anonymous nature or the
sheer size o the population.

1o compensate it is recommended that you become comortable with the concept o the
aerage isitor` and the types o things the aerage isitor` may do on your \eb site. In
uture chapters you will read about metrics that help you to understand the aerage isitor`s
requency o isit, the aerage number o pages that isitor sees in an aerage isit and other
useul but not particularly detailed data. \hile it would be more interesting to be able to say
Lric 1. Peterson, the author o !eb .vat,tic. Dev,.tifiea, came to our \eb site 5 times in
August and iewed 3 pages on his irst isit, 12 on his second, and so on .` this is A, not
practical or most online businesses and B, not aailable in most \eb analytics applications.

1rust in the act that the aerage isitor` is not a bad person and that they too want to
contribute towards your success online, proided you are using the continuous improement
process to make sure your \eb site is easy to use. But be cautious o the aerage
isitor`-make sure that he or she` is an aerage o enough real isitors to proide a
statistically alid distribution. One o the most common mistakes people make when using
aerages is to not ensure that their population is large enough. 1hink back to your college
statistics courses and make sure that you hae a large enough population to generate a
normal distribution- the bell cure. A good rule o thumb is to not start working rom
An nLroducLIon Lo LIe CusLomer IIe CycIe 1oq

aerages until you hae at least 1,000 data points to support the aerage. lortunately, or
many \eb sites, 1,000 data points come ery easily, as the aerage isitor` spends quite a
bit o time online.
11o CIupLer 11

CHAPTIR 11
MIASLRING RIACH


(Although) you might have developed the best communication material on the
market, if your Web site does not deliver on the promise made in this material you
will experience large dropout volumes. Reach techniques must reflect the purpose
and capability of Web sitesbecause anything else misleads people as to the content
of the site.
Hurol Inan in Measuring the Success of your Web site


On the Internet reach is ery diicult to quantiy, our best estimates o reach are built rom
demographic sureys and estimates o the number o people who will pause long enough
while browsing some other message to consider what you hae to oer. Marketing
proessionals attempt to buy adertising in such a way that they will reach the greatest
number o people they beliee are likely to respond to their message and show them that
message with the right` requency-the idea being that i you show qualiied` indiiduals
the right message the right number o times they will respond aorably.

1he problem with translating buying adertising in more traditional media to the Internet is
that oline people are buying media impressions and online they are buying ad`
impressions. 1he oline metric is typically easy to quantiy-the number and demographic
o subscribers to a magazine, the Nielsen ratings or a teleision program, the number
newspaper sales within speciic zip codes, and so on. But because the Internet is anonymous
by design, the same leel o accuracy in numbers is diicult to attain. Oten adertisers are
hard pressed to proide meaningul and accurate numbers to media buyers regarding the
number o unique isitors they reach and the requency with which they reach them. And
requently media buyers are distrustul o the numbers that are reported.
General Metrics to Help Measure Reach
1he ollowing metrics are recommended to help you understand the eect o marketing
programs designed to reach new isitors in an eort to attract them to your \eb site.
Overall Traffic Volumes
Because reach and requency are ery diicult to measure accurately online, \eb analysts
oten look towards proxy` measurements to help them understand how well they are
reaching people. Lxamining the oerall traic olume coming to one`s \eb site is an
eectie proxy or reach in that you cannot attract people that you do not reach in some way
MeusurIng ReucI 111

or another. Put another way, i you are doing anything online or oline to let people know
you exist then you are attempting to reach them and attract them to your \eb site. Oerall
traic olumes are a moderately useul key perormance indicator, depending on your
oerall business goals.
How to Calculate
In most cases some measure o oerall traic olume is reported directly, likely in a page
iews` or total page iews` metric broken down by the hour in a day iew, the day in a
week or month iew and the month in a year iew. Some \eb analytics applications proide
aerages and orecasting with this metric, which can be helpul i or no other reason that it
proides additional isual clues as to how your traic olumes are changing oer time.
Dependence
I you are looking at a page iew` metric, keep in mind the pyramid model o \eb analytics
data ,Chapter 4,. 1here are isitors coming to your \eb site generating these page iews. It
is important to vot assume that a large spike in page iew traic is directly correlated to a
similar increase in the number o isitors. Lspecially i you are using \eb-serer log iles as
a data source, automated agents could potentially cause spikes in page iew traic. \hen
you obsere spikes in page iew traic be careul to look at isit and isitor metrics to
alidate that this increase in traic is concurrent with additional isits or isitors.

Oerall traic olume is one \eb analytics metric that is best receied in real time` ,see
Chapter 3 or more on real-time data deliery,. laing this inormation as close to the time
the data is actually generated allows or you to respond rapidly to changes, either beneicial
or detrimental.
Usage
\ou should pay close attention to oerall traic olumes simply because this metric is the
single best indicator that something has changed in your relationship with your online
isitors. Pronounced peaks and alleys will appear in your traic olume reporting more
quickly than higher-leel data such as leads, purchases or complaints to your customer
serice phone center.
Interpretation
In general you are looking or either spikes or dips in traic that exceed the aerage or your
timerame or simply look dierent.` I you get into the habit o checking this metric two
or three times each day you will easily pick out changes in normal patterns that will alert you
that additional research is necessary. Oerall traic olume is not an interesting metric in
and o itsel, the alue this metric proides is to sere as a guide, letting you know when you
should look more closely at inormation about changes in the relationship between your
\eb site and the isitors who are currently seeing your marketing message.
11 CIupLer 11

Example
BackcountryStore.com watches its traic olumes closely to ensure it is able to capture
emerging trends and actiity exhibited by its customers. One o the things
BackcountryStore.com watches most closely or is the traic pattern exhibited in ligure 23.



Iigure 23: 1raffic pattern measured for a single day on BackcountryStore.com's Web
site.

1he yellow line on the graph in ligure 23 indicates the aerage traic olume, measured in
page iews, or the day in question. 1he large spike in traic occurring at 9:00 A.M. was
strongly correlated to an outbound email campaign adertising large holiday discounts on
gear sold at BackcountryStore.com. Based on nothing more than this pattern, the marketing
group at BackcountryStore.com can be assured that their campaign is working. Still, they
would want to drill-down and determine A, how much o this spike can be attributed
directly to the email campaign and B, what percentage o email recipients responded.
Related Metrics
1he metrics most commonly tied to the interpretation o changes in oerall traic olume
are the reerring source metrics discussed in Chapter 12: reerring domains and URLs, search
engines and search keywords and phrases. Campaign analytics are also ery closely tied to
changes in oerall traic olumes.
Number of Visits
Visits are closely related to oerall traic olume as a direct measure o the number o
people you are able to successully reach online. Visits are important to examine in this
context since page iews do not coney all o the necessary inormation about the traic
coming to your \eb site. Page iews are about the number o clicks` a person makes,
whereas isits are about the people themseles.
MeusurIng ReucI 11

How to Calculate
Similar to oerall traic olume measured by page iews, isits are typically reported directly
by your \eb analytics package. Ideally your application also proides aerages and orecasts
to help isualize the eects o changes in isit and isitor patterns. An important thing to
remember is the deinition o isit rom Chapter 4: a isit is composed o measured page
iews or requests ollowed by at least 30 minutes o inactiity.
Dependence
Like oerall traic olume, isits are dependent on the number o isitors coming to the
\eb site. Oten times when you obsere a large spike in page iews it is worthwhile to
immediately check isits to see whether the spike can be attributed to a similar spike in the
number o isits to the \eb site.
Usage
\hile it is good to use isits as a leel set` or oerall traic olumes, isits are also a key
indicator or how eectiely you are able to reach and begin to acquire isitors. 1here are
two common and useul comparisons you can make with the isits metric:

1otal olume o isits compared to preious similar timerames
1otal olume o isits compared to online marketing actiity

1o get a clear picture o how traic measured in isits is changing you should be sure to
compare current isits to isits rom recent timerames. lor example, compare the current
week`s Monday to last week`s Monday, the current ull week to the last ull week or the
current weekend to the last two or three weekends. 1he measurement should be a
percentage increase or decrease rom period to period. Comparing current olume o isits
to the olume rom last year can be inormatie rom a gross reporting perspectie ,saying
\e hae successully grown traic year oer year by 25 percent, measured by total isits to
the \eb site`,, but this inormation is typically less actionable in the continuous
improement model since your \eb site has, hopeully, changed oer this period o time.

\hen comparing isits to online marketing actiity you simply want to be able to annotate
any chart or graph describing the number o isits to the \eb site with inormation about
which portions o your current marketing program are likely driing isitors to your site. It
is important to keep in mind that not eeryone will take the time to read a book like !eb
.vat,tic. Dev,.tifiea and understand what these numbers mean and what they can tell you.
Annotation and the direct mapping o this data back to inormation that is common to your
business ,marketing campaigns, partnerships, releant keywords being searched or at
Google, and so on, will help others connect with this data in meaningul ways.
Interpretation
Similar to oerall traic olume you should be looking or changes in isit patterns that are
indicatie o a change in the relationship between the people you are trying to reach and
11q CIupLer 11

your \eb site. \hile slightly more useul than page iews by irtue o being about people,
not just clicks, the real alue o isits is ound in comparatie analysis.
Example
loliday traic is perhaps the single most important source o income to online retailers,
including BackcountryStore.com. Growth in the number o isits is a key indicator o the
likelihood o reenue growth, as some known percentage o these isits will conert into
purchases. BackcountryStore.com`s isiting traic pattern leading into Christmas 2003
shows nothing but excellent news ,ligure 24,.



Iigure 24: Graph of visits to BackcountryStore.com in December 2003. 1he yellow
line represents the period average, which BackcountryStore.com clearly exceeded
right up until the last few days before the holiday.

\hile BackcountryStore.com pays close attention to the raw numbers o isits to their \eb
site, o greater interest is the year-oer-year growth ,not shown,. laing this inormation
helps BackcountryStore.com understand not only how it is exceeding recent traic patterns
but also how their business is truly growing oer time.
Related Metrics
Similar to oerall traic olume as measured by page iews, the metrics that are most
commonly tied to interpretation o changes in number o isits are the reerring sources
metrics discussed in the chapters on acquisition and campaign analytics.
Number of New Visitors
1he number o new isitors you are able to attract to your \eb site is a direct measure o
the eicacy o your reach. New isitor metrics reported by \eb analytics packages generally
attempt to quantiy the number o new eer` isitors, that is, those isitors who hae neer
been to your \eb site preiously. Like page iews and isits, new isitors is a measurement
that spans both reach and acquisition.
MeusurIng ReucI 11g

How to Calculate
New isitors is a metric that most reporting packages proide with no additional setup.
Check your solution, but this metric should proide inormation about the number o
isitors who hae neer been to your \eb site preiously and who hae isited in the
timerame under analysis.
Dependence
1he primary dependence o any new isitors report is on the method or determining the
newness` and uniqueness o a isitor. Cookies are the preerred method since they proide
a long-term storage mechanism or keeping track o this inormation. But een with cookies
there is always the issue o whether a isitor is new` on one computer,browser but has
isited preiously on another computer,browser ,see Chapter 3 or more on cookies used by
\eb analytics applications,. 1he only sure way to determine whether a isitor is truly new is
to require isitors to log in-an impractical request or most online businesses.
Usage
1he new isitor metric is typically used in conjunction with a count o returning isitors, or
all isitors, to calculate ratios useul in determining the quality o isitor acquisition. As a
stand-alone metric it is good to watch new isitor acquisition patterns to answer questions
like:

Do you reach new people more eectiely on weekends, early in the week or later
in the week
Are new isitors more likely to isit during the daytime or in the eening
Do new isitors react to new marketing materials immediately or is there lag time
between media drops and new isitor response

Correlating new isitor acquisition to campaign actiity can proide aluable data or
optimizing uture campaigns. Consider two media campaigns, one launched on a Saturday
morning and the other on a Monday aternoon. \atching the growth o new isitor
acquisition ater each drop can help you understand which time is more likely to attract the
greatest percentage o new isitors relatie to oerall isitor traic.
Interpretation
lor any gien time period, the number o reported new isitors is simply that, the number o
isitors who hae isited the \eb site or the irst time eer.
Example
One thing that BackcountryStore.com pays close attention to is its rate o growth or
shrinkage in the olume and percentage o new isitors it attracts to its \eb site. As you can
see in ligure 25, BackcountryStore.com experienced a steady growth in the number o new
isitors coming to the \eb site throughout the ourth quarter o 2003.
116 CIupLer 11




Iigure 2S: Steady, continuous growth in the raw number of new visitors to the
BackcountryStore.com Web site in Q4, 2003. While this growth is clearly good news,
a better analysis would be to compare this growth to the total number of visitors to
the Web site to determine whether the percentage of new visitors is also growing.

Related Metrics
Perhaps more important than the raw number o new isitors is the relationship between
new and returning isitors and that between new isitors and all isitors, discussed below.
Additionally, the number o new isitors acquired is tied to reerring sources metrics
discussed in the chapters on acquisition and campaign analytics.
Ratio of New Visitors to Returning Visitors
1he ratio o new to returning isitors is a key perormance indicator or reach and
acquisition, one that can help you easily determine your \eb site`s current acquisition
mode.` Lssentially a simple calculation, the ratio tells you how many new isitors you reach
and acquire compared to the number o isitors you retain or any gien timerame. 1his
metric is an excellent key perormance indicator.
How to Calculate
Diide the number o new isitors by the number o returning isitors reported or the
timerame under analysis:

NEW VISTORS / RETURNING VISITORS = ACQUISITION MODE

It is ery important to use returning isitors that are directly comparable to your new isitor
metric, most likely day, or daily returning, isitors. Since new isitors are most typically
reported on a daily basis, i you are to use a non-day metric or returning isitors you will
skew the measurement o acquisition mode.

MeusurIng ReucI 11,

Consider 100 new isitors summed oer a single week, 50 daily returning isitors summed
oer the same week and 30 weekly returning isitors or the week. 1he calculation or each
would be:

100 new isitors , 50 daily returning isitors ~ 2.00
100 new isitors , 30 weekly returning isitors ~ 3.33

lrom an interpretation standpoint these two measurements o acquisition mode` are 66
percent dierent, certainly a large enough ariance to result in potentially dierent actions
taken.
Dependence
1he ratio o new to returning isitors depends on an accurate measurement o the indiidual
numbers. 1he use o cookies is preerred or accurately measuring o new and returning
isitors, but you are encouraged to read the section in Chapter 3 on cookie usage in analytics
programs.

It is important to note that i you hae recently moed to a new application or strategy or
measuring new isitors that this metric will be meaningless due to the act that nearly all o
your isitors will be counted as new isitors.` I this is the case you should consider
measuring, but not acting on, this measurement or some period o time, allowing isitors to
be counted as new or the irst time and time or their retention patterns to deelop.
Usage
1his ratio is a key perormance indicator or any \eb site, one that should be an integrated
part o your regular reporting program. \ou should monitor this ratio on a daily or weekly
basis, watching or pronounced changes, which can be leading indicators o a signiicant
change in the relationship between your marketing programs, the \eb site and the audience
you are trying to reach. 1his KPI has the ability to inorm on the eicacy o both acquisition
and retention marketing programs.
Interpretation
Consider three examples:

Company A` reports 100,000 new isitors and 30,000 daily returning isitors
summed or the month. 1he acquisition mode calculation or company A` is
3.33. 1his means that or each isitor that returns to this company`s \eb site, 3.33
new isitors are successully reached and acquired.
Company B` reports 50,000 new isitors and 80,000 daily returning isitors
summed or the month. 1he acquisition mode calculation or company B` is
0.63. 1his means that or eery two new isitors that comes to this company`s
\eb site, three isitors return in the same timerame.
Company C` reports 180,000 new isitors and 16,000 daily returning isitors
summed or the month. 1he acquisition mode calculation or company C` is
11S CIupLer 11

1.02. Company C` gets about one new isitor or eery one returning isitor in
the timerame.

lrom the aboe data we can determine that company A` is in new isitor growth and
acquisition mode, company B` is in actie retention mode and company C` has ound a
healthy balance o growth and retention. Depending on your acquisition model there are
optimal targets or this calculation:

I your business is in a strong growth mode, perhaps because you are new or you
are releasing a new product or product line, this ratio should be between 2.00 and
5.00. Less than 2.00 and you would not describe your new isitor reach and
acquisition as strong.` More than 5.00 and you are likely not doing enough to
retain isitors.
I your business depends on isitors returning, either because you hae a complex
sales process or you are working to deelop an online community o loyal isitors,
this ratio should be between 0.25 and 2.00. Less than 0.25 and you are simply not
attracting many new isitors. More than 2.00 and you are beginning to drit into
positie acquisition mode.
I you are doing little marketing but hae a well-designed \eb site that oers
compelling inormation or products, likely your ratio is somewhere between 0.5
and 1.25. 1his is an indicator o a healthy \eb site but one that is not actiely
trying to acquire either new or returning isitors.
Related Metrics
New isitors reached and returning isitors retained are tied to reerring sources metrics
discussed in the chapters on acquisition and campaign analytics, as well as the retention
metrics discussed in Chapter 14.
Percentage of New Visitors
1he percentage o new isitors to total isitors is perhaps the best measure o how
eectiely you are able to reach people on the \eb. 1he ratio is a straight percentage telling
you how many people, out o your total audience, were new prospects who had somehow
been attracted by your marketing program. 1his metric is an excellent candidate or usage as
a key perormance indicator.
How to Calculate
Diide the number o new isitors reported by your analytics package by the number o
isitors that isited in the same timerame.

NEW VISITORS / UNIQUE VISITORS = PERCENT NEW VISITORS

One subtlety is that many analytics packages will report dierent types o isitors-daily,
weekly, monthly and occasionally or odd timerames. \ou want to be careul you
understand what kind` o isitors you are using or your comparison.
MeusurIng ReucI 11q

Dependence
1his number is dependent on how your particular analytics package measures new and
unique isitors. Ideally someone can only be a new isitor to your \eb site one time eer
,see number o new isitors` discussion,. Visitors, on the other hand, can be measured a
number o dierent ways ,see Chapter 4,. 1he most important thing is that you are using
the same timerame to get both numbers, as ailure to do so will skew the resulting
percentage. In general it is best to use the same type o unique isitors or the reporting
timerame, that is, i you are looking at new isitors or the week use weekly unique isitors.
Usage
1he percentage o new isitors is an excellent key perormance indicator or your marketing
programs. It is a single number that can be tracked on a daily basis that allows you to
quickly determine the eect o actie marketing programs. I the percentage o new isitors
is increasing day-oer-day you know that your marketing programs are haing a positie
eect-you hae successully reached more new prospects today than you did yesterday.
1he conerse is also true.
Interpretation
1he number generated is the percentage o all isitors to your \eb site in the timerame
under examination who were completely new to your site. A alue o 30 percent` means
that 30 percent o your isiting traic was made up o new people successully reached and
brought to your \eb site or the irst time eer.
Example
BackcountryStore.com keeps track o the percentage o new isitors that are coming to its
\eb site to achiee and presere an optimal mix based on past perormance. It knows that
its customers are unlikely to make more than a handul o purchases throughout the year so
BackcountryStore.com depends heaily on attracting new isitors in hopes o conerting
them to loyal, lietime customers.

Oer time, BackcountryStore.com has learned that about 5 to 85 percent is the best mix o
new to returning isitors coming to its \eb site, and, using automated reporting proided by
their \eb analytics endor, it monitors the percent o new isitors coming to the \eb site
,ligure 26,.



Iigure 26: Percent new visitors monitored on BackcountryStore.com via regular KPI
reporting. BackcountryStore.com has optimized its marketing mix to achieve a
roughly 7S to 8S percent mix of new to returning visitors.
1o CIupLer 11


Related Metrics
New isitors are tied to the ratio o new isitors to returning isitors ,acquisition mode`,, as
well as the reerring sources metrics discussed in the chapters on acquisition and campaign
analytics, as well as the retention metrics discussed in Chapter 14.
Entry Pages and Contents
Lntry pages and contents are the way that many \eb analytics applications reer to the irst
pages and irst content groups that a isitor sees when they isit your \eb site. I you
consider the deinition o a isit` ,Chapter 4,, you will see that isit implies a irst and last
page in eery isit, although they may be the same page. Paying close attention to entry
pages and entry content is an excellent way to increase your chances o acquiring and
conerting the isitors you are able to reach.
How to Calculate
Lntry pages and contents are metrics that are typically reported by your \eb analytics
application. Oten grouped under page` and content` metrics, most endors report the
number o isits that begin at each page or content group listed.
Dependence
Lntry pages and contents are dependent on how your application deines a isit or session.`
I your application is not collecting inormation about pages iewed on a per isit` basis
you will likely be unable to make use o these metrics.
Usage
Lntry pages and contents are critical to ensuring that you maintain a good chance to connect
with each isitor you are able to reach. An important actor in reaching and acquiring
isitors to your \eb site is expectation setting-making sure that the message a prospect
sees when they iew your banner, read your email, ind your search results, and so on, is
consistent with the message a isitor sees when they land on your \eb site. I a isitor
thinks they are getting A` when they click on your link or type in your URL but don`t see
any reerence to A` when they arrie on your site, the chances that they will click the back
button increases signiicantly.

Monitor your top entry pages and content areas closely to ensure that the message on these
pages is consistent with your oerall marketing program. I you obsere that a page or
content area is appearing unexpectedly high on this list take a closer look at it to ensure that
the message is consistent.
MeusurIng ReucI 11

Interpretation
I your \eb analytics package reports that entry page A had 10,000 isits` this tells you that
10,000 isits to your \eb site started at page A.` \our particular analytics package may
also report that 5,000 isitors entered your \eb site at page B,` which is slightly dierent
but still a measure o the relatie popularity o page B` as an entry point into your \eb site.
Example
BackcountryStore.com keeps track o the most popular entry pages to its \eb site in order
to make sure that the home page eatures the products that its isitors are most interested in
,ligure 2,. By monitoring the percentage o entries into the \eb site by page and product
category, BackcountryStore.com is able to ensure that the new isitors, who make up the
largest part o their isiting audience, are able to ind the most popular items quickly.



Iigure 27: 1op ten entry pages by percentage of entry page views.

Related Metrics
Lntry pages and contents are closely related to other marketing reach and acquisition metrics
described in the chapters on acquisition and campaign analytics. Also, entry pages and
content are closely reported to the acquisition metrics o immediate abandonment`
reported in Chapter 12.
Visitor Geographic Information
Keeping an eye on isitor geographic inormation is a simple recognition that the Internet is
global, een i your marketing programs are local. Monitoring where your incoming traic is
coming rom geographically will increase your chances o taking adantage o unexpected
traic rom outside your target market.
1 CIupLer 11

How to Calculate
Visitor geographic distribution is oten included in \eb analytics packages, typically
organized with isitor` or demographic` metrics. Dierent analytics packages report this
inormation to dierent leels o granularity-some only reporting to the leel o continent,
others going down to the leel o state, city and een zip code.
Dependence
Visitor geographic inormation is dependent on the method used to assign an anonymous
isitor to a geographic region. Some \eb analytics packages, notably those that depend on
\eb-serer log iles, hae some diiculty with this since isitor IP address is oten not the
best data source or this assignment. Still, some endors hae partnered with content
deliery networks like Akamai and Speedera to reine IP addresses to physical locations with
a high degree o accuracy ,see http:,,www.akamai.com,en,html,serices,edge_how_
it_works.html or an example o the kinds o data that are aailable in a model like this,.
Usage
It is recommended that you monitor the geographic distribution o your isitors occasionally
to watch or spikes in traic rom locations you do not expect. 1he classic example o this
is the online business that was doing no marketing or promotion oerseas but who obsered
a signiicant traic presence coming rom Japan. Knowing this allowed the company to
experiment with landing pages and marketing campaigns to better speak to this speciic
audience. 1he end result was a signiicant increase in oerall online reenue rom Japanese
isitors.
Interpretation
Again, most \eb analytics packages report geographic inormation on either a isitor or isit
basis. I your \eb analytics package reports that you had 100,000 isitors rom Canada` in
the last month or 50,000 isits were rom isitors in China` you can determine i you
should attempt to reach isitors you currently consider to be out o market.`
Example
As BackcountryStore.com expands into the oline world o catalog sales it has an increased
interest in knowing more about its isitors` speciic locations. Knowing more about where
people are geographically when they ind BackcountryStore.com online can help better target
catalog mailings, especially considering the cost o lists o names. BackcountryStore.com`s
analytics proider is able to proide isitor geographic inormation down to the leel o the
city, allowing BackcountryStore.com`s catalog group the ability to watch or cities and states
that may stand out as being more actie online.


MeusurIng ReucI 1


Iigure 28: Visitor cities by percent of visitors visiting BackcountryStore.com.

As you can see in ligure 28, with the exception o New \ork, New \ork, isiting traic is
pretty well distributed throughout North America. It is interesting to note the traic coming
rom Canada ,line 13, and Great Britain ,line 14,, both areas that BackcountryStore.com
currently considers out o market` or shipping.
Error Pages
Oten oerlooked in a \eb analytics program, since it is perceied as being more I1`s
responsibility than marketing`s, is monitoring error pages-typically ile not ound ,404,`
errors. Nothing is more rustrating or a new isitor than getting an error or running into a
dead end. Len i you hae created a nice looking we`re sorry but that documents has
moed .` page you are still making a isitor wait or a page to load that was not what they
were looking or or requesting. Lrror pages make a aluable addition to your key
perormance indicator program or reach and acquisition.
How to Calculate
\our \eb analytics package should either generate error page inormation or you
automatically ,common in \eb-serer log-ile-based applications, or proide an optional
script to allow you to track these pages. \eb-serer log-ile-based solutions will typically
report on a ariety o errors ,ile not ound, serer errors, and so on, since this inormation
is reely aailable rom the log ile. Client-side data collection solutions are typically more
limited in their ability to report, oten depending on a custom error page being deliered by
your \eb serer.
Dependence
1racking error pages is wholly dependent on your \eb analytics applications ability to
record and present this inormation.
1q CIupLer 11

Usage
Lrror page reporting should be monitored on a weekly, i not daily, basis. Considering the
amount o time, eort and money that most online businesses spend on marketing and
marketing optimization, and since nothing good comes o a isitor seeing an error page, it
only makes sense to work diligently on reducing the number o times isitors get errors
during a isit.

lollow your error report closely and do two things to any page that is being requested in
error` as quickly as possible. lirst, implement a redirect o some kind rom the error page
to the correct page or at least a closely related page. Second, i your analytics package also
reports on the pages that contained links to the broken page, correct the linkage on those
pages i they are on your site or, i the links are on another \eb site, write to the owners o
the site and ask them to correct the broken links.
Interpretation
Most error page reports indicate the name or URL o the page that the isitor was ultimately
looking or and the number o times isitors saw the error page. \ou should be watching
closely and looking or error pages that are a receiing a signiicant amount o traic as
measured by total number o page iews or as a large percentage o all error pages.
Interested Visitor Share
Adapted rom Bryan Lisenberg and Jim Noo`s Cviae to !eb .vat,tic. description o
committed isitor share,` this metric can help the online business understand what percent
o isitors spend a moderate period o time interacting with the \eb site. Much like aerage
time spent on site, interested isitor share can be a powerul perormance indictor or reach,
helping you to identiy changes in the composition o the audience you are reaching.
How to Calculate
1he interested isitor share calculation is simply the number o isits that lasted longer than
n` minutes diided by the total number o isits:

VISITS OF MORE THAN N MINUTES / TOTAL VISITS = INTERESTED VISITOR
SHARE

Lisenberg and Noo do not recommend a alue or n` that will speciically indicate that a
isitor is in act interested` in your \eb site. Rather their description o committed isitor
index` discusses isitors who spend more than 19 minutes interacting with the \eb site.
One can reasonably set this number, howeer, by examining the aerage time spent on site,
commonly aailable in most \eb analytics packages. It would be reasonable to set n` to
the aerage number o minutes isitors spend on your \eb site, thus proiding an easy KPI
or the percent o isitors who spend more than an aerage amount o time browsing.
MeusurIng ReucI 1g

Dependence
1o calculate the interested isitor share all you need be able to do is determine the total
number o isits exceeding n` minutes or any period o time. I your \eb analytics
application has a time spent on site` metric or report this is the most likely source or the
data required to calculate interested isitor share.
Usage
1he interested isitor share can be used to understand changes in the audience you are
currently reaching and bringing to the \eb site. I your goal is to bring isitors to the site
that will spend ie or more minutes interacting with the site then it would be useul to set
n` to 5 minutes` and calculate the total share o isitors who are spending more than ie
minutes, showing greater a than aerage interest in your message and content.
Interpretation
1he interested isitor share is a percentage o all isitors who spend more than n` number
o minutes on your \eb site. 1he actual number o minutes will depend on your speciic
business model and \eb site goals. An educational, news or entertainment \eb site would
want to hae a much higher interested isitor share than would a customer support site. In
the ormer case, a isitor spending more time is a good thing as they are likely to connect
more closely with the content and generate additional adertising iews in the process. In
the latter case, someone spending more time is potentially a bad thing, sering as an
indicator that they are struggling to ind the inormation they need to troubleshoot whateer
problem they may be haing at the time.
Example
BackcountryStore.com keeps track o its interested and committed isitor share and
compares these numbers to its oerall conersion rate ,ligure 29,. BackcountryStore.com is
not necessarily interested in isitors spending long periods o time browsing their \eb
site-its hope is that isitors are able to ind what they are looking or and make purchases
quickly and easily-but it recognizes that there is some correlation between the site`s
conersion rate and isitors being interested` and committed.`



Iigure 29: Interested and committed visitor share measured for
BackcountryStore.com.

16 CIupLer 11

Related Metrics
Interested isitor share is related to all time spent on site metrics, including aerage time
spent on site and percent o isits less than 90 seconds presented in Chapter 12.
Campaign Reach Metrics
1hroughout !eb .vat,tic. Dev,.tifiea we will reisit the metrics used to measure the
eectieness o campaigns designed to drie traic to one`s \eb site. \hile the majority o
the alue deried rom the campaign analysis tools described in Chapter 8 is associated with
measuring acquisition and conersion, the ollowing two metrics used to measure the
eectie reach o a campaign are important to creating a holistic iew o how your
campaigns support your online business.
Impressions Served
1he measurement o campaigns crosses oer all stages in the customer lie cycle. Robust
and well-implemented campaign analysis tools will help you to understand the reach and
requency or each ad, the response rate or the campaign as a measure o acquisition, the
conersion rate-een metrics o retention and loyalty. 1o calculate reach and requency,
the primary metric used is the number o impressions sered.
How to Calculate
Measuring and calculating impressions sered is highly dependent on both your \eb
analytics package and the type o impression you are attempting to measure. In many
instances the measurement o impressions sered must be made by the endor sering the
impressions, as in the case o running your ad on a banner network or purchasing search
engine keywords or placement ia Google or Oerture. In other instances \eb-serer log
analysis applications are able to report the number o impressions sered ,i images are
requested rom the local \eb serer, simply by determining the number o requests or
campaign images. In still other instances impressions can be tracked rom within l1ML
email campaigns by sending a simple image request back to a \eb serer.

In general it is best to consult your \eb analytics sotware or serice proider or the details
about how to get this inormation i you are not already able to get it rom the endor
sering the ads.
Dependence
Measuring impressions sered or a particular campaign is dependent on your ability to do so
within or outside o your \eb analytics application. In general it is ar more common to get
this inormation rom an outside endor, such as a banner network, search engine or email
campaign management solution, than it is to be able to measure using your own \eb
analytics application. \ith this in mind it is important to understand how the outside
MeusurIng ReucI 1,

endor makes the impression measurement-is it based on page iews or some isitor
measurement Is it set up to record immediate subsequent impressions ,such as a reload, or
ignore them

Making sure that you can reconcile and understand how these measurements are made will
allow or more meaningul integration o this data into your oerall \eb analytics program.
Usage
Impressions sered are an important metric or calculating the reach and oerall success o
any campaign. \ithout knowing how many people saw the message it is impossible to know
how compelling the message was or calculate the oerall response rate or the campaign.
Again it is important to reconcile any dierences in how impressions sered are measured
with the method or measuring responses to be sure you are making an apples to apples`
comparison.
Interpretation
Impressions sered are critical to determining the quality o any online opportunity to
present your marketing message. \ou want to compare the number o impressions sered
with the number o impressions you purchased or were promised to ensure you are getting
what you are paying or. \ou want to compare this number with the number o isitors who
iewed the impression to calculate reach and requency, that is, i you sere 100,000
impressions to 50,000 isitors to a \eb site oer a month your reach is 50,000 and your
aerage requency is 2.00 impressions per isitor.

In the end the most important reason to measure impressions sered is to proide the
denominator or the calculation o response rate or a campaign, you need to know how
many people saw an ad to determine the rate at which people actually responded.
Open Rates
Open rate is a campaign analysis metric speciic to email indicating the percentage o all
emails sent that were actually iewed. 1he open rate is important, especially when you are
purchasing a list o names and email addresses, as your message only has a chance to reach
people when they actually see and read the message.
How to Calculate
Open rate is calculated by diiding the number o email addresses the message was sent to
by the number o unique isitors who opened the mailing. Ideally:

UNIQUE READERS / TOTAL EMAIL ADDRESSES = OPEN RATE

Since it can be ery diicult to measure the number o unique readers it is reasonable to use
the measure o impressions sered and assume a 1:1 ratio o impressions to unique readers.
1his is potentially dangerous i you consider a reader who is so interested in your message
1S CIupLer 11

that he or she opens the email and reads the message seeral times but with large numbers
this eect is minimized and it`s better to hae a rough` open rate than no inormation
about the number o readers at all. So:

AUDIENCE IMPRESSIONS / TOTAL EMAIL ADDRESSES = ROUGH OPEN RATE
Dependence
Since it is more likely you will be generating a rough` open rate you should also keep in
mind that this calculation depends on the likelihood the email addresses you use are iable,
that is that your emails will actually be deliered to people. Ideally you are able to keep track
o both bounced emails ,or whom no recipient could be located, and your opt-out rate. I
you send a message to 100,000 email addresses and get 18,000 bounce backs and 2,000 opt-
out requests your reised total email addresses should be 80,000 ,100,000 - 18,000 bounces -
2,000 opt-outs ~ 80,000 total email addresses,. Doing so will increase the accuracy o the
calculation.

Measuring open rates is dependent on your ability to accurately measure the number o
emails actually opened, oten using some type o impression measurement tool. Both \eb-
serer log ile and page-tag-based applications are able to support this, typically by
embedding an inisible image into the email. It should be noted that this technique ovt,
works or l1ML emails, not text-based emails. But many businesses hae expressed their
comort in generating an open rate or l1ML email and then using the same percentage
when analyzing text-only email.
Usage
Calculating the open rate or any email message you send to people will let you determine
three things:

Quality o email list
Receptieness o the recipients to the message subject
1rustworthiness or strength o brand` or your company

All a prospect sees when they get an email, unless they hae some kind o preiew pane, is a
subject line and an idea o who was sending the email in the lrom` line. I your subject
line is not well written or the recipient has no idea who the sender is, especially in this age o
spam and email-deliered iruses, chances are your email will be deleted, unopened. I your
brand is a known quantity, you may hae a slightly higher chance o haing your message
iewed. Regardless, open rate proides a good opportunity to test the relatie quality o
subject lines.

Gien a large enough number o addresses it is relatiely simple to test subject lines by
simply diiding your audience up into test groups. A recommended method or testing is to
separate out enough o your audience to test two or three dierent subject lines in an eort
to optimize your open rate and still hae enough audience let oer or a mass-mailing.

MeusurIng ReucI 1q

Consider a total audience o 100,000 email addresses:

Separate out 18,000 or testing, leaing 82,000 email addresses.
Split the 18,000 into three groups o 6,000 test subjects.`
Send each o the three groups the same email, rom the same person or email alias,
with three dierent subject lines.
Measure the open rates or each o the three audiences. 1ake this a step urther
and measure response and conersion rates or each ,Chapters 12 and 13,.

lollowing this simple testing strategy will allow you to determine which subject line your
total audience is most likely to be most receptie to. In a perect world one o the subject
lines will stand out as most likely to be opened` and will hae a response and conersion
rate that is equally high as the other test messages. 1his subject line would be the best
candidate to use or the remaining 82,000 recipients and will, hopeully, delier the largest
number o responses and conersions.

Interpretation
Open rate is simply the percentage o emails receied that were actually opened or iewing,
higher percentages indicate more actual messages iewed. 1his number can be inlated,
dramatically in some instances, by the use o a preiew pane in some email browsing
applications. Unortunately the author knows o no way to measure the percentage o
messages browsed in a preiew pane ersus the number opened intentionally.
About Email Marketing
Open rates are a metric speciic to a ery important category o measurement, that o the
eicacy o your email marketing. Jim Sterne, a recognized authority on email marketing and
the author o .aravcea vait Mar/etivg has this to say about email marketing: Lmail is an
eectie message-testing tool. \hile you hae likely used email to drie \eb traic and
sales, you may not hae used it as a tool or testing your brand message.` Jim makes an
excellent point, that email is both an eectie tool or communication but also a ery
aluable component in the continuous improement process on one`s \eb site.

In his book, Jim discusses a handul o undamental email marketing metrics that are not
explicitly dealt with in !eb .vat,tic. Dev,.tifiea. Briely, these metrics are:

1otal Sends - the number o email addresses you originally sent your message to,
beore any bounces or ailed notiications hae returned.
Successful Sends - the percentage o total sends that are not returned or some
reason.
Hard Bounces - the number o undeliered mails that are due to an address no
longer being alid.
Soft Bounces - the number o undeliered mails that are only temporarily
undeliered due to sotware or serer error. 1he author o !eb .vat,tic. Dev,.tifiea
1o CIupLer 11

also considers out-o-oice` messages sot bounces since some people will
ultimately read your message but alas, many will not.
Messages Missing-in-Action - the ery diicult to track number o messages
that get caught in anti-spam sotware, etc.
1racked Opens - the number o l1ML emails opened tracked using an
impression measurement tool, such as the one described in this chapter.
Lstimated Opens - the estimated number o total emails opened or both l1ML
and text messages.
1racked Clickthroughs - een though the author does not like the term click-
through` ,see Chapter 4,, Mr. Sterne cites this metric as the number o successul
clicks` on a link in an email. 1he author preers to measure the number o
responses and respondents ,described in Chapter 12,.
Iorwards and Referrals - diicult to accurately measure but inaluable in
understanding how your message was ultimately receied by your target audience.

Rather than spend too much time drilling-down into the subtleties o email marketing, the
author instead gleeully reers the reader to Mr. Sterne`s book. I you`re doing any kind o
email marketing you deinitely should get a copy o this book rom Lyris 1echnologies
,www.lyris.com, and make sure that you A, read it and B, understand it.
Key Performance Indicators Recommended for
Measuring Reach
Part o the demystiication process is helping you to understand the metrics listed in this
chapter and make good use o them on a regular basis. 1o this end, seeral o the metrics
hae been identiied as good candidates or regular key perormance indicators.` Ideally
any reader o !eb .vat,tic. Dev,.tifiea will, upon inishing the book, be able to construct a
KPI report or their online business and make these metrics aailable to others within their
organization on a regular basis.

1o summarize, the key perormance indicators the author recommends or reach, keeping in
mind that these are metrics that are typically useul or att business models, include:

Oerall traic olumes, to ensure that you are tracking large spikes or dips in the
number o page iews being requested.
Ratio o new to returning isitors as an indicator o changes in the oerall makeup
o your audience.
Percentage o new isitors, to obsere the eect o changes to your marketing
reach and acquisition eorts.
Aerage pages iewed per isitor, to check whether dramatic changes in your
oerall traic olume are due to a similar change in the number o isits to the site.
Similarly, some businesses will want to monitor interested isitor share.
1op ie or ten error pages and the number o times each is seen by isitors to
quickly resole problems in the isitor experience.
All campaign metrics presented in this chapter.
MeusurIng ReucI 11


As you should see, the KPIs recommended are, or the most part, ratios, percentages or
aerages. 1hese types o numbers are typically more useul as they are able to coney a
greater amount o inormation ia a single number, one which can be easily compared to
preious time periods. Reer to Chapter 15 or additional inormation on these KPIs and
how they should roll up into a regular, automated reporting program.
Metrics You Should Be Asking for Before You Spend
Advertising Dollars
Beore you spend any money on adertising with any other online business you should ask a
handul o ery speciic questions about the type o audience they reach and how they
measure reach and requency. 1hese questions apply to any \eb site you are considering
adertising on to drie traic to your site, regardless o whether you are going to run banner
ads, sponsor content or enter into a ull-blown business partnership. 1here are slightly
dierent measurements o reach or keyword buying opportunities and email campaigns
,discussed below,.

Important questions to ask include:

low many page iews does your \eb site sere in an aerage day
low many page iews did your \eb site sere last month
low many unique isitors did you sere those pages to

1he answers to these questions will help you get a better handle on whether the site is likely
to proide a good opportunity to reach people with enough requency that your ad will likely
be seen and remembered. I you are looking to buy a million impressions sered oer the
month, but the \eb site in question only saw a million page iews last month, then you will
be super-saturating the site and annoying more people than you attract. Alternatiely i you
are looking to buy 10,000 impressions and the site is sering ten billion page iews your
adertising will probably be iewed inrequently and either missed or just not remembered,
unless you are targeting a ery speciic page or content area.

1he ratio o page iews to unique isitors, both the daily aerage and monthly measurement,
will also help you determine whether the \eb site has the right type o audience or you. I
the site reports one million page iews rom 900,000 unique isitors you can determine that
each isitor is iewing a little more than one page, on aerage. In this case, depending on
the quantity o impressions you purchase, you will either be seen only once or not at all.
Conersely, one million page iews and 100,000 unique isitors yield an aerage o 10 pages
per isit. Conentional wisdom dictates that the higher this number the more interested the
site isitor is in the content proided, and the more interested the isitor the more likely they
are to at least .ee your ad, i not respond.

\ou also want to be sure to ask how they measure their \eb site`s traic. Consider what
you learned in chapters 2 and 3 regarding dierences in data sources inorming \eb metrics,
such as the ratio o page iews to isitors. Accuracy is ery much an issue i the inormation
1 CIupLer 11

is based on \eb-serer log iles, unless the \eb site operators are diligent in their attempts
to determine the uniqueness o isitors and excluding non-human iewers when recording
page iews. It is better to hae these numbers based on a client-side data source or to get
audited numbers rom a source like I,PRO.

I you do not trust the alidity o the sources o data proided by the adertising site you
should be extra careul to measure the response and conersion o traic rom that
adertising enue. 1his is not to say that you should choose not to adertise with any \eb
site that does not hae highest-quality metrics reporting, just that you should establish a
igorous program or determining success on your end.

Another critical question to ask is, \hat is the demographic and technographic makeup o
your audience and how do you measure that inormation` More important than the
number o impressions that a \eb site will be able to sere is the type o audience they are
able to reach and the quality o it its audience has with the audience you hae identiied as
being predisposed to respond to your message. Age, race, gender and socioeconomic actors
are just as important online as they are oline when determining how adertising dollars
should be spent. Unortunately, accurately measuring these actors is extremely diicult and
is oten achieed using panel-based measurement serices such as Nielsen,NetRatings.

Many \eb sites are now taking adantage o isitor segmentation tools by capturing
demographic inormation out o their registration process to urther examine numbers and
actiity around key demographic segments. Done well, and proiding the \eb site is able to
segment a statistically releant population, this type o inormation can be ery aluable in
terms o understanding quality o it or adertising buyers. Additional inormation
regarding this segmentation strategy is proided in Chapter 14.
MeusurIng AcquIsILIon 1

CHAPTIR 1
MIASLRING ACQLISITION


1he closer the match between the capabilities of the Web site and the needs of the
customers, the greater the likelihood of success in the acquisition stage of the
customer life cycle.
Hurol Inan in Measuring the Success of your Web site


Compared to measuring reach, measuring acquisition is relatiely easy, there are seeral well-
deined metrics that report on the number o people who hae expressed at least some
interest in your \eb site. 1he most important dierentiator is that, whereas many measures
o reach depend on inormation rom other sources, measurements o acquisition can be
made rom one`s own \eb analytics data.

lor the most part the ocus on acquisition measurement is, low many isitors are we
acquiring` and, \here are we acquiring these isitors rom` 1he answers to these
questions become ery important when we moe on to examine conersion and retention.
Knowing one`s conersion rate without knowing where the most likely conerters come
rom is a good start but it`s not enough to truly be eectie on the Internet.

Note that all o the metrics presented in Chapter 11 on reach make good acquisition metrics
in many instances.
General Metrics to Help Measure Acquisition
1he ollowing are metrics recommended to help better understand how successully your
\eb site and online marketing programs are at helping you acquire isitors.
Percent New Visitors
1he percentage o new isitors, also presented in the Chapter 11, as a reach metric is perhaps
a better measurement o acquisition. Usually reported by \eb analytics packages as the
number o new eer` isitors, those isitors who hae neer been to your \eb site
preiously, new isitors are an audience segment that you want to ocus much o your
eorts on rom an acquisition,conersion standpoint.

1he percentage o new isitors in relation to all isitors is an excellent acquisition
perormance indicator. Please see the description o percent new isitors in Chapter 11 or
details about percent new isitors.
1q CIupLer 1z

Average Number of Visits per Visitor
1he measurement o the aerage number o isits per isitor is used to help you understand
the oerall leel o interest exhibited by actie isitors. 1he more actie or interested a
isitor, the greater the likelihood they will return to your \eb site and isit repeatedly. Keep
in mind that or some business models increased actiity is not necessarily a good thing. 1he
aerage number o isits per isitor is an excellent acquisition perormance indictor.
How to Calculate
1he aerage number o isits per isitor is simply calculated by taking the total number o
isits in the timerame and diiding by the number o unique isitors in the same timerame:

VISITS / VISITORS = AVERAGE VISITS PER VISITOR

1his calculation should always gie you a number greater than 1.00.

I you hae access to the isitor segmentation tools discussed in Chapter you may want to
consider generating this calculation or dierent segments o isitors-new isitors,
returning isitors, existing customers, and so on. Doing so will let you better understand
what actors may be behind any changes obsered. An increasing number o isits per
isitor may be good or new isitors but bad or existing customers on the same \eb site.
Dependence
1he major dependence in this calculation is on the isitors` measurement. As discussed in
Chapter there are seeral types` o isitor:

A daily unique isitor should be counted the irst, and only the irst, time a unique isitor
isits a \eb site on any gien day. A weekly unique isitor should be counted the irst, and
only the irst, time a unique isitor isits a \eb site in a gien week. A monthly unique
isitor should be counted the irst, and only the irst, time a unique isitor isits a \eb site
in a gien month.`

\ith this in mind you should always attempt to use the most releant deinition o isitor
or the timerame in question. I you are trying to determine the aerage number o isits
per isitor or a week, use weekly unique isitors. I you are trying to determine the aerage
number o isits per isitor or a month, use monthly unique isitors, and so on.
Usage
1he aerage number o isits per isitor should be used in conjunction with marketing
actiities to understand how interested dierent audiences are in the content, products or
inormation you proide. Additionally, you should be careul to examine this metric in the
context o your business model-it is not necessarily good news when this ratio increases on
all sites. laing a higher aerage number o isits per isitor is preerred or most online
stores where more isits potentially indicates a greater likelihood that the aerage isitor will
MeusurIng AcquIsILIon 1g

make a purchase. A customer support site with a high aerage o isits per isitor may
indicate an expression o rustration with a product or serice.
Interpretation
1he number proided by this calculation will simply be a number greater than 1.00, which
indicates the aerage o all isits per all isitors, measured in the timerame. A ratio o 3.5`
tells you that the aerage isitor came to the \eb site 3.5 times in the timerame under
examination. Another way to think about the aerage number o isits per isitor is as the
aerage requency o isit or the timerame.
Example
1he olks at BackcountryStore.com know that most o their customers are new isitors and
that isitors return inrequently to purchase new gear. 1hey keep track o the aerage
number o isits per isitor to watch or any changes in this trend but they know that, unless
they oer some dramatically dierent serice, these numbers are unlikely to change ,ligure
30,.



Iigure 30: Weekly reporting on the average number of visits per visitor for
BackcountryStore.com, calculated using weekly unique visitors.
Average Number of Page Views per Visit
1he aerage number o page iews per isit is also an excellent measurement o acquisition,
as well as the likelihood or oerall conersion in some situations. A measure o the likely
depth o isit, measured by the number o pages the aerage person is likely to see in a gien
isit, this measurement is another good proxy or the leel o interest in your content,
products or oers. I one measure o success is getting isitors to look at a large number o
pages, the larger this number the greater number o pages iewed in an aerage` isit. 1he
aerage number o page iews per isit is an excellent acquisition perormance indicator.
How to Calculate
1o calculate the aerage page iews per isit simply diide page iews into isits:

PAGE VIEWS / VISITS = AVERAGE PAGE VIEWS PER VISIT

1his calculation should always gie you a number greater than 1.00.

Much like aerage isits per isitor, i you hae access to the isitor segmentation tools
discussed in Chapter you may want to consider generating this calculation or dierent
16 CIupLer 1z

segments o isitors-new isitors, returning isitors, existing customers, and so on. Doing
so will let you better understand what actors may be behind any changes obsered. Much
like isits per isitor, an increasing number o page iews per isit may be good or new
isitors but bad or existing customers on the same \eb site.
Dependence
1his calculation depends only on haing accurate, consistent measurements or page iews
and isits.
Usage
Aerage page iews per isit is used much like isits per isitor-to help you understand the
changing nature o the isitors you are able to attract to your \eb site. Similarly you want to
be sure to consider your business model and what you are trying to get isitors to do online
when you examine this aerage.

Much like the interested isitor share described in Chapter 11 there are optimal alues or
aerage page iews per isit depending on your particular business goals. Obiously i your
business model depends on isitors iewing a larger number o pages because those pages
are tied to adertising iews you hope this number will be as large as possible. Conersely i
your isitors are looking or speciic inormation on your \eb site, perhaps to learn more
about your business, larger numbers do not necessarily mean good news.

One aluable actiity is to map out dierent scenarios that you`d like isitors to accomplish
on your \eb site and count the number o page iews each would take. Depending on the
scenarios, and the number o scenarios you can reasonably expect a isitor to complete in a
single isit, you will begin to gain an understanding o what the optimal number o page
iews per isit is or your \eb site. 1his number can then be used or a perormance target
and incorporated into your KPI program.
Interpretation
Like isits per isitor this number should always be a number greater than 1.00 and relects
the number o pages the aerage isitor saw in a isit. So a ratio o 10.8` tells you that
during an aerage isit, isitors are seeing slightly less than 11 pages.
Example
Much like monitoring the aerage number o isits to isitors, BackcountryStore.com keeps
track o the ratio o page iews per isit to monitor or any dramatic changes in the number
o pages the aerage` isitor is iewing in a isit ,ligure 31,.
MeusurIng AcquIsILIon 1,




Iigure 3J: Page views per visit, part of BackcountryStore.com's regular site
performance reporting.

1he major inluence on page iews per isit, at least on the BackcountryStore.com \eb site,
appears to be sale actiity. \hen sales are announced ia marketing channels such as banner
ads and the retention email program, BackcountryStore.com is able to see more page iews
per isit. In the example proided in ligure 31 there was a large, adertised sale that ran
rom the week o Noember 23 to the week o December . During this timerame you can
see that the aerage number o page iews per isit was up between 6 and 8 percent oer
non-sale periods.
Average Pages Viewed per Visitor
1he aerage number o pages iewed per isitor is an excellent indicator o the quality o it
to the audience you are trying to acquire. 1he better targeted your marketing program, the
greater the number o pages the responding isitors will likely browse. 1his metric is
reerred to as attraction` since you can loosely state that i you ind the right people they
will be more attracted to your \eb site and your content.

Not all business models are necessarily trying to get isitors to iew many pages o content.
1his metric does sere most business models well, howeer, and it should be used as a key
perormance indicator or reach and acquisition.
How to Calculate
Diide the number o page iews or a time period by the number o isitors:

PAGE VIEWS / VISITORS = AVERAGE PAGES VIEWED PER VISITOR
Dependence
1his number is bigbt, dependent on the reporting period, especially in the denominator.
lopeully your \eb analytics application is able to measure at least daily and monthly
unique isitors. Page iews are a daily statistic, so i you are looking at a month`s page iews
you are looking at the sum o page iews or each day in the month. \ou can calculate both
a daily and a monthly aerage page iews per isitor measurement or your \eb site, but
these numbers will be dierent.
1S CIupLer 1z

Usage
1he daily aerage page iews per isitor measurement can be used to determine, on a short-
term basis, how eectiely a \eb site is able to draw isitors past the home ,or landing,
page. 1he monthly calculation is a better measurement o the true aerage depth o isit or
the \eb site on a per-isitor basis. Use the monthly calculation to help you determine
whether your \eb site`s aerage isitor will hae the opportunity to iew your message with
enough requency.

Practically speaking, unless you dramatically change the content or inormation architecture
o your \eb site this number does not change much month-to-month. \ou should
incorporate this metric into your regular monthly reporting and compare the current month
to the preious month, watching or a percent change o more than 10 percent. 1he one
actiity that has the potential to change your site`s alue or aerage page iews per isitor is
a dramatic change in the audience you are trying to reach and acquire. Imagine that you
hae been adertising to a highly qualiied audience and these respondents are looking at ie
pages on aerage when they come to your \eb site. Imagine now that you begin to
adertise to a much larger but less-qualiied audience. As these isitors come to your \eb
site, assuming they are truly less qualiied, you may see your aerage page iews per isitor
metric aecrea.e. Because o this it is worthwhile to keep an eye on this metric in your regular
reporting program and, i the number decreases, always ask yoursel, Did anything change
in our marketing program recently`
Interpretation
1he ratio or this metric will always be greater than one, and the alue represents the aerage
number o pages that a isitor will iew in the timerame. ligher numbers, at least in terms
o adertising reach and requency, are better. A alue o 1` indicates that the aerage
isitor only sees one page. A alue o 10` indicates that the aerage isitor sees 10 pages.
1here is no optimal range` or aerage pages iewed per isitor, eery \eb site will hae a
dierent alue. 1he range or your particular \eb site is a unction o your inormation
architecture, your content and the audience you attract.
Related Metrics
Aerage pages iewed per isitor can also be calculated or content groups ,discussed
below,. Some \eb analytics applications proide a depth o isit` metric that reports on
the measured number o isitors iewing one page,` two pages,` three pages,` and so on.
1his metric can be used as a supplement to aerage pages iewed per isitor, i it is aailable,
to determine whether there is a sweet spot` o pages iewed. Aerage pages iewed per
isitor can also be compared to time spent on site to see how page iews relate to time spent
on a global scale.
Page Stick and Slip
One o the most important measurements in reach and acquisition is the ability o the
landing page to keep isitors` attention and draw them deeper into the \eb site. \hile
MeusurIng AcquIsILIon 1q

some authors use this term dierently, the author o !eb .vat,tic. Dev,.tifiea uses the term
stickiness` to describe a page`s ability to keep a isitor engaged. 1he stickiness` o any
landing or entry page is critical to acquisition eorts as a direct measure o the ratio o
isitors who start their isit on any gien page to the number o isitors who ovt, see that
page. Put another way, the opposite o stick,` sometimes reerred to as slip,` is the
number o isitors who see your home page,entry page but immediately back up or leae
the \eb site. lor most marketing and \eb site entry pages, page stickiness` is a critical
acquisition perormance indicator.
How to Calculate
lor any gien page under inestigation:

SINGLE ACCESS PAGE VIEWS OF A PAGE / ENTRY PAGE VIEWS OF THE SAME
PAGE = PAGE SLIP

1 (SINGLE ACCESS PAGE VIEWS OF A PAGE / ENTRY PAGE VIEWS OF THE
SAME PAGE) = PAGE STICK

1he ratio o single access page iews- the number o times the page was the only page
iewed in a isit diided by the number o times the page was the irst page in a isit-
proides us a measure o the percentage o times a isitor sees the particular page and no
other, we reer to this as page slip.` Subtracting this rom one proides us a more
optimistic key perormance indicator, the percentage o isitors who start a isit at this page
and get at least one page urther, we reer to this as a page`s ability to create stick.`
Dependence
Page stick` and slip` are dependent on being able to get a single access and entry page
measurement or indiidual pages but most known analytics packages proide both o these
metrics on a page-by-page basis, at least or the most popular pages on a site.
Usage
lrom an acquisition standpoint this metric is one o the most important perormance
indicators or the home page or any page designed to be an entry point into the site. As with
all KPIs you are watching or both lower numbers than expected- a low percentage o
isitors sticking` on the page-as well as or large swings up or down. I, aside rom your
home page, you are unsure which pages should be tracked using this metric you should
examine your entry page` report or the top three to ie most popular points o entry to
your \eb site and then monitor those pages.

\ou should always make this measurement or campaign landing pages since there is a gap
between measuring open rates, response rates and conersion rates. Remember that open
rates or impressions tell you only how many people iewed a message and that response rate
tells you how many people clicked to your \eb site. Conersion rate, which you will learn
more about in Chapter 13, tells you how many respondents completed a speciic action or
actions. Making a page stickiness` measurement or the campaign landing pages, you can
begin to estimate the percentage o isitors who are going to be aailable to conert- i you
1qo CIupLer 1z

hae 1,000 respondents and a page stickiness` o 20 percent you only hae 200 potential
conerts, not 1,000.
Interpretation
I the page stickiness` metric or your home page is 45 percent` ,0.45, this tells you that
only 45 percent o the isits to your home page during the timerame under analysis went
any urther than the home page. Since most pages are designed to drie traic to additional
pages, the ideal situation is where the stickiness` metric is relatiely high ,in the 0 to 80
percent range,. Unortunately this is rarely the case, some ery common reasons that
stickiness` is low or entry pages include:

Long load times or entry pages, especially critical or modem users
Poor it between the message that prompted the isitor to isit the page and the
message presented on the page itsel
Poor page design in general
Poor reach or marketing campaigns driing high olumes o traic

lortunately all o these issues are easily corrected. Any time you see lower than expected
stickiness,` certainly or any entry page where at least 50 percent o all isitors immediately
bail out, you should perorm additional analysis to determine the cause and implement
correctie action.

Occasionally there are pages that are designed to hae low stickiness` but most oten these
are portal-type pages designed to drie traic directly to dierent \eb sites that are outside
o the realm o the analysis, dierent domains or log-ile analysis or dierently coded pages
or tagged pages. In these situations you can perorm a similar analysis i you hae some
type o exit link` measurement aailable to you to determine whether isitors who are
leaing that page are, in act, clicking the appropriate links.
Example
1he olks at BackcountryStore.com pay close attention to the stickiness o key entry pages
on their \eb site, especially those that they drie isitors directly to ia marketing actiities.
1o this end, in their regular KPI reporting they hae built-in alerts to let them know when
pages hae slipped below acceptable thresholds ,ligure 32,.


MeusurIng AcquIsILIon 1q1


Iigure 32: Page stickiness reporting for BackcountryStore.com, part of their
regular, automated KPI report. Note that Winter Packs falls well short of a 60
percent threshold for the percent of visitors who click at least one-page further,
highlighted in bold. Also note that Women's Ileece Jackets are of some concern
with only 43.J percent of visitors going any further.

BackcountryStore.com has set one threshold at 60 percent and another at 40 percent to
indicate pages that need attention and pages that need .eriov. attention. I a page is designed
to moe isitors more deeply into the \eb site, it should be able to do so at least 60 percent
o the time. 1he 40 percent threshold is an emergency warning, indicating that perhaps
something has gone wrong with the page being monitored, as eery page on the \eb site
reasonably leads somewhere else that should be o interest to isitors.

laing regular reporting such as that shown in ligure 28 helps BackcountryStore.com
respond proactiely, rather than reactiely, towards its \eb site. Being proactie lets the
company ensure that the customer experience is always a positie one, keeping conersion
rates higher than they would be i customers were constantly experiencing errors and ending
up at dead-ends.
Related metrics
1he most closely related metrics are single access page iews and entry page iews, the
critical components o this metric. Also related are the reach and acquisition metrics or
campaign analysis as well as reerring source metrics, both o which are driing traic to
entry pages. Lisenberg and Noo in their Cviae to !eb .vat,tic. reer to this metric as the
site penetration index` when it is calculated or the home page.
Cost per Visitor
Lisenberg and Noo state that ,cost per isitor, is a ery useul way to measure what your
traic is costing you. ,A, simple calculation to get an important number that can`t be
oerlooked.` Lssentially cost per isitor can be used to determine how eectie your oerall
marketing program is at driing traic to your \eb site, a measurement o reach and
acquisition. 1he ultimate goal o online businesses, especially those selling directly ia the
1q CIupLer 1z

Internet, is to lower cost per isitor while increasing sales per isitor ,see Chapter 13 or
more,.

Gien the ability to accurately and easily determine the cost` part o this equation, cost per
isitor can be an important acquisition perormance indicator.
How to Calculate
Cost per isitor is based on being able to measure your costs or any marketing program
designed to drie traic to your \eb site.

MARKETING EXPENSES / VISITORS = COST PER VISITOR
Dependence
As you can see, this calculation is dependent on being able to easily determine your
marketing expenses or the same period o time you make the isitors measurement. Some
businesses are able to grab this number with ease, others are unable to assign a ixed dollar
alue to marketing expenses. 1he problem with this number is that A, i you are spending
any appreciable amount o money it can be diicult to know how much was actually spent in
any short period o time, such as a day, and B, oten it can take more time than is reasonable
to determine this number and so the act o building this ratio eectiely sidetracks the \eb
analytics process.

1here appears to be a direct correlation between the size o a business and the diiculty in
determining the marketing spend or a gien day, week or month. Keep this in mind when
determining whether to use this number or not, i it will take you much more than 15 or 20
minutes to get an answer to how much did we spend on marketing last |timerame|` and
you cannot automate the process, consider passing on cost per isitor and ocusing on
metrics more easily calculated.
Usage
Dependence aside, knowing your cost per isitor and watching it increase and decrease oer
time can help you understand the eect that decreasing marketing spend can hae on
acquiring isitors. Ideally, at some point, you can decrease your monthly spend on
marketing without also decreasing the number o isitors coming to your \eb site. \ou
would know that you are at this point because your cost per isitor will begin to decrease
while your aerage order size, purchase conersion rate and sales per isitor will stay roughly
the same.
Interpretation
Cost per isitor is a dollar amount, such as, >0.50 cost per isitor.` Obiously lower
amounts are better. lopeully you would neer end up in a situation where your cost per
isitor is the same as, or higher than, your sales per isitor, this situation would mean that
you lose money with eery isitor who comes to your \eb site.
MeusurIng AcquIsILIon 1q

Related metrics
Cost per isitor is closely related to sales per isitor, aerage order alue and site-wide
,purchase, conersion rates, all described in Chapter 13, Measuring Conersion.
Ratio of New Visitors to All Visitors
See number o new isitors,` aboe, or discussion o this ratio.
Heavy User Share
1aken rom Bryan Lisenberg`s Cviae to !eb .vat,tic., heay user share is the number o
isitors who iew more than n` pages in a isit diided by the total number o isits. As
long as your \eb analytics application is able to proide you the distribution o number o
pages per isit, n` in this case, you should be able to calculate this ratio.

leay user share can be a aluable perormance indicator or online businesses ocused on
haing isitors stay engaged or longer isits.
How to Calculate
Simply, or any n` pages per isit:

NUMBER OF VISITS OF N OR MORE PAGES / TOTAL VISITS = HEAVY USER
SHARE

In 1be Cviae to !eb .vat,tic. Mr. Lisenberg recommends setting n` to 11 pages to account
or isits o 10 or more page iews.
Dependence
1he only dependence is your application`s ability to report on the number o pages iewed
per isit.
Usage
Lisenberg states that heay user share is useul in determining the percentage o users who
are highly attracted to content on your \eb site and, as a metric, to help you understand the
quality o it or the audience you are currently reaching. 1he assumption is that a more
qualiied audience will iew more pages in a gien isit, thus increasing the likelihood they
will be counted as a heay user.`

Depending on how you deine n` you can tweak this ratio to sere your particular
businesses needs. I you determine that the aerage number o page iews per isit ,see
aerage page iews per isit, is ie pages per isit then you could set n` to seen or eight
to measure the percentage o isits that are 50 percent greater than your aerage number o
1qq CIupLer 1z

page iews per isit. Conersely you could hale the aerage number o page iews per isit
and calculate a light user share,` depending on your speciic business needs.
Interpretation
Depending on your \eb site`s goals you may or may not want to hae a high heay user
share. I you present content it is likely that you would preer to hae a higher heay user
share, which indicates that you are attracting isitors who are interested in your content.
Conersely i you are making this measurement or a support site you may want isitors to
be able to ind the inormation they need in as ew clicks as possible, leading to a lower
heay user share.

\our particular business model will dictate whether heay user share can be used as a
perormance indicator to help you determine i current and emerging marketing campaigns
are haing the desired eect in terms o audience reach and acquisition.
Top Pages and Content Requested By New Visitors
One o the most important data points or acquisition is the pages and content that new
isitors are iewing on your \eb site. 1his is the inormation that you proide that they
ind most compelling` and thus the area on which you may want to ocus your acquisition
eorts. Keep in mind that the new` or irst time` isitor experience may dier rom that
o a returning isitor or someone who was purchased or otherwise made a commitment to
you in the past. 1racking what early-stage prospects are doing on your site can help you to
signiicantly improe the quality o outbound marketing you do to attract other new isitors
in the uture.
How to Calculate
Since all known analytics applications report on some kind o page analysis,` and most
proide similar analysis or hierarchical content, the important component or this metric is
being able to tie page and content iews to new` isitor status. In many applications this is
accomplished using the isitor segmentation tools described in Chapter , using the
ollowing rough strategy:

1. Create a system to cookie isitors based on the number o isits they make to your
site ,see below,.
2. \hen any isitor comes to your site, irst test to see i this cookie exists. I it does
not, assume they are a irst time` isitor and segment as such.

One adantage o creating a system to cookie isitors on a per isit` basis is that you can
then likely generate a number o useul reports based on the number o lietime isits your
isitors are making.
MeusurIng AcquIsILIon 1qg

Dependence
1his metric is wholly, but only, dependent on your analytics application`s ability to
dierentiate new rom returning isitors and report on the pages and content new isitors
are most interested in. Ideally your application proides page iews, content iews,
campaign responses and reerring source ,domains, search engines and search terms, or
new and returning isitor segments.
Usage
1he most typical use or this inormation is to determine what content and inormation
presented on your \eb site is most compelling to newly acquired isitors. 1his same logic
will apply to returning isitors and isitors in other customer lie cycle categories ,conerted,
retained,, but by mining this type o inormation you can determine what products, oers
and,or inormation you proide are most likely to A, help you moe these isitors along the
customer lie cycle and B, acquire more isitors in the uture.
Interpretation
laing this inormation aailable can highlight particular pages and content that are isited
more requently by new isitors as opposed to returning isitors. By knowing what
percentage o time key inormation is iewed based on percentage o page iews or new
isitors and comparing that to the total number o page iews or the content area or all
isitors you can begin to interpret the goals o newly acquired isitors in an eort to increase
new isitor conersion.
Related metrics
Closely related to inormation preerences or new isitors are both isitor segmentation and
page,content iew metrics. Please reer to the section below on setting up a isit cookie
system or speciics on how this metric can be made i your analytics application does not
proide it automatically.
An Important Note about Setting up a Visit Cookie System
One aspect o tracking acquisition and retention is the ability to dierentiate new rom
returning isitors and to be able to see what pages, content, processes, and so on, these
dierent classes o isitors engage in. Ideally your analytics package will proide this
inormation to you without any additional setup. I it does not, but does proide some
ability to segment isitors within the reporting interace, you may want to consider setting up
a isit tracking system o your own using cookies.

All that is required to do this is the ability to set and read cookies rom a isitor`s browser
and the ability to determine the number o minutes since the last page click.` 1he basic
logic inoled is:

1. Check and see i the cookie exists already:
1q6 CIupLer 1z

2. I it does, grab the number o isits and the date,time o the last click.
3. I it does not, write the current date,time and the number 1` or number o isits
to the cookie. 1his isitor would be classiied as a irst isit` isitor.
4. I in step 1 there was a cookie, compare the current date,time with the date,time
o the last click:
5. I the number o minutes since the last click was greater than 30 minutes increase
the number o isits by 1` and write that number and the current date,time to the
cookie.
6. I the number o minutes since the last click was less than 30 minutes simply write
the current date,time to the cookie ,but do not increase the number o isits,.


// Set cutoff amounts
var durationCutoff = 23; // Set the duration high/low to 23 minutes
var visitCutoff = 3; // Set the visit high/low to 3 visits
var sessionCutoff = 30; // Set the session timeout in minutes to 30 m

function _read_cookie( begin ) {
var end=document.cookie.indexOf(";,begin);
return unescape(document.cookie.substring
(begin,end<0?document.cookie.length:end));
}
function _write_cookie(name,value,expires,path,domain,secure) {
document.cookie= name+"="+escape(value) +
((expires == null) ? "" : ("; expires=" + expires.toGMTString())) +
((path == null) ? "" : ("; path=" + path)) +
((domain == null) ? "" : ("; domain=" + domain)) +
((secure == true) ? "; secure" : "");
}
function GetCookie( name ){
var arg = name + "=";
var alen = arg.length;
var clen = document.cookie.length;
var i = 0;
while( i<clen ) {
var j = i + alen;
if(document.cookie.substring(i, j) == arg)
return _read_cookie(j);
i = document.cookie.indexOf(" , i) + 1;
if (i == 0) break;
}
return null;
}
function SetCookie(name,value,expires_in_minutes,path,domain,secure) {
var expdate = new Date ();
expdate.setTime(expdate.getTime() + (expires_in_minutes * 60 * 1000));
_write_cookie(name,value,expdate,path,domain,secure);
}

// Load the cookies with previous values if present
var oldTime = GetCookie('visit_time');
var oldVisitCount = GetCookie(' visit_count');
var oldMonth = GetCookie(' visit_month');
var oldSessionStartTime = GetCookie(' visit_ss_time');
var timeDiff = 0;
var thisCount = 1;
var thisDate = new Date();
var thisTime = thisDate.getTime();
MeusurIng AcquIsILIon 1q,

var thisMonth = thisDate.getMonth();
var thisSessionTime = oldSessionStartTime; // Don't update this unless it's a
NEW VISIT

// Test to see if this is a first monthly visit
if (oldMonth == thisMonth) {

// Set thisCount to the value in the cookie
var thisCount = oldVisitCount;

// No, they've been this month so set their QUAD level
var difference = thisTime - oldTime;
var minutes = Math.round(difference / (1000 * 60));
var session_difference = thisTime - oldSessionStartTime;
var session_minutes = Math.round(session_difference / (1000 * 60));

timeDiff = minutes;
session_timeDiff = session_minutes; // Use this to test for SESSION
duration

if (timeDiff > sessionCutoff) { // If there has been "sessionCutoff" mins of
inactivity BETWEEN CLICKS, increase visit count and reset the session time
thisCount = (oldVisitCount/1) + 1;
thisSessionTime = thisTime;

} else {

// Yes, first visit this month so reset
thisCount = 1;
thisSessionTime = thisTime;
session_timeDiff = 0;

};

// Update cookies
var expires_in_days = 30 * 24 * 60 * 1000;
SetCookie('track_time',thisTime,expires_in_days);
SetCookie('track_ss_time',thisSessionTime,expires_in_days);
SetCookie('track_count',thisCount,expires_in_days);
SetCookie('track_month',thisMonth,expires_in_days);

Iigure 33: Sample code for setting up a cookie-based visit tracking system,
written in JavaScript by the author in as inelegant a fashion as humanly possible.

\ith each test run-and you should run this test on eery page, i possible-this logic will
update the time o last click` in all cases and in some cases ,i this is a new isit, or
example, will update the total number o isits. One ariation on this is to determine
whether the isit is in a new month ,that is, the last isit was in a preious calendar month,,
in which case you can count monthly isits.

lrom this logic you can easily derie whether the isitor is a new` or returning` isitor
and also begin to classiy the isitor into isit categories, 2 to 5 preious isits,` 6 to 10
preious isits,` and so on. Using this in tandem with whateer system or isitor
segmentation your analytics application proides can be a ery powerul strategy or
understanding isitor acquisition and retention.
1qS CIupLer 1z

Content Focus
1he concept o content ocus` was irst presented by Sterne and Cutler in Metric.: v.ive..
Metric. for tbe ^er covov,, and relates the aerage number o pages isited in a content area
to the total number o pages in the section.
How to Calculate
In order to calculate content ocus you irst need to calculate the aerage number o pages
iewed per isitor or the content group o interest. Similar to the aerage number o page
iews per isit, this ratio is speciic to a single content grouping on your \eb site.

CONTENT PAGE VIEWS / CONTENT UNIQUE VISITORS = AVERAGE NUMBER OF
PAGES VIEWED PER VISITOR BY CONTENT

Once you hae made this calculation, content ocus can be ound by diiding this alue by
the total number o pages contained in the content group.

AVERAGE NUMBER OF PAGES VIEWED PER VISITOR BY CONTENT/ TOTAL NUMBER
OF CONTENT PAGES = CONTENT FOCUS
Dependence
1his metric is dependent on being able to determine the number o page iews and isitors
or an indiidual content group, numbers that are not necessarily aailable in all \eb
analytics packages. \ou should consult your endor to determine i these numbers are
aailable.

Obiously this metric is also dependent on knowing how many total pages are contained in a
content grouping, which is oten diicult to determine-especially or content-rich sites or
online businesses that sell many dierent products,
Usage
According to Sterne and Cutler, this calculation will help you determine whether the aerage
isitor has a riae or varror ocus regarding the content in question. \hile the desire to hae
a wide or narrow ocus will likely change by content presented, making this calculation or
key areas o your \eb site will allow you to understand when your audience makeup changes
signiicantly, thereby allowing you to react to those changes.
Interpretation
According to Sterne and Cutler smaller alues indicate a varror focv. while larger alues
indicate a riaer focv. on the content in question. 1hese authors cite the example o narrow
ocus being good in customer serice and support areas o a \eb site, where it is important
that isitors are able to quickly ind inormation and moe on, whereas a wider ocus is
preerable in an online auction or product catalog, where one hopes that isitors browse a
number o products, increasing their likelihood o conersion.
MeusurIng AcquIsILIon 1qq


Sterne and Cutler also relate content ocus to the stickiness` o the \eb site but it is worth
noting that their deinition o stickiness` and that used in !eb .vat,tic. Dev,.tifiea are
dierent.
Percent of Visits Under 90 Seconds
lor most \eb sites it takes isitors at least 90 seconds to become engaged in reading
content or inding inormation. Certainly there are exceptions-news and sports \eb sites
with regularly changing headlines on the home page are a good example-but or the most
part any isit lasting less than 90 seconds can be considered a isit that was probably not
long enough or a isitor to become connected with the site.

1he percent o isits under 90 seconds is an important acquisition perormance indicator.
How to Calculate
1he calculation is simply the number o isits lasting 90 seconds or less on the \eb site
diided by the total number o isits in the timerame.

VISITS UNDER 90 SECONDS / ALL VISITS = PERCENT OF VISITS UNDER 90
SECONDS
Dependence
1he only real dependence is your analytics package`s ability to report on the distribution o
isits in hal-minute increments or groupings that include 90 seconds and under. I only
minute increments are aailable it is acceptable to calculate percent o isits lasting less than
a minute.
Usage
Percent o isits under 90 seconds is an important acquisition perormance indicator helping
you to obsere gross changes in the type o audience you are bringing to your \eb site. I
you are consistently seeing this number reported in the 25 to 30 percent range and then
suddenly it spikes to 50 percent you know that either you hae just attracted a large number
o isitors who are not interested in what you hae to oer or or some reason you are no
longer attracting the isitors who had been interested in your site ,the percentage will change
due to changes in either the numerator or the denominator,.

Any time you see dramatic changes in this metric you want to immediately ask yoursel what
may hae changed in your marketing program and the type o isitors you are trying to
reach. I you start a marketing campaign that is bringing you thousands o respondents, but
your percent o isits under 90 seconds spikes dramatically, you can reasonably assume that
most o the respondents are not as interested as they thought they might be. Conersely i
your campaign is bringing you large olumes o isitors and your percent o isits under 90
1go CIupLer 1z

seconds stays relatiely stable or decreases you know you hae identiied a good adertising
opportunity.
Interpretation
1he percent o isits under 90 seconds is simply a percentage describing the total olume o
traic spending little or no time interacting with your \eb site.
Example
BackcountryStore.com pays close attention to how isitors interact with its \eb site. As
shown in ligure 29, the marketing group pays close attention to the interested` and
committed` isitors who isit the site. Because they are realists, they also pay attention to
the number o isits that are under 90 seconds in length ,ligure 34,.



Iigure 34: Percent of visits under 90 seconds included with interested and
committed visitor shares in BackcountryStore.com's regular and automated KPI
reporting.

\hile BackcountryStore.com knows that it will neer hae 0.0 percent isits under 90
seconds` it works to ensure that its \eb site loads quickly and presents compelling products
and inormation on key entry pages. 1he act that its isits under 90 seconds runs roughly
parallel to its interested isitor share means that or eery isitor it loses to disinterest it
retains one who may be interested enough to make a purchase.
Average Time Spent on Site
Aerage time spent on site is less a metric that needs to be calculated and more a metric that
needs to be watched as part o a regular reporting program. Considered here under the
umbrella o acquisition` this measurement truly spans all phases o the customer lie cycle.
In act, a strong use o the isitor segmentation tools described in Chapter would be to
segment isitors based on their phase in the customer lie cycle and obsere how the aerage
time spent on the site changes as the isitor`s progress.

How to Calculate
1he rough calculation is the total time spent suring all pages in a isit diided by the
number o isitors in the same timerame. In nearly all instances the aerage time spent on
MeusurIng AcquIsILIon 1g1

site is calculated and presented by the analytics application. Any application that does not
proide this metric should be considered suspect and replaced.
Dependence
1he only major dependence aerage time spent is known to hae is how the calculation
treats the inal page in the isit ,the exit page in some applications,. Because there is no
subsequent click` it is impossible or the analytics application to assign an amount o time
spent on the inal page in a isit. It is best i the application excludes the inal page iew in a
isit rom the calculation, i possible.

Consult your application endor to determine how they treat the inal page in a isit when it
comes to both aerage time spent on site and also time spent on pages.
Usage
Aerage time spent on site is used to watch or gross changes in the total audience isiting
the \eb site. Not limited to isitor acquisition programs, this alue can be heaily
inluenced by the type o isitors you reach and acquire ia marketing programs. I, much
like percent o isits lasting under 90 seconds, new inbound marketing actiity can be
correlated to a decrease in the aerage time spent on site, the quality o the audience being
acquired should be more closely examined or true quality o it.
Interpretation
Simply a report o the aerage number o hours, minutes and seconds all isitors spend
interacting with your \eb site.
Campaign Response Metrics
One o the most important aspects o measuring reach, acquisition and conersion on any
\eb site is the measurement o marketing actiities designed to bring isitors to the \eb
site. Regardless o whether your online strategy uses banner ads, opt-in email, non-opt-in
email ,nice word or spam`,, direct mail, pop-up ads, keyword buying, print adertising,
teleision, billboards or roing gangs o thugs wearing t-shirts bearing your logo and URL,
you should be measuring the eect this campaigning has on your online business. Not to do
so is irresponsible, period.

1o this end, hopeully the \eb analytics package you hae chosen supports the type o
robust campaign analysis tools described in Chapter 8. Key to measuring the eect online
marketing has on customer acquisition is the ability to somehow tag` media and links in
such a way that you are able to dierentiate campaign responders rom other traic.

Note that ultimately the concept o other traic` is allacious, since eeryone had to learn
about your \eb site ia some marketing eort. But it can be diicult to measure returning
1g CIupLer 1z

isitors with anything other than lietime returning isitor` metrics ,which will be described
in Chapter 14,.
Responses and Respondents
1he second most important metric when measuring the eectieness o any marketing
campaign is the response rate or that campaign. \hen online businesses irst start to
attempt to quantiy the eect o marketing, response rate is one o the irst metrics
examined and tracked. \hile this is tremendously important, it is perhaps more important
to understand what these respondents are doing on the \eb site ,see campaign discussion in
Chapter 13,.

How to Calculate
Any \eb analytics application that supports campaign analysis o some kind should be able
to proide the number o responses and respondents to a gien campaign. Remember rom
the discussion in Chapter 8, respondents are real people who perorm some actiity you
hae designed to bring them to your \eb site ,click a link, type a URL, and so on,. \hen
these people perorm this actiity it is said to be a campaign response` and the percentage
o campaign responses diided by the total impressions is the response rate.

RESPONSES / TOTAL IMPRESSIONS = RESPONSE RATE

\ou should calculate the response rate or any actie campaign on a daily basis.
Dependence
1he response rate calculation depends on you being able to accurately track or know the
number o impressions sered and the number o responses or a gien campaign. I you
are unable to measure the number o impressions sered or a campaign but you hae a
rough idea o how many impressions were deliered ,you sent an email to 100,000 addresses
but hae no idea what the open rate was, you can calculate response rate based on this
estimate but keep the act that you are working rom an estimate in mind when you are
tempted to use the resulting data.
Usage
Response rate is critical to the continuous improement process when it comes to
acquisition and conersion. Presumably you hae some experience with online marketing
and hae discoered independently that not all campaigns are created equally. I open rate is
an indicator o how compelling your subject line is or an email, response rate is a measure
o how compelling the oerall message is once it has been iewed. Response rate is critical
in helping you diagnose issues o message and packaging, as well as the quality o reach
through a gien deliery mechanism.

MeusurIng AcquIsILIon 1g

\hile response rate is not the most important metric o campaign success it is an important
indicator o whether a campaign will be successul. 1he more respondents you are able to
acquire the greater the likelihood you will ind qualiied respondents who will potentially
conert.
Interpretation
Response rate is simply a percentage that tells you, \hat percentage o all impressions
sered or this campaign turned into isits to my \eb site` ligher percentages can tell you
either that the message was well-receied by the audience you were attempting to reach ,a
strong it between audience and message, or that the message was so compelling that people
were responding regardless o the quality o it between audience and message. Lower
percentages are, o course, an indicator that there is some disconnect between the message
and the audience.
Example
1he author o !eb .vat,tic. Dev,.tifiea was engaged in a handul o marketing campaigns to
drie traic to the companion \eb site or this book throughout the last hal o 2003.
1hese campaigns included email, keyword buying through Google.com and a handul o
anity URL landing pages ,ligure 35,.



Iigure 3S: Sample data from the author's Web site showing active campaigns and
number of responses, broken down by new respondents and repeat responses.

Depending on the campaign type, the author calculated response rates dierently. lor the
KNL-Google` campaign, impression data came rom the Google Ad\ords application
interace ,ligure 36,. lor email programs, response rate was generated based on the total
number o emails sent. lor anity URL landing pages, rates were generated using page
iews.


1gq CIupLer 1z


Iigure 36: Response rates (called click-through-rate in Google-ese) for a handful of
keywords that the author purchased in the last half of 2003 (data provided by Google
AdWords).

O each o the dierent kinds o campaigns that the author was running at the time, direct
email campaigning had the best response rates, aeraging more than 20 percent. 1his was
ollowed by anity URLs, aeraging around to 9 percent and then keyword buying,
aeraging around 0.5 percent. In the context o the author`s marketing program or driing
people to download a preiew copy o !eb .vat,tic. Dev,.tifiea this made perect sense. 1he
people whom the author had preiously had the most contact ,through email, had the
highest conersion rate, ollowed by people with whom he`d had peripheral contact ,anity
URLs were being ed by others in the \eb analytics space such as Jim Sterne and the
Stanord Publishing group,, ollowed by complete strangers who happened to be interested
in the topic o \eb analytics.`
Cost-per-Acquisition / Cost-per-Click
Cost-per-acquisition is one o the irst metrics discussed in !eb .vat,tic. Dev,.tifiea that tie
\eb analytic data to return on inestment. One o the reasons this book was written is to
help marketing managers understand how spending money online translates into either
making money back-or at least not throwing money away. Cost-per-acquisition is critical
to help ensure you are not paying too much or leads or prospectie customers.

How to Calculate
Cost-per-acquisition is the dollar amount spent on the campaign diided by number o
unique prospects acquired.

COST OF CAMPAIGN / CAMPAIGN RESPONDENTS = COST PER ACQUISITION

It is important to calculate this metric or a speciic timerame, or the day, the week, the lie
o the campaign, and to use campaign costs that are as accurate as possible. It is also
important to use re.ovaevt. not re.ov.e. or this calculation since you are working to
MeusurIng AcquIsILIon 1gg

determine not the simple cost-per-click but rather the cost o acquiring people that you can
urther market to and that, hopeully, will ultimately conert.

I you wanted to calculate the cost-per-click you would simply make the same calculation
using the number o campaign re.ov.e.:

COST OF CAMPAIGN / CAMPAIGN RESPONSES = COST PER CLICK
Dependence
Depending on the type o campaign, costs can be more or less diicult to determine. In
general, or campaigns that hae a ixed deliery cost ,CPM banner adertising, some
partnerships, you can simply diide the total cost by the number o days under examination
to get the cost o campaign or the calculation. Variable deliery cost campaigns ,cost-per-
click, pay-per-click, are slightly trickier in that you need to sum the cost or each day in the
timerame under examination.

1his calculation also depends on haing tools to allow you to measure responses and
respondents to a gien campaign.
Usage
Cost-per-click and cost-per-acquisition are major actors in helping the online business
understand which campaigns are cost eectie or the organization. \hile not the inal
word in whether a campaign is a success or not, used in combination with response rate,
conersion rate and cost per conersion, these metrics will help smart business managers
determine which campaigns to keep and which to drop.

Ideally you will be able to hae or create a iew o these important campaign metrics-open
rate, response rate, cost-per-acquisition, conersion rate, cost-per-conersion and campaign
return on inestment-that allow or the easy comparison o actie campaigns.
Interpretation
Cost-per-click and cost-per-acquisition are monetary metrics describing the indiidual cost
or each. A cost-per-click o >2.50 means that or each response to the campaign in
question you are paying >2.50. A cost-per-acquisition o >0.49 means that each person you
acquire ia this campaign costs you >0.49. 1aken alone these numbers are less interesting
than when you compare them to the predicted alue o the campaign oerall. I you are
paying >2.50 or each response to the campaign but the ultimate goal is to sell a product or
which you only make >25.00 per sale, the campaign will need to conert at 10 percent just to
break een:

100 responses costing >2.50 each ~ >250.00
250.00 diided by >25.00 order alue ~ 10 conersions to break een
10 conersions diided by 100 responses ~ 10 percent conersion rate to break
een

1g6 CIupLer 1z

\ith all o the aboe, this campaign will not be proitable at a two-time or three-time
expenditure target unless the campaign conerts at 20 or 30 percent ,which in most instances
is unlikely,.

1he best way to back into the answer to \hat is a good target or cost per acquisition` is
to do the ollowing:

1. ligure out the total amount you`d like to make rom the campaign ,be realistic,
2. Make sure you know your aerage order alue ,AOV, or aerage alue per
conersion
3. ligure out what multiple you`d like to enorce or your campaigns
4. Diide the total amount you`d like to make ,step 1, by your target multiple , step 3,
to get your maximum campaign spend
5. ligure out your campaign conersion rate, either rom an aerage or rom preious
similar campaigns
6. Diide your earning target or the campaign by your AOV to get the number o
conersions required to make your earning target
. Multiple the number o conersions by your conersion rate to determine how
many respondents you need
8. Diide the maximum campaign spend ,step 4, by the number o respondents
required ,step , to get the maximum spend per respondent ,cost-per-acquisition,

As an example:

1. \e`d like to make >10,000 rom the campaign
2. Our AOV is >100.00
3. \e like our campaigns to produce at a 5x multiple
4. \e`d like to spend no more than >2,000 on the campaign
5. Our conersion rate is 5 percent
6. \e need 100 conersions to make >10,000
. \ith a conersion rate o 5 percent we need 2,000 respondents
8. \ith a target spend o >2,000 and needing 2,000 respondents, we can spend no
more than >1.00 per acquisition on the campaign

In the example aboe, i your conersion rate were double ,10 percent, you could also aord
to spend twice as much ,>2.00 per acquisition, on the campaign. \ith this inormation in
mind you can use your response rate to determine whether you can actually aord to run a
cost-per-click or cost-per-acquisition campaign or not.
Example
lor an example o cost-per-acquisition in action, take another look at the author`s marketing
campaign strategy described earlier in this chapter. Part o this strategy was buying releant
keywords through Google`s Ad\ords program where the author was paying or each click
to www.webanalyticsdemystiied.com rom Google ,ligure 3,.


MeusurIng AcquIsILIon 1g,


Iigure 37: Cost-per-acquisition (cost-per-conversion in Google-ese, which is
technically correct, although the author treated conversion differently in this
campaign) for a handful of keywords that the author was purchasing in the last half
of 2003 (data provided by Google AdWords).

As you can see in ligure 3, the author was paying between >0.19 and >0.44 per isitor
acquired, with the act o acquisition being a lead generated when the respondent submitted
their email address to get a copy o !eb .vat,tic. Dev,.tifiea. Based on the calculations
presented aboe the author was losing money, as there was no money made on the ree`
download. 1he alue o each lead, howeer, was predicted to be somewhere between >5.00
and >9.00, depending on how well this book ultimately sells online. \ith this in mind, the
Google program was working quite well or the author ,perhaps you een ound this book
ia a paid placement on Google or a related \eb site,.
Referring Sources and Marketing Mix
Much like campaigns and campaign analysis, understanding which reerring sources are
sending you traic is critical to deeloping a clear picture o how you acquire isitors with an
eye towards conerting them. Keeping track o which reerring domains, URLs, search
engines and search keywords drie the greatest olume o traic to your site is an important
irst step in being able to deelop new online marketing programs and partnerships. Unlike
most campaigns, reerring traic rom domains is oten ree and simply a unction o how
the Internet is designed to work-\eb sites link to other \eb sites.

\hen you are considering the source o traic there are two major incoming sources-sites
that somehow link to your \eb site and people who type your URL directly into a browser.
\ithin each o these there are subtypes. lor sites linked to yours these subtypes are search
engines and those sites that are not search engines ,such as your partners and the online
media,. lor people who type your URL directly into a browser there are those who enter
the URL each time and those who hae your site bookmarked` ,listed as a aorite site`,.

1he relationship between directly reerred traic,` search engines` and normally reerred
traic is a unction o your oerall marketing program ,your marketing mix`,. 1he more
oline marketing you do that prominently displays your URL the greater the eect on your
directly reerred traic.` 1he more energy and money you spend on search engine
1gS CIupLer 1z

optimization and placement programs the greater the eect on your search engine`
reerrals. 1he greater the eort you put into online marketing and partnerships the greater
the eect on normally reerred domains,URLs. 1he particular blend o these three areas
that is right or your business depends entirely on your brand, your ocus and the amount o
money you hae to spend. Most small businesses ind that it is easiest and least expensie to
ocus on the latter as, comparatiely speaking, online marketing is much less expensie than
oline marketing. Most truly large businesses ,think lortune 500, ind it easiest to ocus on
the ormer, haing built their businesses on old world` adertising and marketing.
Leryone struggles, more or less, with search engine optimization, although the recent
addition o cost-per-click,pay-per-click and paid placement programs rom the major search
engines hae allowed businesses to enter into search results in a way that make it easy to
calculate return on inestment.

Regardless o your particular marketing mix there is always room or improement and
optimization. Understanding the relationship between traic drien by marketing campaigns
and otherwise reerred traic is the irst step in the optimization process. Making sure you
understand the concepts presented in !eb .vat,tic. Dev,.tifiea is critical to optimizing you
marketing mix because, as the old adage states, you cannot change what you do not measure
and you should not measure what you will not change.
Referring Domains
1he reerring domains discussed in this section all into the broad category o those \eb
sites that are not search engines that link directly to your \eb site and thereby drie traic
to you. 1here are many reasons these links hae been created, including:

1he reerrer is a partner o yours and you hae worked directly with them to create
the link.
1he reerrer is a list o some kind and they`e included you in the list.
1he reerrer is a content site o some kind and they hae written about you.
1he reerrer is unhappy and they hae included a link to create headaches or you.

In actuality the list o reasons, both good and bad, that a \eb site may link to yours is
endless. 1he adantage o closely watching reerring domains as they help you acquire
isitors is that it may allow you to capture an opportunity-or head o a liability.

Reerring domains, at least the top three to ie, are an excellent addition to an acquisition
KPI report.
How to Calculate
All known \eb analytics applications hae some type o reerring domain reporting included
and typically report either the number o isits or isitors rom the reerring domain. Please
consult your application`s documentation or speciics on how to generate this report.
MeusurIng AcquIsILIon 1gq

Dependence
1he only dependence or an application to report on reerring domains is the presence o a
reerring URL in the header o the document request. In most instances where there is no
reerring URL in the header o the request, the request has been made directly ,a isitor
typed in the URL or used a browser bookmark,. Occasionally this is not the case-some
orwarding applications and methods or redirection strip the original header when the URL
request is passed along. 1he most common culprit is an l11P 3xx redirect request` ,see
http:,,www.w3.org,Protocols,rc2616,rc2616-sec10.html, or a custom, serer-side
redirect script that, rather than passing along the original reerrer now includes the redirect
script as the reerring source.

In instances where the reerrer is incorrectly remoed the only recourse ,known to this
author, is to attempt to either insert the original reerring URL back into the l11P request
or to pass the URL along to the application ater the redirect. lailure to do so will result in
skewed data regarding reerring domains and URLs, sometimes dramatically so, oten
completely inalidating the inormation contained in a reerring domains or reerring URLs
report.
Usage
1he primary use or a reerring domains report is to ensure that you know rb, another \eb
site is sending you traic. Any time a \eb site sends you traic you hae the opportunity to
establish a relationship with the isitors as they arrie. 1echnically you are already acquiring
them but, as you hae already read, simply acquiring these isitors is not enough. I you
understand what it was that brought the isitor to you, you can work to continue to connect
with the isitor along those lines, and perhaps drie the isitor to conert.

Most \eb analytics packages will report two related metrics when discussing reerring
domains-reerring URL and entry page. 1he reerring URLs are the exact pages on the
reerring \eb site that contained links to your site, the entry pages are the pages on your
\eb site that the reerring URL sent traic to. 1he general strategy is to examine the
reerring URLs to understand what the traic partner is saying about your \eb site that
would compel a isitor to click the link to your site, and then make sure that the content on
the associated entry pages is consistent with what the partner is saying. 1he worst-case
scenario is where you hae worked with an online partner to crat a message that dries a
large olume o isitors to your \eb site but you send those isitors to a page that has
nothing to do with the message. Situations like this create conusion in the minds o online
isitors and decrease the likelihood that those isitors will conert.

Ideally you are always working to create a message in your marketing materials and program
that is consistent when isitors arrie at your \eb site. Regardless o whether it is email,
banner ads, text in cost-per-click,pay-per-click adertising, language on partner sites, listings
in search placements, and so on. that you are using to reach people, acquisition only leads to
conersion when the message is consistent ,or at least not ivcov.i.tevt, with the inormation
you present when the isitor arries at your \eb site.
16o CIupLer 1z

Interpretation
Most analytics packages report on the reerring domain and the number o either isits or
isitors that come rom each. 1ypically a reerring domain report will include line items or
Directly Reerred 1raic` and Search Lngines` and also perhaps Internal 1raic`
depending on the exact package you use. In most cases, directly reerred traic` reers to
isits or isitors arriing rom people typing your URL into a browser directly or using a
bookmarked link in their \eb browser. Oten this directly reerred traic` can be thought
o as a measure o the oerall strength o your brand and your oline brand awareness
marketing.

It is worthwhile to watch the traic rom reerring domains or large spikes or dips in the
olume o traic your top domains are sending. A large increase can indicate a change in
placement or message, a large decrease can mean the same. Lither way, any changes should
be inestigated by drilling down into the associated reerring URLs and entry pages these
URLs are driing traic to.
Example
BackcountryStore.com pays close attention to its incoming traic to keep track o the eect
o its search engine optimization ,SLO, programs, brand awareness and the business
relationships it has established. 1o this end, they watch a Change` report that is
proided by their \eb analytics application requently, watching or sites that exhibit a high
degree o moement, either up or down ,ligure 38,.



Iigure 38: Domains referring traffic to BackcountryStore.com and the percent
change from comparable timeframes. WebSideStory, BackcountryStore.com's Web
analytics vendor, refers to visitors who have typed the URL directly into the browser
as having come from Bookmarks or directly referred URLs.

MeusurIng AcquIsILIon 161

In the report shown in ligure 38 you can see traic olumes decreasing quite sharply rom
three domains ,winternext.com,` tgoemall.com` and hot-deals-online.com`, and
increasing dramatically rom store4outdoors.com.` \hile the marketing group at
BackcountryStore.com would most certainly look at what may hae changed with these
domains, they will also work to understand why traic rom search engines and isitors who
know their URL are down 9.9 percent and 6.1 percent respectiely. Lspecially or the search
engine traic, BackcountryStore.com depends on isitors being able to ind them online
when they are looking or outdoor gear.
Related Metrics
As mentioned, reerring URLs, search engines and entry pages are all closely tied to reerring
domains, being either a peer data type ,as in the case o search engines, or an associated data
type useul or drilling down ,as in the case o reerring URLs and entry pages,.
Additionally, reerring domains oten include those domains that you are running online
marketing on and so many o the campaign analysis metrics are closely related as well.
Referring URLs
Reerring URLs are similar to reerring domains in that reerring URLs are the source o
data rom which reerring domains are deried. Reerring domains are typically presented
something like ligure 39:



Iigure 39: Sample of traffic from major search engines to BackcountryStore.com.

lor any line item, these entries are usually deried rom something like ligure 40:



Iigure 40: Lxample referring URLs that are the basis for the referring search engines
shown in Iigure 39.

16 CIupLer 1z

As you can see, the reerring domain is simpliied rom the original reerring URL and
aggregated. In some instances you will need to get back to the reerring URL, most oten to
see the message or link that droe traic to your \eb site as described under reerring
domains. A reerring URL report is the most common way to access this inormation.
How to Calculate
All known \eb analytics applications hae some type o reerring domain URLs included
and typically report either the number o isits or isitors rom the reerring URL, please
consult your application`s documentation or speciics on how to generate this report.
Dependence
Reerring URLs hae the same dependence as reerring domains. Please see the discussion
o dependence` under reerring domains.
Usage
1ypically when aced with the choice between examining reerring domains and reerring
URLs you should start with reerring domains. In many examples, especially search engines,
there are an ininite number o possible reerring URLs or any single reerring domain.
Rather than sit and sort through hundreds o indiidual reerring URLs it is better to
examine the inormation at the highest leel ,the domain, and dele deeper i necessary.

As described under reerring domains, the primary use or a reerring URL report is to
determine exactly what marketing message isitors were responding to when they decided to
click to your \eb site. \ou would want to drill-down on this to ensure that the message is
consistent with the inormation you proide when a isitor actually arries at the site.
Inconsistency in the message rom the click to the isit is one o the places where
abandonment most requently occurs. \ou get the nope, that`s not it.` syndrome, where
a isitor clicks to your \eb site but does not immediately see what they are looking or and
so they click back and keep searching.
Interpretation
Similar to reerring domains, reerring URLs are reported in isits or isitors. 1his metric
tells you the exact number o isits or isitors coming to your site rom each indiidual URL.
Reerring URLs are usually used as a diagnostic tool rather than a regularly obsered
perormance indicator.
Navigation Path by Referring Source
Naigation path ,or click stream`, by reerring source is a metric oered by some analytics
packages and is oten reerred to in a ariety o dierent ways. 1he gist o paths rom
reerrer` is that you hae some way to search the exhaustie set o naigational paths
ollowed by each isitor in each session and relate those paths to the original reerring source
MeusurIng AcquIsILIon 16

,be it a domain or search engine,. 1he alue o doing so is that you can, without any type o
campaign identiication, begin to quantiy the alue o a particular reerring source.
How to Calculate
Naigation path by reerring source is a metric that some \eb analytics packages hae and
others simply do not and those that do hae wide ariations in their sophistication. Ideally
your application proides a simple interace to search or speciy which pages you are
interested in and which reerrers you`d like to examine. I your particular application does
not support this unctionality but does support campaign analysis o some kind you should
be ine. I your application supports neither you may want to consider switching
applications.
Dependence
Aside rom a dependence on the application supporting this unctionality the only major
dependence is on the completeness` o the dataset supporting the analysis. I the
application you use does not keep an exhaustie list o all naigation paths traersed by
isitors you run the risk o making decisions based on incomplete data. 1here are many
reasons that this dataset may be incomplete-log iles may hae been rotated out or you may
be missing data rom serers in a clustered enironment, your ASP may not keep all paths,`
the data may be too large to analyze completely, and so on.

Beore undergoing an analysis using this type o tool it is a good idea to make sure you know
how complete the supporting data model is so you can estimate the accuracy i you are
working rom a sample.
Usage
Proiding you hae access to a complete set o data you can begin to examine which
reerring sources drie traic to key pages on your \eb site. As an example, you could
search a report like this or all reerred` naigational paths that traerse a key page, perhaps
the purchase conirmation` page on a commerce site. 1his would allow you to quickly
determine which reerring partners are sending you the most qualiied traic in the absence
o commerce analysis tools like those discussed in Chapter 9.

\ou can also search these paths or a particular reerring source and work to understand the
most common paths and goals or these reerred isitors, proiding you the opportunity to
look or trends and patterns that may inorm uture marketing actiities with that partner or
site.
Interpretation
Most oten naigation path by reerring source is reported in isits and aggregates similar
paths. So i 30 people came rom yahoo.com,` landed on page A` then clicked to page
B` and page C` you might see something like ligure 41:

16q CIupLer 1z



Iigure 4J: Lxample of navigation path from referring source from the
BackcountryStore.com Web site.

A common problem with any naigational path or click-stream` reporting is isualization
and aggregation. 1he most common paths to multiple isitors are ery short, say three-page
paths. \hen paths get longer it is much more likely that these paths will begin to become
unique. \hile it is reasonable to mine click-stream data or short paths and path ragments
it is oten unclear how much alue there is in knowing about detailed actiity or a small
number o isitors. In most applications the response to this issue is to proide a process
analysis` toolset similar or same as that described in Chapter 6.
Search Engines
Search engines are another large category o reerred traic ,not directly reerred`, that is
coming speciically rom those \eb sites commonly reerred to as search engines.` \hile
this list is potentially ery, ery large, it commonly includes the major search engines such as
Google, MSN Search, \ahoo! Search, AOL, Allthe\eb, Ask Jeees, lotBot and Lycos ,a
more complete list can be ound at SearchLngine\atch.com,. Companies commonly spend
large sums o money in an attempt to increase and improe their ranking in these search
engines, oten reerred to as search engine optimization` or SLO,` and the desire to hae
optimized placement in search engines has eoled pay-per-click and pay-or-placement
options rom the major engines. Clearly traic rom search engines is tremendously
important to online marketing and the Internet in general.

Because o the adent o pay-per-click and pay-or-placement options rom the major search
engines it is important to consider the mix o traic when you examine the traic rom a
gien search engine. \ears ago we considered all traic rom search engines to be organic`
and resulting rom normal indexing as opposed to be rom a paid placement. Organic
placement was optimized by either making good \eb pages that were well indexed by the
search engine robots` or by somehow tricking the search engine robots` into thinking that
your pages were good,releant,and so on. Now, nearly all o the popular search engines
oer some type o paid placement, either pay-per-click or pay-or-placement. 1he issue is
that traic rom each type o program ,organic or paid, is grouped together in a search
engines reerral report.

Because, hopeully, the \eb analytics application you are using supports the types o
campaign analysis tools described in Chapter 8, you should deinitely be tagging or
identiying any paid search engine placement to help you isolate paid placement rom
organic. Doing so will allow you to determine how well indexed you are normally in the
search engines. Knowing this helps you understand how much traic you can likely expect
rom this piece o your marketing mix.`
MeusurIng AcquIsILIon 16g

How to Calculate
Most known \eb analytics packages automatically report on traic rom search engines by
engine based on isits or isitors.
Dependence
Search engines hae the same dependence on data described under reerring domains.
Additionally, the indiidual engines that are included in a search engines` traic report
must be determined and maintained somehow. An important question to ask your analytics
proider is how the list o known search engines is generated and who maintains this list.
Usage
1he major reason or understanding search engine traic is to ensure that any money you
spend on search engine optimization and paid inclusion is paying o. \hile a search
engines report won`t tell you what isitors are doing when they land on your \eb site it will
help you understand the total olume o traic you get rom each engine. Knowing that
conersion is a unction o the pool o isitors you hae to potentially conert, monitoring
the olume o traic is important.
Interpretation
Similar to reerring domains, watching traic olumes rom search engines is important to
ensure that you are getting isitors rom these important sources o traic and to make sure
that you understand why any large spikes or dips are occurring. Any time you spend time,
money or energy to optimize your site`s placement in a search engine ,or all search engines,
you hope and expect to see an increase in the olume o traic coming rom this source.
Similarly, i you are seeing a decrease, either sudden or gradual, in the olume o traic rom
any engine that had been sending you traic preiously, it is worthwhile to explore the
reason or this decrease and the potential eect this will hae on your online business.
Example
As shown in ligure 38, BackcountryStore.com pays close attention to the olume o traic
reerred to its \eb site rom search engines and engages actiely in search engine
optimization ,SLO, actiity. Regularly, the marketing sta at BackcountryStore.com will use
a report similar to that shown in ligure 42 to make sure that they know how changes in
traic olumes rom the major search engines aect their reenue stream.


166 CIupLer 1z


Iigure 42: 1raffic to BackcountryStore.com from some of the major search engines
showing the Change in visits from each.

\ithout question, the report shown in ligure 43 proides much more important
inormation to BackcountryStore.com about the traic they get rom search engines.



Iigure 43: Search engine traffic to BackcountryStore.com that drives visitors to
purchase (data provided by HitBox Commerce). 1he B/B column is the
browse-to-buy ratio, an indicator of how likely visitors from each search engine are
to purchase. 1he Revenue and Visits columns have been blurred to protect
BackcountryStore.com's privacy.

By closely watching both the traic olume and reenue drien by search engines,
BackcountryStore.com is able to make the best use o search engine optimization and
understand which actiities work well and which do not.
Search Keywords and Phrases
1he traic that comes rom each o the search engines carries important inormation about
the intentions o the isitor you are hopeully about to acquire-the word or phrase that the
isitor searched or just beore they clicked to your \eb site. Carried in ia the reerring
URL, this text proides a great deal o insight into the inormation that the isitor hopes to
ind on your site. Ideally the page that the isitor lands on has inormation releant to the
word or phrase they used to ind your site, i not you run the risk that the isitor will
immediately back up to the search engine and ind what they are looking or somewhere else.
MeusurIng AcquIsILIon 16,


Search keywords and phrases, at least the top ie to ten, are an excellent addition to an
acquisition KPI report.
How to Calculate
Most known \eb analytics packages report on search keywords and phrases by isit or
isitor. Ideally your analytics package will report on the entry pages to which each search
engine is sending search keyword traic, to proide insight into the quality o it between the
search and the resulting page.
Dependence
Search keywords hae the same dependence on data described under search engines and
reerring domains. Additionally, the indiidual engines that are included in a search
engines` traic report must be determined and maintained somehow in order to determine
which URLs will contain keywords and phrases. An important question to ask your analytics
proider is how the list o known search engines is generated and who maintains this list.
Usage
Search keywords and phrases are useul in that they are a direct relection o what your
isitors are thinking about or looking or when they click to your \eb site. Depending on
how your site is indexed in the major search engines and whether you are paying or
placement ia a pay-per-click program your site will appear at some location in each o the
search engines. I you appear high enough in these listings you will ineitably acquire
isitors rom search engine listings. In an ideal world people would be searching or words
and phrases that are consistent with your marketing messaging and goals, unortunately this
is not always the case. Oten when you are reiewing a report on the top search terms and
phrases driing traic to your \eb site you will likely think, Now why did someone search
or that!`

\ords and phrases that will appear in this list will typically all into three categories: your
brand name and URL, words you expect and words you do not expect. People search or
your brand name and \eb site URL or mysterious reasons-perhaps they are lazy and they
hae a search engine or a home page ,Google, MSN, \ahoo!, so it is easier to type in your
name or URL and hit Search` than it is to type your URL into a browser directly. \ithout
extensie ocus groups it is likely that we will neer know why isitors do this so you should
simply know that this happens.

1he second category, words and phrases that you do expect, typically includes the names o
your products or brands that you sell, the kinds o serices you proide, the type o
inormation you oer, and so on. 1hese are the words that you sincerely hope will help you
acquire isitors because it means that your marketing programs are working and that when
people think about X` they are willing to consider you as a proider o X.` 1oo ew or
no isits rom these words and phrases indicates that your site is poorly indexed and you
should consider some kind o search engine optimization. Search engines hae become a
driing orce in Internet marketing oer the last ew years and not being indexed or listed in
16S CIupLer 1z

search engines will put you at a seere disadantage ,considering that your competition is
likely listed,.

1he third category, words and phrases you do not expect, are a source o opportunity and
insight into how your isitors think. 1his group o terms can proide some insight into how
you tell people about your company and oerings and how potential customers receie that
message. Mining this list can help you to reine your message and uture marketing materials
to capitalize on emerging trends or changes in the competitie landscape.
Interpretation
Keyword and phrases reporting is typically done on a per isit` or per isitor` basis, much
like each o the other reerring source reports. An important actor to keep in mind when
interpreting this report is that i you are making keyword purchases through one o the
popular search marketing programs ,Google, Oerture, most likely your search results
contain botb the words and phrases you are purchasing and those that you simply get
organically` ,without purchase,. In most instances this is a good thing, allowing you to
quickly see which words and phrases are driing the greatest olume o traic to your \eb
site regardless o how those words show up in the search engines. I you do hae mixed
paid,organic search results it is a good idea to track the keywords and phrases you are
purchasing using campaign analysis tools ,see Campaign Response Metrics` in this chapter
and Campaign Conersion` in Chapter 13,.

Similar to other reerring source metrics the important thing to watch or are dramatic
changes in the olume o traic being drien by search terms. A dramatic upward
moement or a word or group o words can mean that you hae been re-indexed by one or
more search engines and that your new placement is more aorable. It can also mean that
someone in your marketing group has started aggressiely purchasing those words and
phrases. 1he irst example is good news since there is little or no cost with most search
engine optimization ,when done properly,. 1he second example is potentially good news,
proiding isitors responding to those keyword purchases are doing something on your \eb
site that is making you more money than was spent on the keyword buys.

One strategy or interpretation is to take adantage o any search tools that your analytics
package may hae to group similar terms and phrases. Doing so can help you isualize any
trends that may be emerging.
Example
In early 2002 the nice olks at BackcountryStore.com had noticed that a number o searches
that were bringing isitors to their \eb site implied that isitors were looking or
paddleboats, kayaks and related gear. \hile they had some small items o this type aailable
in their product selection they had not really considered stocking kayaks and paddle boats
since they were A, expensie to purchase and ship and B, space-intensie to keep in stock.
Still, there appeared to be a trend and so being the smart analytics users they are they started
to monitor their internally executed searches or a similar trend.


MeusurIng AcquIsILIon 16q


Iigure 44: Search terms from BackcountryStore.com's internal search that suggested
that visitors were interested in buying paddle boats and gear from them online.

\hat they ound was that signiicant numbers o isitors were in act looking or these types
o items ,ligure 44,, large enough numbers that a relatiely modest conersion rate would
yield a airly proitable business selling paddle boats and gear through
BackcountryStore.com. Ater working out the details, BackcountryStore.com was able to
stock paddle boats and currently enjoys a healthy reenue stream rom this line o business
,ligure 45,.



Iigure 4S: BackcountryStore.com's paddleboat category, prompted by referring
search keyword and internal search activity.
1,o CIupLer 1z


Related Metrics
Search keywords and phrases are closely related to both search engines and reerring
domains,URLs in that they all share the same primary source o inormation ,the reerring
URL,. 1his metric should also be compared to campaign analysis metrics o response and
conersion i you are purchasing cost-per-click,pay-per-click search results.
Notes on Search Engine Marketing and Search
Engine Optimization
Search engine optimization and search engine marketing are big business, especially lately.
At the time this book was being written, Jupiter Research had pegged the paid search market
at >1.6 billion in 2003 ater haing grown almost 50 percent rom 2002
,cyberatlas.internet.com,. 1hey also predict that online spending or search engine
marketing will grow 20 percent year-oer-year until 2008, at which point it will be a >4.3
billion dollar business accounting or roughly 30 percent o all online adertising. 1his,
coupled with \ahoo!`s purchase o Oerture ,www.oerture.com,-the organization that
pioneered pay-or-perormance search marketing in the late 1990s as Go1o.com-or more
than >1.6 billion ,ironically the same amount as Jupiter estimates or the entire market,
www.seoconsultants.com,, points to a long and ery lucratie lie or companies that proide
pay-or-perormance search engine marketing.

\hy

1he answer, at least anecdotally, is that search engine marketing works. It is easy to buy,
easy to understand and, thanks to \eb analytics applications that support campaign analysis,
easy to measure return on inestment. Go ahead and Google the phrase case study on
search engine marketing` and you will get pages o results discussing why search engine
optimization and search marketing is an eectie tool or growing an online business. Or,
take a look at what the olks at Marketing Sherpa hae to say on the subject
,http:,,library.marketingsherpa.com,. Lssentially what you will ind is that the pay-or-
perormance model o search engine marketing is an extremely aluable tool or connecting
people`s needs with the companies that can help them-it`s that simple.

Moreoer, because you can easily measure the response and conersion rate or any word or
phrase you bid on and purchase using most popular \eb analytics applications you can
essentially remoe all o the guesswork rom this type o online marketing, words and
phrases either make you more money than they cost you or you simply bid less. 1he olks at
litBox hae nearly completely automated this system. \ou simply identiy the traic as
coming rom a keyword network` and the application does the rest, measuring the
response and conersion rate or all purchased keywords automatically. 1he inal step in this
process is to tie conersion inormation to a commerce measurement package to allow
online marketing managers to know exactly how much reenue each indiidual term dries
and the distribution o products actually sold ,to calculate margins on product sales,.
MeusurIng AcquIsILIon 1,1


So assuming you hae tools to measure which words bring you isitors who ultimately
proide some kind o return on inestment, why would you not inest in keywords at
Google, Oerture or a similar program 1he unortunate answer may ultimately be that you
cannot aord to bid on words that bring you enough traic to be worth your while.
Consider the calculation outlined under cost-per-acquisition , cost-per-click` in this
chapter. I you run numbers through the equations described in that metric you will learn
that there is an upper limit on the amount you can aord to pay or each click. As search
engine marketing continues to become more popular the likelihood is that many o the terms
releant to your online business, especially generic terms, will simply be priced out o your
reach by companies that are willing to pay more. Much like in the time o banner
adertising, when some deep pocket` companies droe the cost-per-thousand ,CPM, up on
popular destination and portal sites, the same is likely to happen on search engines as we
approach the saturation point.

1he only recommendation the author has on this point is to get in now and start iguring out
which words work or you and which ones do not. Capture as much market share rom
those terms as possible, conert those isitors and retain them as best you can. 1hen, as
terms are slowly priced out o your reach, use your analytics tools to explore other, less
expensie terms and ocus your energies there. By continually watching the acquisition,
conersion and retention metrics or search engine keywords you can easily determine where
your money is best spent ,essentially, to quote the great bard Kenny Rogers, know when to
hold, know when to old em .`,
Key Performance Indicators Recommended for
Measuring Acquisition
\hile the majority o the key perormance indicators recommended or measuring reach are
simply percentages and ratios, within acquisition we start to see the emergence o what are
oten reerred to as string statistics,` such as reerring domains and search keywords and
phrases, as aluable indicators.

1he ollowing are excellent acquisition perormance indicators and should be considered or
inclusion in regular reporting regardless o business model:

Percent new isitors to alert you to signiicant changes in your new isitor
acquisition programs and their eect on your oerall traic.
Aerage number o page iews per isit or all isitors to ensure that your content
consumption stays relatiely stable. Stability in this metric is an indirect
measurement o the user experience on your \eb site.
Page stickiness` or your home page and key entry points into the site to watch
or dramatic changes in stick` and immediate bailout.
Response rates or all actie campaigns to determine when a campaign has
eectiely run its course and needs to be replaced, rereshed or simply ended.
Cost-per-acquisition and cost-per-click or actie campaigns to ensure that you are
not paying more than you can aord or any campaign. \ou should create a
1, CIupLer 1z

column or maximum CPA,CPC` so that you can clearly see i you hae
exceeded that limit.
1he top three to ie reerring domains and olume o traic you receie rom
each to watch or changes in reerring traic patterns.
1he top ie to ten search keywords and phrases and the olume o traic you
receie rom each to watch or changes in isitor interest as expressed by external
search.
Aerage time spent on site and percent o isits under 90 seconds should be
watched closely and correlated to any inbound marketing actiity.

Keep in mind that each o these will not be necessarily interesting as raw numbers, the most
important aspect o any perormance indicator is the relationship between the current and
preious measurement. Simply a leading indicator o changes happening, each o these KPIs
is designed to alert you to the act that some aspect o your isiting traic has changed and
more research is warranted.

MeusurIng ConversIon 1,

CHAPTIR 1
MIASLRING CONVIRSION


Lvery Web site is different, and nobody is selling exactly the same thing in the same
way to the same audience. We've found over J,J00 variables that affect conversion
rates on a Web site, and those variables can be broken down into even more detail.
Bryan Lisenberg in Converting Search Lngine 1raffic


lor most businesses with an online presence that is used to help grow the oerall business-
especially those businesses that exist solely online-getting a isitor to somehow conert`
is the ra.iov a`etre. Regardless o what the act o conersion actually is, it is this act that keeps
the business growing and moing towards proitability ,or increased proitability,. \hile
many \eb sites express that they seek to reach and acquire traic in order to build brand`
it is the author`s opinion that this is A, not enough and B, not likely to be a successul long-
term goal or one`s online marketing programs unless the company has extremely deep
pockets. 1he customer lie cycle is applicable to many, i not all, business models or a good
reason and conersion is the critical step in this process.

Measuring conersion is really about measuring abandonment, the people that start any
gien actiity but or whateer reason do not complete it. Abandonment, described in
greater detail below, is an expression o a isitor`s lack o commitment to or understanding
o the process you are trying to get them to complete. Online isitors, except in ery special
situations ,such as where the oer is simply too good to pass up,, will seize nearly any
opportunity to bail out o a process and seek ulillment elsewhere. Len though published
conersion rates are creeping upwards as \eb site operators become more adept at using
analytics data like that described in this book to improe online processes, industry aerage
site-wide conersion rates` still hoer around 3 percent ,Shop.org,. Assuming that your
online business exists to either sell something or proide some kind o aluable inormation,
i you are aerage,` is making a sale to three in a hundred people good enough or you

1he issue o industry aerage comparisons brings up a ery important point on the topic o
conersion-you should only compare your conersion rate to itsel in an eort to increase
said rate. \hile it is ery tempting to try and determine what your competitors` conersion
rates are and then beat that rate it is most likely an exercise in utility. lirst, you`d be lucky
to be able to igure out what your competitors` conersion rates are as smart businesses keep
these numbers secret and only share ague inormation about increases or decreases.
Second, unless you market the exact same way to the exact same audience, hae the same
products at the same prices, and so on, there is no reason to make this comparison. Smart
business managers will ocus on making use o the continuous improement process as
described in Chapter 2 to design, implement and measure changes with the goal o
1,q CIupLer 1

increasing key conersion rates without regard to published aerages and competitie
inormation.
Which Conversion Rates Should Be Measured?
As stated in Chapter 8 there are many dierent conersion rates that can be measured and
they exist or eery \eb site regardless o business model. 1he most obious actiity to
measure to completion,conersion is the online purchase process. \hether you are selling
shoes, CDs or airline tickets, applying the continuous improement process to the online
purchase process is one o the astest paths to achieing return on inestment or your
inestment in any analytics application. 1hat being said, there are three major categories o
conersions that can and should be measured online:

Actiities leading to an online purchase
Actiities leading to the collection o aluable` inormation, such as registration
Actiities leading the isitor to inormation that will reduce operational costs, such
as calls to your technical support group, and so on

Lach o these actiities can be easily deined as a series o discreet steps, although any single
step may contain more than a single page. 1his deinition in tandem with the process
measurement tools described in Chapter 6 will enable accurate measurement o
abandonment and conersion rates help you improe these rates.

\ith the categories aboe in mind, there are a handul o common actiities that are worth
discussing regarding abandonment and conersion.
Site-wide Conversion Rate
1he site-wide conersion rate, also commonly reerred to as the order conersion rate,` or
purchase conersion rate,` is the standard conersion rate that is oten used in conersion
rate studies. Calculated as orders ,or appropriate goal, diided by total site-wide isits or
isitors, this conersion rate will gie you a eel or how well you conince isitors to
complete goals regardless o their entry point on your \eb site.
Home to Purchase
1he home page to purchase rate is the number o orders taken diided by the number o
isitors ,or isits, to a site`s home page. lor an online store this would indicate isitors
traersing the ollowing rough path:

1. lome page
2. Product or category page,s,
3. Shopping cart
4. Checkout
5. Order complete
MeusurIng ConversIon 1,g


Between the home page and the completed order there can be any number o pages isited
in any order. Lssentially what this actiity captures is the number o isitors who enter the
site ia the most common` entry point and the abandonment rate or those isitors
through the generalized purchase process.
Search to Purchase
Search to purchase is a ariation on the home to purchase` actiity or those sites that rely
heaily on internal search to guide isitors towards online purchases. 1his is essentially the
same as the home to purchase` actiity except that the irst step would be a search results`
page:

1. Search results
2. Product or category page,s,
3. Shopping cart
4. Checkout
5. Order complete

1he reason you`d start measuring this actiity at the search results` page and not a search
orm is that most sites now oer search orms on most, i not all, pages. I you hae a single
dedicated page or search that isitors need to intentionally click to in order to search you
would make that the irst page in a six-step process.

Measuring search to purchase` and comparing that to A, your home to purchase` process
and B, your oerall all isits to all orders` conersion rate will help you assign alue to
search on your \eb site. It is not uncommon or sites haing robust, accurate and well-
presented search results to hae a higher conersion rate or search to purchase` than
home to purchase` and the site-wide aerage ,keeping in mind that this actiity contributes
orders to the site-wide aerage,.
Special Offer to Purchase
Special oer to purchase` is again nearly the same as home to purchase` and search to
purchase` except that here you would be measuring the eect o dierent merchandising
and pricing options. 1he rough path describing this actiity:

1. Special oer or pricing page,s,
2. Shopping cart
3. Checkout
4. Order complete

1his actiity as described assumes that you hae an add to cart` option on the pages in step
1. I you don`t you`d likely want to add a step to capture that inormation, helping you to
see how likely isitors are to be interested enough in your special oers or pricing to actually
put an item in the shopping cart.
1,6 CIupLer 1

Lead Generation
1he attempt to entice isitors to proide your business with some type o personal
inormation is arguably the most popular actiity on the Internet. 1his process is essentially
the measurement o abandonment rom the point where you tell the isitor why they should
proide this inormation to the point where you hae successully gathered that inormation.
1he shortest ersion o this process is:

1. Proide orm and incentie to ill it out
2. 1hank the isitor or proiding the inormation

In reality this process is more oten something like:

1. 1ell the isitor why they should complete the process
2. lorm 1
3. lorm 2
4. lorm 3
5. 1hank the isitor or proiding the inormation

\ou can see the alue in measuring abandonment through this actiity, especially or longer
processes. Imagine in the second example aboe that you implement your particular \eb
analytics package`s ersion o process measurement and discoer that you hae 50 percent
abandonment between orm 2` and orm 3.` I you are practicing continuous
improement you will immediately re-examine orm 2` to make sure that you really veea
each piece o inormation you ask or. I the answer is no,` you should redesign the orm
to only ask or inormation you vv.t bare, deploy the changes and measure the eect on
abandonment at this step and on the oerall conersion rate.

MeusurIng ConversIon 1,,

Web Analytics Tip: How to Ask Visitors for Information and
Get Results
In the author`s experience, the single most common reason that isitors abandon lead
generation processes is the request or intrusie or seemingly irreleant inormation.
Regarding the ormer, unless you are a inancial institution you should neer ask a isitor or
their social security number or mother`s maiden name, eer. \ith reports o identity thet
on the rise the likelihood that a stranger will proide this inormation willingly is extremely
low. Regarding the latter, consider that isitors understand that nothing good comes or ree
on the Internet-they are likely to proide some inormation to get something that has
perceied alue. 1hat being said, there appears to be an anecdotal correlation between the
number o orm ields you ask a isitor to complete and the likelihood they will abandon the
orm. Marketers always seem to want to ask questions like low did you hear about us`
and low many people do you superise`-questions that are ery easy to answer with
garbage and may simply make the orm look too long to bother completing.

In general, unless you are oering some amazing delierable that a isitor is unlikely to be
able to get elsewhere, ollowing these simple rules will help you improe completion or this
type o actiity:

Do not ask or any inormation that is not ab.otvtet, reqvirea, period.
Unless you are a trusted brand and a inancial institution, neer ask or any personally
identiiable inormation that could lead to identity thet. I you are a inancial
institution only ask or this inormation i the isitor has intentionally started a
process.
I the number o required ields is especially large, consider breaking the orm up into a
small number o logically grouped questions oer a number o pages. 1ry and
ensure that each o these pages is quick to load and clearly identiy or the isitor
how much o the process they hae completed ,such as Step 2 o 4`,.
I possible, perorm client-side orm alidation rather than orcing the page to reload to
display any problems with the inormation the isitor has proided. Client-side
orm alidation saes isitors time and reduces the risk that other required orm
ields will be cleared ,password ields are the most common problem,.
Do not ask or any inormation that is not ab.otvtet, reqvirea, period.

As you can see, the most important thing is to not ask or optional` inormation unless
you`re oering the most amazing thing eer and it is absolutely ree o charge. lollowing
these simple rules should help you increase your registration and orm completion rates
signiicantly.

Navigation to Important Information
1he process o naigating to important inormation is the most generic actiity o all since
you can easily classiy a purchase conirmation page as important inormation.` Still, it is
worth treating this process as special or the legions o \eb sites that exist to help isitors
1,S CIupLer 1

ind inormation in requently asked question ,lAQ, documents. In this model, the isitor
is likely ollowing a low similar to:

1. Support site home page
2. Page containing lists o support categories
3. Page containing lists o commonly asked questions
4. Page containing a speciic answer to an lAQ

\hile your site may hae a completely dierent series o pages the idea is to measure the
low o isitors moing along sequential, required steps towards speciic inormation. 1he
reason you`d measure abandonment and conersion along a path like this is that theoretically
there is some adantage to your isitors inding this inormation. \hile there are many
estimates or the operational costs associated with maintaining support sta nearly eery
analyst agrees that sel-serice ia the Internet is the most cost-eectie solution or
proiding support. 1his statement is only true i isitors can actually ind inormation on
your \eb site, hence the need to measure low rom less to more speciic inormation.
Using Process Measurement Tools
\ith all o the aboe in mind the only challenge that remains is to identiy A, which
processes on your \eb site you need to measure and B, how the particular \eb analytics
application you hae allows you to do that. Some questions you may want to ask when
deciding which processes you need to measure include:

\hen isitors complete the process does your business make money
I isitors ail to complete the process might they pick up the phone and call you
instead
\hen isitors complete the process do you know more about them and can you
market directly to them in the uture
I isitors ail to complete the process might they complete a similar process on
your competitor`s \eb site
lae you spent money on an enabling step in the process or which you need to
quantiy your return on inestment

I you answer yes` to any o these questions that process is an excellent candidate or actie
measurement. Once you hae made the decision to measure a process the next step is to
setup your application to do so. 1he best adice the author can gie you regarding setup is
to contact your application endor and explain to them what you are trying to measure.
Make sure they understand you would like to be able to measure isitor low through a
sequential process and make measurements based on isits ,best, or isitors ,second best,.
Metrics to Help Measure Conversion
Once you hae decided what processes to measure and how to measure them, the ollowing
metrics will help you understand isitor success and ailure in each.
MeusurIng ConversIon 1,q

Conversion Rates
1he conersion rate or any process is the critical measurement o how likely isitors are to
complete the processes. lrom conersion rate you can estimate isitor satisaction, the
oerall eect o your marketing campaigns and the likelihood that your online business will
be successul.

Conersion rates are one o the most aluable perormance indicators that should be tracked
on a regular basis
How to Calculate
lor any multi-step process you are measuring, the conersion rate or the process is simply:

COMPLETIONS / STARTS = CONVERSION RATE

Keep in mind that there are macro conersion rates, such as ratio o all orders to all isitors
to the \eb site, and micro conersion rates, such as the number o leads generated rom
respondents to a single campaign or campaign component. It is ery likely that you will
measure a number o dierent conersion rates on your \eb site.

Also keep in mind that when you are making a conersion rate measurement you want to
use like` metrics, that is, isits or both the numerator ava denominator, and not isits or
the numerator and page iews or the denominator.
Dependence
1he major dependence or measuring conersion rate is being able to get accurate and
similar numbers or both the number o completers` and starters.` Some examples o
similar data types or conersion rate measurements include:

Visits to inal step , isits to irst step
Visitors to inal step , isitors to irst step
Orders , Visitors
Orders , Visits ,both isits and isitors are unique measurements,
Registrations , Visitors or Visits

Conersely, some examples o dissimilar data types include:

Visits to inal step , page iews to irst step
Visits to inal step , isitors to irst step
Orders , page iews
Registrations , page iews

In general i is not recommended to use page iews or any conersion rate measurement
because page iews are easily inlated ,page reloads inlate page iew stats,. Unless page
iews are the ovt, metric aailable to your analytics program they should not be used.
1So CIupLer 1

Usage
Conersion rate measurements are most requently used to determine the rolling success o
both the \eb site and indiidual processes contained in it. As one o the most aluable key
perormance indicators, conersion rates should be watched daily to ensure that marketing
actiities continue to reach qualiied isitors and bring them to the \eb site. Any dramatic
change in conersion rates should be immediately researched to determine what caused a
rise or a all in the rate.
Interpretation
Conersion rates describe the percentage o people who successully complete a designed
process. So i the conersion rate or a process is reported as 10 percent this tells you that
one in ten isitors completed the process.
Abandonment Rates
Abandonment rates or any process describe the number o isitors who ail to moe along
successie required steps in that process, typically measured on a step-to-step basis.
Measuring the number o isitors lost rom step to step is extremely important to the
continuous improement processes-knowing which steps in any process isitors bail out at
high rates is the single best way to ocus resources towards improing oerall conersion or
that process.
How to Calculate
1he calculation or abandonment rate should be made or each step in a multiple-step
process. So i you are measuring the conersion rate or a ie-step process you should also
be using the process measurement tools that your \eb analytics package to measure our
dierent abandonment rates, between steps one and two, steps two and three, steps three
and our, and steps our and ie. Depending on whether you are using isits or isitors to
make process analysis measurements, the calculation is ,using isits as an example,:

1 (NUMBER OF VISITS AT CURRENT STEP / NUMBER OF VISITS AT PREVIOUS
STEP) = ABANDONMENT RATE FOR CURRENT STEP

As you can see there is no abandonment rate or the irst step in any multiple-step process
since there is no preious step.`
Dependence
Abandonment rates hae similar dependence as conersion rates, using similar types o data
measured oer same timerames.
MeusurIng ConversIon 1S1

Usage
lor any multiple-step process abandonment rates are critical to know how to improe the
processes conersion rate. It is one thing to know what the conersion rate is or an actiity
or processes, it is another to know at which point isitors are deciding to no longer moe
along in the process. Knowing which pages or steps in a process contribute most
signiicantly to loss is the irst step in stemming that loss and increasing the conersion rate.
Abandonment rate measurement is critical to the continuous improement process, as
described in Chapter 2.

In general, when you are able to accurately calculate abandonment rates or any gien step in
a process you want to ask yoursel whether there are marketing or process changes that you
could make to pages in that step to encourage more isitors to moe to the next step.
Important questions to ask include:

Are there orms in this step that may be asking or too much, unnecessary or
personal inormation
Is the action or actiity you are asking the isitor to complete at this step easy to
identiy Are buttons obious
Is the marketing material clear and compelling so that isitors are likely to begin or
continue the process
Is the page quick to load and consistent in look and eel to preious pages in the
process

In a situation where you are using the continuous improement process you may also want
to run an A,B test on the existing site and the changed site to ensure that any changes you
make hae lower abandonment than the current site. See Chapter 6 or more inormation
on how A,B testing makes use o abandonment and conersion rates.
Interpretation
I an abandonment rate between two steps is reported as 85 percent this tells you that 85 in
100 isitors ailed to moe rom step to step. But with an 85percent abandonment rate, you
know that 15 in 100 isitors aia successully moe rom step to step.
Campaign Conversion Metrics
As discussed in Chapter 12, measuring campaigns as they drie traic to your \eb site is
extremely important. Perhaps more important is measuring how these acquired isitors
conert rom simply being browsers to being connected` to your online business. 1he
ollowing are conersion metrics speciic to marketing campaigns you may be running to
drie traic to your site:
1S CIupLer 1

Campaign Conversion Rate
1he conersion rate or a campaign is the same as the general conersion rates discussed
aboe but is speciic to a single source o traic to your \eb site. Recall rom Chapter 8
that any marketing ehicle that dries traic to your \eb site as a campaign` and employ
campaign analysis tools to measure the eect o that traic. Measuring campaign
conersion rates is critical in that, hopeully, you are trying to reach and acquire isitors to
your \eb site vot ovt, to build brand or name awareness but also to entice those isitors to
complete a speciic goal or series o goals. \ou should be measuring what campaign
respondents are doing on your site in terms o their completion or conersion through those
named goals.

Much as your \eb site has many dierent processes that can be measured using the process
measurement tools described in this book, a gien campaign can hae multiple goals. I you
hae a typical online business that is selling products and serices on the Internet you likely
hae the ollowing named goals somewhere on your \eb site:

Purchase process complete
Item added to shopping cart
Shopping cart checkout initiated
Registration process complete
Newsletter signup process complete
Send to a riend` unctionality has been used

\ou may also hae some soter` goals on your site including:

1hree or more product iews in a isit
lie or more minutes spent browsing in a isit
1wo or more isits in a day, three or more isits in a week, and so on

\hen you are setting up your online marketing campaigns you want to be sure to gie
careul thought to what named goals you want to measure or your campaigns. Some
endors proide unctionality to allow conersion through speciic goals, others proide
unctionality to measure conersion through all named goals. Ideally the endor you use will
proide at least the ability to measure conersion through all named goals, as haing only a
single goal aailable to a gien campaign is limiting and will not paint the complete picture
about how isitors respond to your campaigns.

Campaign conersion rates are excellent conersion perormance indicators, at least or
actiely running campaigns.
How to Calculate
1he campaign conersion rate calculation is simply the number o isitors completing
speciic goals on your site diided by the number o respondents or isitors responding to a
speciic campaign:

MeusurIng ConversIon 1S

CAMPAIGN COMPLETERS / CAMPAIGN RESPONDENTS = CAMPAIGN CONVERSION
RATE

It is important to note that you should only make this measurement or isitors responding
to a speciic campaign unless you are attempting to generate an all campaigns` conersion
rate.
Dependence
1he major dependence in calculating campaign conersion rates is ensuring that your
analytics application is able to proide you both the number o completers and the number
o respondents to an indiidual campaign. \ou need to be careul not to mix and match
campaigns and also be careul not to mix and match metrics, that is, do not use
completions` diided by respondents` as the ormer is a non-unique metric, the
equialent o a page iew, and the latter is a unique metric, the equialent o a isitor.
Usage
1he use o campaign conersion rates is similar to that o generalized conersion rates in
that you should work towards improing these rates through judicious use o the continuous
improement process. Campaign conersion rates go one step urther, howeer, and you
should calculate the minimum conersion rate or any campaign you are paying or, using
the strategy described in Chapter 12 under cost-per-click,cost-per-acquisition, and consider
eliminating or ending campaigns that do not meet minimum perormance requirements.

\hen you hae multiple named goals, it is important to examine how isitors responding to
a gien campaign low through these multiple goals. It is not uncommon or a marketing
group to design a campaign intended to attract isitors who are likely to complete goal A,`
and then actually attract isitors who complete goals B and C.` laing isibility into this
actiity allows marketers to learn rom their mistakes and better understand the audience
they reach and acquire rom dierent marketing outlets.
Interpretation
Interpretation o campaign conersion rates is similar to that o generalized conersion rates,
but speciic to an indiidual campaign. I your campaign conersion rate is 2.5 percent this
tells you that 25 isitors in 1,000 completed one or more o the dierent named goals you
hae established on your \eb site.
Example
As mentioned in Chapter 12, the author engaged in online campaigning in the last hal o
2003 in an attempt to drie isitors to the !eb .vat,tic. Dev,.tifiea \eb site and download
the preiew copy o this book. Lmail, anity URLs and search keyword buying through
Google were all employed to create leads.


1Sq CIupLer 1


Iigure 46: Campaign conversion rates from the author's Web site. 1he act of
conversion at the time was the download of a preview copy of Web Analytics
Demystified.

As you can see in ligure 46, most o the methods that the author used to generate leads
were eectie rom a conersion rate standpoint. Len though ery ew isitors came
through the Stanord Publishing group` anity URL ,the O1L-Stanord` campaign,, more
than hal o these respondents downloaded a copy o the book. Len the Google
campaigning, or which the author spent out-o-pocket against uture returns on book sales,
was eectie, conerting at more than 12 percent.
Cost-per-Conversion
Cost-per-conersion is a simple, yet important, calculation that helps marketing managers
know their campaign money is being spent wisely. Much like cost-per-click and cost-per-
acquisition, cost-per-conersion is a key perormance indicator or campaigns.
How to Calculate
Cost-per-conersion is simply the running cost o the campaign to date diided by the
number o conersions in the same timerame:

COST OF CAMPAIGN TO DATE / CONVERSIONS TO DATE = COST-PER-
CONVERSION

It is important to be careul to ensure that you are using numbers rom the same timerame,
so i the campaign has been running or 10 days and you hae agreed to pay >30,000 or the
month you would use the number o conersions rom the last 10 days as a denominator
and >10,000 as the numerator. I you were to use the ull alue o the campaign ,>30,000,
you would be artiicially inlating the cost o the each conersion.
Dependence
1he primary dependence or cost-per-conersion is using data rom the same timerame.
Also, or some types o campaigns, cost can be diicult to calculate. In cost-per-click ,CPC,
campaigns you would need to calculate the total cost or the timerame by multiplying the
number o clicks by the cost-per-click unless you are able to get a daily summary o costs
rom some administratie interace.
Usage
Cost-per-conersion is used in conjunction with your marketing goals and known alues or
conersion and customer acquisition. Imagine the scenario where in the course o a month
MeusurIng ConversIon 1Sg

,30 days, you get 1,000 conersions and your aerage order alue is >10.00. In this case the
total ,estimated, alue o the campaign in the month is >10,000. Next imagine that the total
adertising costs or the campaign were >30,000 or the month. 1he measured cost-per-
conersion would be >30.00 per conersion, three times your aerage order alue!
Interpretation
Since cost-per-conersion is a straightorward calculation it is perhaps more useul to look at
a way to back into an appropriate cost-per-conersion:

1. Determine the aerage alue o a conersion.
2. Determine the multiple you`d like to make rom the campaign.
3. Diide the results rom step 1 into the results rom step 2. 1his is the maximum
cost-per-conersion you should pay or this campaign.

lor example:

1. Our aerage alue o a conersion is >30.00.
2. \e like to make a 5x multiple on our campaigns.
3. 1he most we can aord to pay or each conersion is >6.00.

I, when you calculate your cost-per-conersion or the campaign oer time, you are paying
more than >6.00 per conersion you know that you are paying more than your target spend
on the campaign. I this is the case you can either A, decide to end the campaign or B,
decide that the multiple you hae picked is too aggressie and scale that back. I, in the
example aboe you only wanted a 2x multiple you could pay up to >15.00 per conersion.
Example
1he author o !eb .vat,tic. Dev,.tifiea used Google`s Ad\ords program to generate leads
or the book, collecting names and email addresses or people whom he hoped would
ultimately purchase a copy o the book. Although this data has already been presented in the
example in Chapter 12 or cost-per-acquisition, in the context o the author`s campaigns the
data is interchangeable with cost-per-conersion.
1S6 CIupLer 1




Iigure 47: Cost-per-conversion information for a sample of keywords the author
purchased to drive leads to www.webanalyticsdemystified.com (data provided by
Google AdWords).

As you can see in ligure 4, the Google Ad\ords application automatically calculates cost-
per-conersion ,Cost,Con.`, or those keywords that were successul in driing isitors
who conerted into leads. \hile at the time the alue o a lead was technically >0.00, the
author estimated that each lead generated would ultimately be worth between >5.00 and
>9.00. Based on this inormation, the reported costs-per-conersion o between >0.19 and
>0.44 were well within the author`s budget or lead acquisition and conersion.
Campaign Return on Investment (ROI)
Compared to other campaign metrics described in this book, campaign return on inestment
is perhaps one o the most important measurements you can make i you are doing any
online marketing. \hile it is important to measure your cost-per-acquisition and cost-per-
conersion it is the relationship, between the total amount you spend and the total amount
you make that will help you determine the oerall alue o your online marketing.
How to Calculate
Campaign return on inestment is ery simple to calculate or any gien timerame but there
are two ariations:

AMOUNT MADE BY CAMPAIGN TO DATE / AMOUNT SPENT ON CAMPAIGN TO DATE
= ROLLING CAMPAIGN RETURN ON INVESTMENT

AMOUNT MADE BY CAMPAIGN TO DATE / AMOUNT SPENT ON CAMPAIGN IN TOTAL
= TOTAL CAMPAIGN RETURN ON INVESTMENT

1he rolling` campaign ROI is simply an expression o how well you are doing so ar in the
campaign, likely to become a positie number more quickly than the total` campaign ROI.
1he two are dierentiated to help you understand both how the campaign is currently
MeusurIng ConversIon 1S,

perorming and when you can expect the campaign to break een i the campaign perorms
at a roughly linear rate.

Again, the amount made by the campaign would simply be the gross proits the campaign
has deliered, the amount spent on the campaign would be the cost-per-acquisition times the
number o responses or the rolling` number, or the total cost o the campaign ,or a
suitable estimate or upper-limit,. Keep in mind that the ormer can be an estimate based on
the aerage alue o a customer times the number o conerters in the campaign i you are
not actually selling online.
Dependence
1he only real dependence or campaign ROI is haing accurate numbers to work with,
which should not be so diicult-or should it Depending on who controls the low o
campaign spending and campaign conersion these numbers can be diicult to come by. In
cases like this it is worthwhile to work with an aerage cost-per-acquisition and an aerage
alue-per-conersion. \ith these numbers and access to the number o respondents and
number o conerters you can make a rough estimate o your campaign ROI, which is better
than nothing.
Usage
Campaign return on inestment should be used to ensure that you are not losing your shirt
on a campaign on a daily or weekly basis, to predict at what point the campaign will begin to
become proitable or the online business and to determine when you beliee you will start
to see the multiples o return that you are looking or.
Interpretation
1he best way to think about return on inestment is to think, lor eery dollar I hae spent
I hae made back X.` I the rolling` calculation yields a number like 0.50` this tells you
that or eery dollar you hae spent in the timerame you are looking at you hae made back
>0.50. I the total` calculation yields the same alue this tells you that or eery dollar you
hae committed to spending on the campaign you hae made hal o that dollar back. In
this example the rolling ROI` calculation is not great news, unless something pretty
dramatic changes a loss like this will make it ery diicult to realize a positie ROI on the
entire campaign. In the same example the total ROI` calculation is not bad in and o itsel
since all campaigns start o in the red and ,hopeully, moe towards the black.

1he interpretation o the total ROI case is that you are halway towards break-een on the
campaign, which tells you roughly how long you will need to run the campaign.
Unortunately you cannot just say to yoursel, Ok, i we`re halway there then we`ll just run
the campaign or three more similar time periods and we`ll make a 2x return.` Reality
dictates that campaign perormance decreases oer time so unless you are proiding
continuous input you cannot be sure the response and conersion rates will be linear. 1hat
being said, i the inputs remain roughly the same you will be able to generate a relatiely
accurate picture o when the campaign will contribute positiely to your online business.
1SS CIupLer 1

Segment, Commerce and Search Related Conversion
Metrics
1he isitor segmentation and commerce tools described earlier in this book hae
conersion-related attributes. lor isitor segments it is the rate at which a isitor moes
rom a lower-alue segment ,browsers`, to a higher-alue segment ,buyers`,. lor
commerce there are a number o dierent metrics that are important and nearly all reole
around the key conersion actiity o making a purchase.
Segment Conversion Rates
One o the components cited as important in the chapter on isitor segmentation was the
ability to measure the moement rom one segment to another. 1his measurement is
essentially the conersion rate or groups o isitors completing any set or number o
actiities, depending on how you assign isitors to segments. Depending on your particular
\eb analytics package`s support or segments you can perorm more or less complex
segmentations. Some examples o segment conersions that are commonly tracked include:

Unregistered` to Registered` isitors
Browsers` to Buyers` in an online store
Anonymous` to Known` isitors
1hose isitors reerred rom a particular source to Customers` or Registered`
isitors
Visitors rom a particular geographic region to Customers` or Registered`
isitors
Light` content users to leay` content users
Inrequent` isitors to lrequent` isitors
Campaign Responders` to Directly Reerred` isitors ,and ice ersa,

1he list is nearly as ininite as the number o online businesses. 1he most important things
to consider when setting up isitor segments are A, does the segment you are creating make
sense and B, will you be able to reasonably act on the inormation once you hae collected
it.
How to Calculate
1hose \eb analytics packages that support this type o analysis automatically calculate
segment conersion or you. I you hae the ability to segment but not conert segments
you may be unable to measure this metric. I you are unsure about how your particular
analytics package supports segment conersion please consult your endor directly.

1o generate a segment conersion rate you would simply diide the number o segment
conerters into the number o isits or isitors in the lower-alue segment during the
timerame being analyzed:

MeusurIng ConversIon 1Sq

SEGMENT CONVERTERS / VISITORS TO LOWER VALUE SEGMENT = SEGMENT
CONVERSION RATE

It is important to use the same metric or both the numerator and denominator, i the
segment conersions are measured in isits use isits, i the segment conersions are
measured in isitors, use isitors, and so on.
Dependence
Segment conersion depends entirely on the analytics package`s ability to record and present
this inormation. Because, or the most part, isitor segmentation occurs behind the scenes,
either in cookies or in query strings passed around in \eb serer log iles, you depend on
the application`s ability to track segment conersions. Some applications, particularly some
outsourced solutions, proide the ability to denote segment conersions in the page tag,
other applications proide the ability to perorm an aa boc analysis o preiously collected
data to report on segment conersion.
Usage
Segment conersion is typically used to keep track o process conersions occurring oer
longer periods o time than just the isit` or the day. Because isitor segmentation is
designed, in most cases, to keep track o isitors oer the lietime o their isits ,once you
are segmented you stay in that segment until the application or analysis is told otherwise,,
these tools allow us to obsere isitors oer long periods o time.
Interpretation
\our application will hopeully proide at least two o three useul metrics or tracking
segment conersion. 1hese are:

Number o conerters rom segment A` to segment B`
Aerage amount o time it took or the segmentation to occur in days, hours, minutes
and seconds
Aerage number o isits between the time the isitors entered segment A` and when
they entered segment B`

\hile we reer to the aerages or the second or third items aboe it would probably be
more useul to hae the distributions or these numbers but, as is oten the case in \eb
analytics, you take what you can get.

laing the inormation mentioned aboe aailable to you will allow you to assess how
eectiely and how quickly you moe isitors rom lower- to higher-alue segments. I you
were to see that you were conerting isitors rom segment A` to segment B` at a rate o
20 percent but the aerage time to conersion was 20 days and the aerage number o isits
was 30 you may reasonably assume that you hae a airly complex conersion process or that
marketing materials encouraging isitors to conert is not entirely clear. See Chapter or
an example o how BackcountryStore.com used segment conersion to improe its checkout
process.
1qo CIupLer 1

Example
An excellent example o how segment conersion metrics can help in the continuous
improement process is presented in Chapter under the heading Lxamples o Uses or
Visitor Segmentation 1ools.` Lssentially, BackcountryStore.com`s sta questioned whether
they should require registration in their purchase checkout process. 1o alidate whether
isitors wanted to register or not, they created an anonymous` test checkout process. 1hey
then let isitors go through the registration whicheer way they desired, segmenting and
conerting isitors as they did so.



Iigure 48: Segment conversion report for BackcountryStore.com. Of interest is the
difference in conversions from Unregistered => Customer and Registered =>
Customer.

As you can see in ligure 48, a clear majority o isitors were opting to go through the
purchase checkout process anonymously, ollowing the Unregistered ~ Customer` path.
Based on this inormation, BackcountryStore.com kept the anonymous checkout process in
place. 1his streamlined checkout process was in part responsible or BackcountryStore.com
increasing its site-wide conersion rate by nearly 30 percent, a notable accomplishment or
any online business.
Average Order Value (AOV)
Aerage order alue is a commerce metric that should be watched with the same religious
eror that is most oten resered or conersion rates i you sell anything on your \eb site.
Depending on the number o dierent items you sell, this alue can help you understand
how your business is doing rom a 10,000-oot iew. Aerage order alue can be calculated
at both a macro and a micro leel, that is, you can calculate the AOV or all orders taken
online, AOV or a category o items and AOV or a single product or line o products.

At a minimum, i you sell online, AOV should be calculated at the site leel as a key
perormance indicator or the health o the online store.
How to Calculate
Very simple to calculate, the site-wide AOV is the total reenue generated diided by the
number o orders taken.

TOTAL REVENUE / TOTAL ORDERS = AVERAGE ORDER VALUE
MeusurIng ConversIon 1q1


I you are interested in the AOV or a product category or a single product you would make
the same calculation or the orders and reenue taken or that particular group.
Dependence
Aerage order alue only depends on haing an accurate and timely way to determine how
many orders hae been taken and or how much reenue-undamental data or any store,
online or oline.
Usage
Aerage order alue is best used as a key perormance indicator and or making simple
conersion rate calculations. An example o the latter would be to quickly determine the
eect o a 10 percent increase in your site-wide conersion rate, so i your current
conersion rate is 5 percent and your AOV is >100.00, i you increase your conersion rate
by 10 percent to 5.5 percent you stand to make an additional >500.00 per 1,000 isitors to
your site.
Interpretation
Aerage order alue is simply the aerage sales or the items you hae sold through your
\eb site. I you only sell one item or one price the AOV or your site will be the same as
the price o that one item. I you sell two things or two dierent prices, the closer the
AOV is to one price or the other the greater the number o either o those items you are
selling.
Related metrics
Aerage order alue is related to a number o dierent commerce metrics that are, hopeully,
core unctionality in your online commerce tracking toolset, including number o orders, the
number o items that were sold, site-wide conersion rate and total online reenue. See
Chapter 9 or additional inormation on common commerce metrics.
AOV for New and Returning Customers
In addition to calculating the aerage order alue or all items sold online, both oerall and
by product or category, it is useul to make this same calculation or new and returning
customers. Doing so will help you to better understand the alue o isitor and customer
retention. I you sell online, these calculations are ery important conersion perormance
indicators.
How to Calculate
Both are calculated in the same way as aerage order alue except the data used is or either
new or returning customers. See aerage order alue ,AOV, or the calculation.
1q CIupLer 1

Dependence
Being able to make these calculations depends on your analytics application`s commerce
package being able to dierentiate new` and returning` customers. Oten done ia the
passing o a unique customer ID during the purchase process, the accuracy o these
calculations is entirely dependent on knowing how many orders and how much reenue
should be attributed to each o these customer segments.
Usage
New and returning customer AOV is extremely useul in determining the oerall makeup o
your marketing programs. \ere you to discoer that your new customer AOV was >100.00
but your return customer AOV was only >5.00 you could reasonably justiy ocusing your
marketing eorts on new customer acquisition. But i your site`s new customer AOV is
>50.00 and your return customer AOV is >500.00 you would immediately ly into customer
retention mode and attempt to lure as many preious customers back to your \eb site as
possible.
Interpretation
See aerage order alue ,AOV,.
Example
Aerage order alue is an excellent metric to watch in your key perormance indicator report,
one that can tell you a great deal about your online business. An example o how AOV
should be reported is presented in ligure 49.



Iigure 49: Key performance indicator worksheet for an anonymous online business
showing AOV for all, new and repeat visiting customers.

Based on the numbers presented in ligure 49 you can reasonably assume that something has
changed or the better between reporting periods, with all AOV metrics up between 10 and
21 percent. 1his inormation would easily alidate any marketing actiity, especially
retention marketing actiity that the online business was engaged in at the time.
MeusurIng ConversIon 1q

Percent of Orders from New and Repeat Customers
I your \eb site is ocused on selling more than a single product and your hope is that
isitors will purchase and then return to purchase again, it is useul to keep an eye on the
percent o order you are receiing rom new and repeat customers. Complimentary to AOV
or new and returning customers, these calculations will help you better determine how
much eort you should be putting into arious customer acquisition and retention programs
and whether they are paying o or not.
How to Calculate
1he percent o orders rom either customer type is simply the number o orders rom new
or returning customer diided by all orders.

NEW CUSTOMER ORDERS / ALL ORDERS = PERCENT ORDERS FROM NEW
CUSTOMERS

RETURN CUSTOMER ORDERS / ALL ORDERS = PERCENT ORDERS FROM RETURNING
CUSTOMERS
Dependence
1hese calculations hae a similar dependence as AOV or new and returning customers,
your analytics package`s commerce tools ability to determine and report on whether a
customer is new or returning. See the dependence` discussion under AOV or New or
Returning Customer` or additional inormation.
Usage
Percent o orders rom new or returning customers is useul or helping determine whether
new customer acquisition marketing programs or customer retention marketing programs
are being eectie. I you spend 50 percent o your marketing budget on retention
marketing but only 30 percent o your orders are coming rom return customers something
is not working. Conersely, i you spend 50 percent o your marketing budget on retention
marketing and 65 percent o your customers hae purchased preiously you are doing better
than aerage or retention but haing problems with new customer acquisition.

Ideally these percentages are pretty close to your marketing spend on new ersus return
customer acquisition, proiding you separate out your marketing that way.
Interpretation
1his measurement proides a strict percentage o the distribution o orders between new
ersus returning customers. Similarly, you could make the same calculation based on
reenue to see percentage reenue by new ersus returning customers.
1qq CIupLer 1

Example
In ligure 49 we can see that the AOV or repeat customers is increasing more than 20
percent while the percent o orders rom repeat customers is down roughly 2.5 percent.
\hile the marked increase in AOV is good news, it is slightly less so in the context that the
oerall percentage o orders taken rom repeat customers is down. \ere this anonymous
online business actiely engaged in retention marketing they would likely iew these results
as a qualiied success`-they had increased AOV, likely through merchandising eorts, but
ailed to attract signiicantly increased numbers o return isitors.
New and Repeat Customer Conversion Rates
Closely related to AOV or new and returning customers and the percentage distribution o
orders along the same lines, online stores should calculate the dierential conersion rates
or new and returning customers, i possible. As a measurement o the eicacy o new
ersus return customer acquisition marketing programs, these rates will oten surprise you in
how dierent they are and these dierences are telling when trying to understand barriers to
purchase on a \eb site.

I they are easily calculated, these conersion rates are key conersion perormance
indicators and should be watched closely.
How to Calculate
1his metric is calculated in the same way as a normal site-wide conersion rate ,orders ,
isitors, only or only new or returning customers and isitors:

NEW CUSTOMER ORDERS / NEW VISITORS = NEW CUSTOMER CONVERSION RATE

RETURN CUSTOMER ORDERS / RETURNING VISITORS = RETURNING CUSTOMER
CONVERSION RATE
Dependence
It is important when making these calculations to ensure that you hae accurate
measurements or both the orders and isitors components. Any mixing will necessarily
aect these numbers and potentially skew one calculation or the other.
Usage
Perhaps more than any other metric described in this book, these two conersion rates hae
the potential to help you understand barriers to making purchases on your \eb site. Aside
rom their comparatie alue, each are interesting in their own right as they help you gain
insight into how compelling your products or serices are and how good a job you do at the
oerall sales and ulillment process.
MeusurIng ConversIon 1qg

Interpretation
I these two conersion rates are relatiely ar apart you can determine that there may be a
learning cure to your purchase or checkout process, so that when isitors complete the
process once they are willing to come back and purchase again more reely. Consider i you
had 100,000 new and returning isitors and your new isitor conersion rate was 0.5 percent,
yielding 500 customers, while your returning isitor conersion rate is 2.0 percent, yielding
2,000 customers. In this situation, it is reasonable to assume that there is some barrier to
making the irst purchase but once the isitor has done that they are more likely to purchase
i they return.
Sales per Visitor
Sales per isitor, according to Lisenberg and Noo in 1be Cviae to !eb .vat,tic., is a simple
metric or measuring your marketing eiciency. Similar to the purchase conersion rate,
sales per isitor is a measurement o the amount o reenue you receie, compared to the
total number o isitors to the \eb site ,not only isitors who purchase,. As a comparatie
point with your purchase conersion rate, sales per isitor can be a useul conersion
perormance indicator.
How to Calculate
Sales per isitor are exactly that, gross sales diided by the number o unique isitors in the
same timerame.

GROSS SALES / VISITORS = SALES PER VISITOR
Dependence
Any online business should hae easy access to both o these numbers so sales per isitor
should be easily calculated. \ou should be careul to use de-duplicated unique isitors, not a
sum o daily isitors that may include repeat isitors.
Usage
Sales per isitor is a simple metric or measuring marketing eiciency and also as a check-
and-balance against any changes in your purchase conersion rate. Ideally i your conersion
rate increases your sales per isitor is increasing at roughly the same rate. I it is not, then
you are conerting more browsers into buyers but you are doing so at the expense o your
aerage order alue ,AOV,.
Interpretation
Sales per isitor is a simple dollar alue, such as >10 sales per isitor.` 1his number, by
necessity, will always be much smaller than the aerage order alue, which uses the same
numerator ,gross sales, but a much smaller denominator, in most cases. Sales per isitor will
1q6 CIupLer 1

ary widely rom business to business depending on the aerage order alue and site-wide
conersion rate.
Searches Yielding Results to Search No Results
1he ratio o searches yielding results to those searches yielding no results can be tricky to
calculate, especially i you are using \eb-serer log iles as a data source. But i you can
calculate it you can learn interesting things about your internal search engine. Lssentially this
is a ratio that quickly describes i isitors are successully inding the products or inormation
they are looking or and whether that pattern is changing oer time. 1his ratio makes a good
conersion perormance indicator proiding you hae internal search and the ability to
measure whether searches yield results or not.






Iigure S0: Searches yielding results (above) and one yielding no results (below) from
BackcountryStore.com.

How to Calculate
1his calculation is relatiely simple as long as you hae a way to count page iews ,or isits
or isitors, to search results ound` and no results ound` pages.
MeusurIng ConversIon 1q,


PAGE VIEWS TO RESULTS FOUND PAGE / PAGE VIEWS TO NO RESULTS
FOUND PAGE = RATIO SEARCH RESULTS TO NO RESULTS
Dependence
1his calculation depends on your analytics package`s ability to count the number o searches
that are successul in yielding a set o results ersus the number o searches that yield a no
results ound` message. Many client-side data collection solutions can be tricked into doing
this ia either a page name or custom ariable that simply counts page iews, isits or
isitors to a page called results` or no results.`
Usage
1his ratio is a quick perormance indicator telling you whether more no results ound`
pages are being sered up rom timerame to timerame. 1he closer this ratio is to 1.00-
God help you i it is greater than 1.00-the more likely isitors are searching or terms that
are not indexed in your search engine or simply looking or stu you don`t hae. Ideally you
hae a way to see what people are searching or when they get these no results ound`
pages and can ealuate whether to make changes to your search engine, your site or your
product line to proide better search results in the uture.

I you are using this ratio as a regular perormance indicator, large changes can also indicate
technology problems with your search engine. It is not unheard o or search engines to
return empty or error results sets when they are under particularly heay load, especially
around the holidays. \atching this metric closely can help you better respond to emerging
issues surrounding your internal search unctionality.
Interpretation
I the calculation tells you that the ratio o results ound to no results ound is 0.10` then
you know that one in ten searchers is not inding any results. It is more likely, depending on
your business model, that these numbers will be more like 0.001` or thereabouts, especially
i you hae a well-indexed site, a search engine that is good at interpreting query requests or
a large catalog o products that your isitors understand.

Web Analytics Tip: How to Increase Your Purchase
Conversion Rate
Perhaps the question I get asked most oten rom owners o online stores is, low do we
increase our conersion rate` lortunately I hae deeloped some experience in this realm
and hae discoered the low hanging ruit` that many people oten oerlook. By checking
whether you are making some o these common mistakes, and then using the continuous
improement process to implement, test and measure the eects o correcting the mistakes
you ind, you can be well on your way to increasing conersion rates.

1qS CIupLer 1


lere are three o the lowest-hanging ruit, and thanks to Bryan Lisenberg or alidation
rom his book, 1be Mar/eter`. Covvov ev.e Cviae to Covrertivg !eb ite 1raffic:
Dont Require Registration
lorced registration is perhaps the single most common impediment to any conersion
process. I oten reer to this as the Amazon.com`iication` o the checkout process, that is,
Amazon.com does it and they`re the best in the business so it must be the right thing to
do.`

Not necessarily.

Online businesses require registration in the checkout processes or a ariety o reasons-
the marketing people want the inormation, the company beliees that registration will
simpliy the purchase process the next time the isitor returns, the transaction application
requires a unique user record or the transaction, our competitors ,or Amazon.com, require
registration ,but see below,-none o which are good reasons to reqvire registration. In act,
in my opinion, there are no good reasons to require registration, period. Registration is an
action that should either occur silently during the purchase process or that your isitor
should opt-in to based on their desire.

1here is nothing more annoying or a new online isitor than to ind the products that they
want, to add them to the shopping cart, to click checkout now` and to be taken to a page
telling them, \ou must irst register with our site.` 1his is essentially saying, \e require
you to proide us personal inormation that appears to be unrelated to the purchase process
beore we`ll let you spend your money with us online, and, i you are unwilling to proide
this inormation we`d rather not hae your money.` Sounds ridiculous when it`s stated
that way but consider that the new isitor has no ested interest in your online business until
you`e actually successully and pleasantly shipped the items they want. All o the data I
hae eer seen strongly indicates that gien the choice between registering in a purchase
process or simply walking away and inding a dierent site to purchase rom, the majority o
isitors will walk away.

1o continue beating a dead horse-which is strangely necessary since nearly eery \eb site I
try to purchase rom requires registration-consider this. Imagine that you owned a grocery
store and that you were really good at initing people in, proiding well-lit aisles, a wide


ariety o products at competitie prices, well-oiled grocery carts and helpul employees that
encourage you to ill your shopping cart. Requiring registration on your \eb site is just like
letting each o these ull shopping carts wait in line, only to be met by a cashier telling them
beore I can check you out I irst need you to ill out this page o inormation.` It would
neer work! Customers would be labbergasted. Carts would be abandoned where they
stood. Customers would ow neer to return,` and so on.

Crazy, huh

MeusurIng ConversIon 1qq

So what can you do \ou want ,or need, to collect certain types o inormation rom your
isitors but it sounds like I`m telling you not to. Not true. \hat I`m telling you is to do it
elegantly. Consider what Amazon.com actually tells people when they begin the checkout
process ,ligure 51,-you`ll create a password later.` 1his is an elegant, behind-the-scenes
way to collect the inormation they need to process the transaction ava get the isitor to
register. 1he ery next step ater you proide your email address is to proide shipping
inormation, which is exactly what the potential customer expects-I click checkout` and
then proide my shipping inormation.



Iigure SJ: Amazon.com's method for starting a checkout for new visitors. All that is
required is an email address and you are told that eventually they'll want you to add a
password.

lere are three things I recommend you consider regarding registration and the purchase
process:

J. Don't require registration, period.

Consider the popularity o Google and the tools they proide or the Internet. One o these
tools is the Google 1oolbar ,toolbar.google.com,. 1he Google 1oolbar is an ingenious
application that lies inside your browser that allows search rom your browser, blocks
annoying pop-up windows, and, germane to this discussion, proides an easy way to
automatically ill in most orms. As long as you ensure that your orms are Google 1oolbar
compliant,` that is, all o the ields are named such that the toolbar will be able to auto-
complete them, then you can rest assured that your isitors hae handled the how to
simpliy the next purchase` question.



2. Don't require registration, but provide an optional registration process that
visitors can opt into.

By simply allowing or anonymous checkout you proide the simplest process possible.
Visitor clicks checkout` and you ask them where they want the product shipped-no email
address, no password, nothing. I you`re still dead-set on getting the isitor registered,
proide them an optional path and bribe them. Oer the isitor 10 o` or ree
shipping` on their vet oraer i they take the time to register with you. I you do this,
consider haing a slightly more inoled registration process and ask them eerything you
oo CIupLer 1

would eer want to know. loweer, make sure you do this after you hae processed the
current transaction.

3. Register the customer silently during the checkout process and send them
an email telling them that you've done this and how to take advantage of it.

1his is perhaps my personal aorite because it is the best o both worlds. 1he customer
gets to check out quickly and easily and the marketing group gets to eel like they are
creating loyal ,registered, customers. I you just create a temporary password or the isitor
and then send them an email telling them how to change the password and the adantages o
doing so you will likely get some percentage o isitors actually changing their password and
some larger percentage o isitors at least coming back to your site, thus creating additional
opportunities to sell to them.
Think about Your Buttons
Make sure that your buttons are easy to see, easy to read and that the most important button
stands out like a sore thumb! 1his may seem, well, obious, but you don`t hae to look ery
hard at the Internet to ind examples o \eb sites ailing to ollow this guidance. Oten the
culprit is the graphic designer, wanting eery element on the page to blend together in isual
harmony, showing o their imagistic prowess. 1he problem is that most o your customers
are not graphic designers, they are most likely regular people who are simply trying to igure
out what to do next. Consider the two buttons shown in ligure 52:





Iigure S2: Lxample buttons from an anonymous Luropean Web site (left) and
BackcountryStore.com (right).


Most people will agree that the Add to Shopping Cart` button ound on
BackcountryStore.com is much easier to see than the Add to basket` ound on an
anonymous Luropean shopping site. Len in the context o the entire page the Add to
Shopping Cart` button stands out and is an easily seen next step` ,ligure 53,.


MeusurIng ConversIon o1


Iigure S3: Whole screen from BackcountryStore.com, note the Add to Shopping
Cart button on the right side of the page. Note also the Send Lmail button near
the lower-right portion of the screen, consistent in look but less pronounced.

I you do a good job at designing buttons you will easily be able to create a small palette o
options with optional, required and reatt, required button types. \our graphic designers may
cry and complain about but adding buttons that big and ugly will positiely ruin the
aesthetic appeal o the entire page and undo all o the work that I`e done!` I this is the
response you get, I highly recommend that you smile, nod and go ind graphic designers
who are as committed to your businesses success as they are to their portolio.
Include a Progress Indictor
Progress indicators in the checkout process are just one o those things that you really hae
to hae to make isitors eel comortable, at least the irst time they purchase. 1he problem
with progress indicators is that some o them lie. \ou may see a three-step process but not
realize that there are three sub-steps at each step and actually you`re about to engage in a
nine-step process. \hen you do this to a isitor you are essentially creating doubt in their
minds about your integrity, increasing the likelihood that they will bail out on the purchase
process prior to completion.



Because my amily taught me it`s better to say nothing at all i you hae nothing nice to say I
will rerain rom proiding examples o large, well-known online businesses that hae
extremely misleading progress indicators. Rather, I would reer the reader to ligure 54
which has two examples o nice-looking and accurate progress indicators rom
BackcountryStore.com and Amazon.com.

o CIupLer 1





Iigure S4: Progress indicators from BackcountryStore.com (top) and Amazon.com
(bottom), both of which are easy to read and are accurate in terms of describing the
number of pages you'll have to get though to complete your purchase.

lor an excellent example o how the purchase process .bovta work or any \eb site selling
durable goods, check out what the olks at BackcountryStore.com are doing
,www.backcountrystore.com,. Additionally, Bryan Lisenberg`s book, 1be Mar/eter`. Covvov
ev.e Cviae to Covrertivg !eb ite 1raffic, which was not aailable in its inal orm as o the
writing o this book, has a more complete list o the low hanging ruit` and is deinitely
worth reading. Look or it at Mr. Lisenberg`s \eb site, www.uturenowinc.com.

Key Performance Indicators Recommended for
Measuring Conversion
1he key perormance indicators discussed in this chapter are some o the most important
you will measure and monitor, i you choose to do so. Because conersion rates are easy to
measure and relatiely easy to inluence, the online business should watch these numbers
ery closely and make sure there is a response plan in place should key conersion rates dip
suddenly and dramatically. \hich conersion rates you watch depends entirely on what kind
o \eb site you hae and which multiple-step processes are most critical to your isitors`
success.

\ith that in mind, key indicators that should be tracked regardless o your particular
business model include:

Conersion rates or any processes that can make or sae you money or that are
critical to your customer`s experience on your \eb site
Campaign conersion rate or any actie campaigns, or i you hae too many to
reasonably watch, or your most expensie actie campaigns
Cost-per-conersion or the campaigns you hae decided to track
Segment conersion rates or key or critical segment conersions

Additionally, i you are selling online you should also be watching the ollowing:

\our site-wide conersion rate, all purchases to all isits or isitors
New and repeat site-wide customer conersion rates
Percent o orders rom new and returning customers
MeusurIng ConversIon o

Site-wide aerage order alue and that or new and returning customers
Sales per isitor, or comparison to your site-wide conersion rate

Again, keep in mind that it is not the numbers but changes in these numbers that you are
watching or.
oq CIupLer 1q

CHAPTIR 1q
MIASLRING RITINTION


1o get repeat business you have to do more than satisfy your customers, you have
to delight them. If you do, they'll not only come back, they'll recommend your site
to colleagues, friends and families. 1he result: more sales with zero additional
marketing expense.
Lisenberg and Novo, 1he Guide to Web Analytics


It is a rare, rare \eb site that would go through all the work described in the last three
chapters to reach, acquire and conert isitors and not want these isitors to remain loyal
and return to conert again. \hile your particular \eb site may not be actiely running
campaigns, hae clear conersion goals or sell anything online, the desire to hae isitors
return is nearly uniersal. \hether it is studies that tell us that current customers are much
easier to sell to and more proitable or just the good eeling you get when someone takes the
time to write you and tell you they enjoy your \eb site and will tell their riends about you-
and this does happen-customer retention is important to any \eb site.

1hat being said, retention is easily the most diicult o all the metrics described in this book
to measure. Because eery business model is dierent, and eery isitor is roughly dierent
,barring the concept o the aerage isitor` described in Chapter 10,, the span o time oer
which isitor retention should be measured can ary rom days to months to years.
Consider two examples, one site that proides news and another that sells cars. 1he irst site
will likely be able to measure isitor retention oer the span o a month-it is likely that a
isitor that had a good experience getting their news at that site will return at least once a
month to read more news. 1he latter may not see a isitor who researched the buying
decision or a car on their site again or three to ie years, i eer-easily longer than the
aerage liespan o a \eb analytics product on any gien \eb site.

1here will always be attrition and churn, this is just a act o lie. Len i you are the best at
what you do online today that may not be the case tomorrow. In a world where your
competition is only a mouse-click away retention can be diicult, i not nearly impossible.
\hen you are making the measurements listed in this chapter be sure and A, be reasonable
about the amount o time you gie isitors to return to your site and B, use the loyalty,
requency and recency measurements described in this chapter to help you hone and reine
that period o time.
MeusurIng ReLenLIon og

Metrics to Help Measure Retention
1he ollowing metrics and ratios will help you better understand how good a job you do at
isitor retention. O course, ideally you would hae a consistent unique user identiier
,UUID, written to eery isitor`s browser so you could track them personally when they
return, but just in case you are not that ar these metrics will help you deelop at least a
picture o how good a job you do.
Number of Returning Visitors
Just like the measurement o the number o new isitors described in Chapter 11, measuring
the number o returning isitors in a gien timerame will help you understand how well
your retention marketing programs are doing. \hile the raw numbers can be misleading i
not compared to oerall traic olumes, the number o returning isitors is an important
metric to understand.
How to Calculate
Most, i not all, \eb analytics packages will measure some kind o returning isitor metric
or you automatically. 1he returning isitor measurement usually depends on a timerame,
such as daily returning isitors,` weekly returning isitors` or monthly returning isitors,`
to put their return in context. I you are interested in how many times a isitor is likely to
return you should see the requency o return` metric described in this chapter. I you are
interested in the percentage o returning isitors, simply diide the number o returning
isitors into the total number o unique isitors in the timerame.
Dependence
Getting an accurate number or the returning isitor measurement depends on haing a
system to identiy that a isitor has been to the \eb site preiously. \ou should employ a
system that uses cookies to keep track o returning isitors, as IP addresses are subject to
change as how your isitors connect to the Internet aries widely or mobile users or
business traelers. Obiously a system built on cookies is not oolproo ,see disadantages
o page-tagging data collection described in Chapter 3,, but cookies are the standard or
keeping track o isitors and will likely continue to be so or some time.
Usage
1he number o returning isitors should, in general, be used in conjunction with a ratio, such
as acquisition mode,` or the ratio o returning isitors to all isitors, in order to put the
number in context. It would be great to know that your \eb site had 100,000 returning
isitors last month unless you had 100,500 total isitors to the site, in which case you
acquired nearly no new isitors, or unless you had 100,000,000 total isitors to the site, in
which case those 100,000 returning isitors was a relatiely small group.
o6 CIupLer 1q

Interpretation
1he number o returning isitors or a timerame is exactly that, the number o isitors who
had been to your site at least one time preiously and returned in the timerame you are
looking at. I your analytics package oers daily,` weekly` and monthly` returning
isitors, as many do, then 100 daily returning isitors on Monday` would tell you that 100
isitors came to your site on Monday and all o these people had been to your site at least
one time beore.

I your analytics application supports returning isitors or odd timerames this can be ery
helpul to help you understand how isitors return during eents, such as the NlL ootball
season or between 1hanksgiing and Christmas, or whether you are likely to get more
returning isitors on weekdays ersus weekends. Consult with your endor to see i this is
possible with your speciic package.
Example
1he normal pattern o monthly returning isitors to the BackcountryStore.com \eb site is
loaded more heaily toward isits earlier in the month ,ligure 55,.



Iigure SS: Monthly returning visitor pattern for BackcountryStore.com. Note the
spike in returning visitors mid-month, attributed to retention marketing activities.

1he interesting piece o inormation in ligure 55 is the second spike near the middle o the
month. 1his spike is tied directly to BackcountryStore.com retention marketing actiities
and this graph proides quick isual eidence that the campaign had some success. O
course, to truly determine the success o the campaign, BackcountryStore.com would need
to drill-down into its campaigns and commerce measurements to see how many o these
returning isitors actually made purchases.
Average Frequency of Returning Visitors
Once you hae learned how your analytics package presents returning isitors, you are aced
with haing to make sense o what this metric is telling you. As with reach, acquisition and
MeusurIng ReLenLIon o,

conersion, many retention metrics are ratios and percentages. 1he ratio o daily to monthly
returning isitors will gie you an idea o the aerage number o times isitors return in a
month and how that number changes month-oer month.
How to Calculate
As long as your analytics package proides you with daily and monthly returning isitors this
ratio is simply:

DAILY RETURNING VISITORS / MONTHLY RETURNING VISITORS = RATIO OF
DAILY TO MONTHLY RETURNING VISITORS
Dependence
Calculating this ratio is only dependent on your \eb analytics package proiding you with
both daily and monthly returning isitors. I you application does not hae a monthly
returning isitor` speciically, but you can calculate de-duplicated returning isitors or
custom timerames you can simply generate the monthly metric on your own.
Usage
1he ratio is useul in understanding whether changes in your site or marketing aect the
aerage number o times isitors are likely to return in a gien month.
Interpretation
1his number represents the aerage number o days in the month that isitors are likely to
return, so i the ratio calculated is 4.5` this tells you that the aerage isitor returns to the
site oer more than times. Keep in mind the concept o the aerage isitor` described in
Chapter 10-not all isitors are aerage but when you hae enough o them the requency o
return isits is belieed to hae a normal distribution.

I yours is a business model that depends on haing isitors return with increasing
requency-as is the case with the online commerce model as isitors build purchase
momentum, or i you proide news and inormation as isitors build brand loyalty-the
higher this number the better. Conersely, i your site proides some kind o product
support higher numbers could indicate that either isitors are haing increasing diiculty
with your products or simply that they are not inding the support inormation they are
looking or and so hae to return to the site repeatedly.
Related metrics
1his measurement is closely related to the aerage requency o return or retained`
isitors, described in this chapter under Visitor Segmentation Metrics or Retained
Visitors.`
oS CIupLer 1q

Ratio of Returning Visitors to All Visitors
Much like the calculation o percent new isitors ,new isitor reach`, calculated in Chapter
11, the ratio o returning isitors to all isitors is an easy-to-read metric describing how you
are doing at retaining isitors. 1he percent o returning isitors is an excellent retention KPI
or any \eb site.
How to Calculate
Percent returning isitors is calculated easily:

RETURNING VISITORS / ALL VISITORS = PERCENT RETURNING VISITORS
Dependence
Making this calculation properly depends on haing numbers rom identical timerames and
haing de-duplicated numbers or each. I the longest timerame or which you can get
unique returning isitors and unique isitors is a month, then a month is the longest period
o time or which you should calculate percent returning isitors.
Usage
Monitoring the percent o returning isitors on a regular basis can help you understand i
something has changed about your isitors or your retention marketing, in which case more
research is warranted.
Interpretation
As an example, i 40 percent o your monthly isitors are returning isitors or three months
in a row and then suddenly that number decreases to 20 percent then one o two things has
happened. Lither you hae suddenly increased your new isitor acquisition programs or you
hae experienced signiicant attrition in your already conerted audience. In the irst case
you simply need to be sure you understand which marketing programs are responsible or
the sudden increase in new isitor acquisition, in the latter you need to attempt to
understand what might hae changed in the relationship with your existing audience.
Loyalty metrics
Loyalty on the Internet is a measurement o how many times a isitor returns to your \eb
site oer time. 1ypically measured in number o isits per isitor,` loyalty metrics proided
by most \eb analytics applications simply inorm you on the number o times isitors hae
returned since you irst implemented your particular analytics solution. Some categories o
loyalty measurement make good retention perormance indicators.
MeusurIng ReLenLIon oq

How to Calculate
lortunately, most analytics applications measure and present loyalty automatically, remoing
the need or you to count and aggregate isits by isitors oer time. \ou should consult
your \eb analytics endor or their documentation to understand how your application
reports on loyalty metrics.
Dependence
1he major dependences or loyalty are A, your particular application`s ability to measure
loyalty accurately and B, the length o time you hae been using your current application.
Because the loyalty measurement nearly always depends on a cookie or determining
uniqueness, the measurement also depends on using a single application consistently oer
time. So i you change your \eb analytics application then you need to start oer again in
making the measurement.

1his perhaps makes more sense i you consider that when you implement a new analytics
application eery isitor is a new isitor` with their irst return. Len i they hae been
coming to the \eb site eery day or a year, once you change measurement techniques they
will be counted as a new isitor ,see the deinition o new isitors in Chapter 4 and about
their use in Chapters 11 and 12,. \ith this in mind, it is a good idea to pay only cursory
attention to loyalty and requency measurements until you are conident your preiously
retained isitors hae had enough chance to return to the \eb site at least once ater you
implement the new solution, so that they hae been counted as new.`
Usage
Loyalty measurements are used to better understand how good a job you do oerall in
retaining isitors. 1he major groupings in any loyalty measurement are 1
st
isit` and more
than one isit` with the ormer grouping describing all new isitors and the latter describing
all retained` isitors. By understanding how the percentage o new ersus returning isitors
change oer time, especially in groups that may be considered sweet spots` or your online
business, you will be able to better understand how well you retain isitors.
Interpretation
Again, most \eb analytics packages present loyalty as the number o isitors coming to your
\eb site or the n
th
isit in some type o table ,ligure 56,.


1o CIupLer 1q


Iigure S6: Loyalty measurement taken from BackcountryStore.com showing the
number and percentage of visitors who come to the Web site n times.

1he example in ligure 56 shows that 14,41 or 4.08 percent o all isitors returned or the
third time in the timerame in question ,row highlighted in blue,. 1his is interesting to the
olks at BackcountryStore.com in that they can see that the majority o isitors who came to
their \eb site in this timerame were new isitors ,coming or the 1
st
isit, and that ater that
the majority o returning isitors had returned ie or less times ,roughly 18.5 percent o the
total number o site isitors,.

Again, as is a recurring theme throughout !eb .vat,tic. Dev,.tifiea, it is not necessarily these
numbers that are interesting but how these numbers change oer time.
BackcountryStore.com would be smart to keep track o the percentage o isitors who return
between two and ie times month-oer-month and watch or increases or decreases in this
number. It may choose to build a key perormance indicator out o the number o 1
st
isit`
isitors and the aggregate o 2
nd
through 5
th
time` isitors to monitor or any large changes.
Frequency of Visit
Closely related the loyalty o isitors you reach, acquire and conert is the requency at
which they return to your site throughout this process and then ater they hae been
conerted. Speciic business models drie the ideal requency o return or isitors, online
stores typically desire that isitors return and purchase monthly whereas sotware endors
may want isitors to return and upgrade their applications once yearly. 1he ideal requency
o return or your \eb site will be determined ultimately by what you are trying to get your
audience to do and how you are trying to get them to do it.
How to Calculate
Much like loyalty and recency, requency is most oten calculated or you by your analytics
package. Most oten requency is calculated as the aerage number o days between isits`
and is calculated or isitors returning in the timerame in question but considers all lietime
isits ,lietime` meaning since you implemented the application you are currently using`,.

MeusurIng ReLenLIon 11

1he calculation is typically something like:

(DAYS BETWEEN 1st AND 2nd VISITS) + (DAYS BETWEEN 2nd AND 3rd
VISITS) + (DAYS BETWEEN N - 1th AND Nth VISITS) / (N -1) =
FREQUENCY OF VISIT

So you sum all o the numbers o days between isits or the lietime o measured isits and
then diide by the number o isits minus 1` to get the aerage number o days between
isits. \ou will be looking or the package to report back to you the number o isitors who
return in a timerame who hae an aerage number o days between isits` o n` days ,or
hal-days,.
Dependence
lrequency has the same dependence as loyalty, your application`s ability to accurately
measure the number o days between isits as well as the amount o time you hae used your
current application. As with loyalty it is a good idea to gie your isitors time to establish
their retention patterns beore placing too much emphasis on what requency measurements
are telling you. 1he author typically recommends waiting six months to measure requency
o isit.
Usage
Once your requency measurements hae had a chance to normalize, these numbers can be
used much like loyalty, to better understand how you connect with your audience. Consider
the data rom BackcountryStore.com, presented in ligure 5, demonstrating graphically the
requency o return or isitors.



Iigure S7: Lxample of average frequency of return for BackcountryStore.com.

As you can see in the igure most isitors are returning roughly eery three weeks ,rom 0.5
to 21.5 days between isits,. It should be noted that the largest single category is or oer
90 days` as represented by the large red bar on the right.
Interpretation
Aerage requency o isit can help you to understand how oten you should expect isitors
to return to your \eb site gien all the marketing you do. In general you want to look or
1 CIupLer 1q

groups o returning isitors, the percent o isitors who return, on aerage, eery seen days
,weekly on aerage, or less ersus the percent that return between seen and 14 days ,eery
other week on aerage,. Knowing at which requency the majority o your retained audience
is returning is ery helpul in planning or marketing actiity and site-wide changes.

One useul example is that i you determine that the majority o your isitors are returning
eery two weeks on aerage you should consider updating your home page at least eery two
weeks. One common method or keeping isitors engaged is to keep presenting new
inormation that they may ind compelling. By the same token i you hae allocated
resources to make changes to your home page on a daily basis, but you can determine that
only a small percentage o your audience returns daily, you may sae money by scaling back
the requency with which you update your \eb site.
Example
As you can see in ligure 5, BackcountryStore.com can reasonably expect that most isitors
will return at least once a month but that a large part o their audience will return perhaps
only two or three times a year. 1his helps the BackcountryStore.com marketing group plan
or outbound marketing programs like opt-in email campaigns, bi-annual storewide sales and
timing o catalog printing and mailing.
Recency of Visit
Recency is closely related to requency in that it is about the time that elapses between isits
to the \eb site. 1he dierence between requency and recency, howeer, is that recency is
the measured amount o time elapsed since the last isit a isitor made to the \eb site. So i
your application reports that 1,000 isitors had a recency o 3 days,` then you know that
these 1,000 isitors had been to your site three days ago and hae returned on the day under
examination.
How to Calculate
As with loyalty and requency, recency is a metric that is calculated and reported
automatically by your analytics application. It is unclear to the author how many o the
currently aailable \eb analytics oerings support recency measurement but it is likely that
oer time this will become a widely supported metric. One aluable reason or supporting
recency in a \eb analytics application is the online business` desire to be able to segment
one`s customer base using the recency, requency, monetary alue ,RlM, model. lor a good
explanation o the RlM model see either Jim Sterne`s book on !eb Metric., or this article at
ClickZ by Mark Sakalosky: http:,,www.clickz.com,res,analyze_data,article.php,961901.

I you are selling products online, it is best i your analytics application supports recency
measurements ,and requency and loyalty or that matter, rom both a non-commerce and
commerce perspectie.
MeusurIng ReLenLIon 1

Dependence
Recency has the same dependence as loyalty and requency, albeit slightly less so. Because
recency is only dealing with the current and preious isits a isitor made to your site you
can likely begin using recency measurements much more quickly than those or requency
and loyalty.
Usage
According to Jim Noo, recency is the most powerul predictor o uture behaior, based on
the reasonable assumption that the more recently a isitor or customer has done
something` on your \eb site, the more likely they are to return and do that thing again.
\ith this in mind recency can be a aluable ally, especially i you are able to determine the
recency or dierent isitor segments and campaign respondents. 1he greater the number
o recent isitors on your \eb site, the higher the likelihood that actions will occur in the
short term. I you hae an online business, and the actions` are purchases, you can see
where recency can be a aluable metric.

I you are selling products on your \eb site you may want to establish recency KPIs to keep
track o the percentage o isitors who had isited in the last three days. 1heoretically as
that percentage increases your total online reenue is increasing as well.
Interpretation
Recency is easily interpreted. I your application reports that 500 isitors hae a recency o
20 days` you are being told that these 500 isitors had isited your site 20 days ago.
Relating recency into an RlM model, or perorming useul segmentations based on recency
measurements, can be slightly more diicult. 1he author recommends reading what others
in the \eb analytics space, such as Jim Sterne, Jim Noo and Mark Sakalosky, hae to say
on using recency.
Visitor Segmentation Metrics for Retained Visitors
One o the most practical strategies or learning about your retained isitors is to segment
them as such-as haing been successully acquired and,or conerted-so that you can
monitor their actiity as they return to your site. Knowing why isitors come back is
paramount in successully conerting and retaining. Being able to see where retained isitors
are coming rom, what pages or content they are consuming and what actiities they are
engaging in will allow you to create better content and naigation systems or this group.
1he ollowing are metrics that you should be looking at i you are able to segment isitors as
retained` using your particular analytics package.
1q CIupLer 1q

Activity of Retained Visitors
1he actiity o retained isitors is really a group o measurements similar to those we`e
looked at in preious chapters. Metrics like page iews, isits and isitors will help you to
determine what percentage o your total audience is retained and how much content they are
likely to consume.

1he typical metrics that are important to watch or retained isitors include:

Number o page iews
Page iew distribution by day o week
Number o isits
Number o isitors
Aerage time spent on the \eb site per isit
lrequency and recency o return to the site
Any campaigns that they may be responding to
Any commerce actiity they may be engaged in
1he pages and content that they are browsing

It is important to dierentiate between returning` isitors and retained` isitors, i
possible. A returning isitor, as measured by most \eb analytics applications, is simply a
isitor who has been to the \eb site at least one time preiously-this measurement has no
regard to the actiities they engaged in during preious isits. But a retained isitor is
someone who has completed an actiity on your \eb site and continues to return to the site.
It is this distinction that dictates that you segment these retained isitors and measure them
as separate rom all returning isitors.

1he interpretation o all o the aorementioned metrics diers greatly by business model,
some businesses, such as those selling products online, desire retained isitors return with
great requency. Other businesses expect that retained isitors will only come back
sporadically, especially i the retention piece is technical support or a product. \ou should
create a series o perormance indicators around the actiity o retained isitors and set
thresholds as appropriate to your online business.
Views of Key Pages and Contents
One o the most practical isitor retention metrics to watch is which pages and content
groups this segment o isitors is interested in and how these page iews dier rom those
or your global and unconerted` audience. By keeping track o the sections o your \eb
site that this isitor segment is isiting most oten you will be able to learn a great deal about
their speciic interests. Knowing about retained isitor interests will enable you to modiy
your \eb site and marketing materials to better speak to this class o isitor.
MeusurIng ReLenLIon 1g

Example
As an example, the olks at BackcountryStore.com keep track o their customer` isitor
segment, people who hae purchased with them at some time in the past and that they
consider retained when they return to the \eb site. Based on this data, they are able to mine
the most popular pages or both the retained and un-retained segments and look or
dierences ,ligure 58,.





Iigure S8: Page activity for BackcountryStore.com's retained customer segment
(left) and all visitors segment (right).

As you can see, there are tangible dierences in preerences or the retained isitor segment
when compared to that o all isitors. 1he irst thing you notice is that retained isitors are
logging in ,,Grow Community,Customer Login`, and tracking their purchases ,,Lducate
Customers,UPS 1racking`,, actiity that BackcountryStore.com expects.

One thing they do not necessarily expect is that the general audience is interested in skiing
and snowboarding ,,Sell Products,Ski,Ski` and ,Sell
Products,Snowboard,Snowboard`, but retained isitors are not. 1his inormation
potentially has an impact on BackcountryStore.com`s marketing and merchandising,
suggesting that isitors are initially attracted to the \eb site looking or ski and
snowboarding gear but that attraction is not translating well in terms o repeat purchases.
Len i BackcountryStore.com does not make any changes to its outward marketing and
messaging, haing this inormation helps to set expectations internally in regards to the
likelihood that retained customers will be purchasing expensie skis and snowboards during
repeat isits.
Retained Visitor Conversion Rate
1here are a number o studies outlining how much less expensie it is to market to existing
customers, although these numbers seem to be changing all the time. It is, howeer, not a
great stretch to imagine that your likelihood o success in getting someone who has already
conerted or purchased to repeat the process is high, especially i the isitor had a pleasant
experience the irst time through. Most online businesses that measure the dierence in
conersion rates or new and retained customers see radically dierent numbers-one site
that the author has had the pleasure o working with in the past conerted new isitors at a
rate o roughly 10 percent when the isitor added an item to the shopping cart. 1his same
site had an 80 percent conersion rate or returning customers adding an item to the cart
16 CIupLer 1q

,see a note about why repeat and retained customers purchase at a higher rate than new
isitors,` below,.

Measuring your retained isitor conersion rates, the approximation or how likely retained
isitors are to repeat and conert again, is a key retention perormance indicator. Much like
the return customer conersion rate` described in Chapter 13, understanding the impact
retained isitors hae on conersion is ery important to deeloping and reining your
marketing programs. 1his rate is an important retention perormance indicator i you are
trying to get retained isitors to conert again.
How to Calculate
1his metric is calculated the same way as a normal site-wide conersion rate ,orders ,
isitors,, only or retained customers or isitors:

RETAINED VISITOR CONVERSIONS / RETAINED VISITORS = RETAINED VISITOR
CONVERSION RATE
Dependence
It is important when making these calculations to ensure that you hae accurate
measurements or both the retained isitor conersion and retained isitors components.
Any mixing will necessarily aect these numbers and potentially skew one calculation or the
other.
Usage
Being able to determine the dierences in how retained isitors conert through any
important actiity on your \eb site will help you make determinations regarding how much
emphasis to place on retained isitors in your \eb site design and outbound marketing
programs. 1he dierences in conersion or new and retained isitors, along with your
acquisition costs will help to drie your oerall marketing strategy. As an example, i you are
able to determine that while retained isitors are much less expensie to market to they also
hae a much lower conersion rate through key processes, you are unlikely to spend as much
money on retention marketing as you would i you determined that retained isitors
conerted at a rate our times the new isitor conersion rate.
Interpretation
As long as you are careul in making your calculation ,see Dependence` aboe, then this
measurement is a simple conersion rate that can be compared to other conersion rates on
your \eb site.
MeusurIng ReLenLIon 1,

Customer Retention Rate
1he customer retention rate is, according to Lisenberg and Noo, the ability o your site to
generate repeat business. A simple calculation and important retention perormance
indicator, the customer retention rate is a must measure` number or any online business.
How to Calculate
lor an online business, the customer retention rate is simply the number o repeat customers
diided by total customers or a timerame.

REPEAT CUSTOMERS / ALL CUSTOMERS = CUSTOMER RETENTION RATE

1his rate can be extended easily or any retained isitor and used to generate a isitor
retention rate.`:

RETAINED VISITORS / ALL VISITORS = VISITOR RETENTION RATE

\ou should note that the second calculation is simply the percent returning isitors`
described elsewhere in this chapter.
Dependence
1hese calculations depend only on your application`s ability to supply the necessary metrics,
either ia commerce measurement or isitor segmentation tools as described earlier in this
book.
Usage
1he customer retention rate is used to determine how good a job you are doing at attracting
repeat business or repeat conersions i that is a goal o your business. lopeully you are
able to establish a baseline or this metric and then use the continuous improement process
to slowly but surely increase your customer retention.
Interpretation
Lisenberg and Noo comment that to get repeat business you hae to do more than satisy
your customers, you hae to delight them.` I you subscribe to this line o thinking, as does
the author, you can think o your customer retention rate as a measure o delight`
expressed by your online isitors. Lisenberg and Noo go on to point out that a key actor
or an online business to be successul and proitable is to increase the customer retention
rate. 1his, in tandem with the notion that existing customers are less expensie to market to
than new customers, will eectiely lower your customer acquisition costs ,Chapter 13,.
1S CIupLer 1q

Average Frequency of Return for Retained Visitors
One aluable metric or understanding how eectie your site is at isitor and customer
retention is the requency o return or this online isitor segment. Some \eb analytics
packages will calculate requency o return or isitor segments automatically, others will
proide requency metrics or customers` speciically ,those isitors who hae placed
orders online,. I, or some reason your application does not support this measurement, but
does proide unique isitor measurements or segments, you will still be able to calculate an
aerage requency o return.
How to Calculate
Proiding your analytics package proides at least daily and monthly unique isitor metrics
or segmented isitors you can simply diide the number o daily isitors into the number o
monthly isitors to get an approximation o the aerage number o isitors or isits per
month.

SUM OF DAILY VISITORS / MONTHLY VISITORS = AVERAGE VISITS PER MONTH
FOR THE SEGMENT
Dependence
1his metric only depends on your application`s ability to proide the sum o daily or weekly
isitors as well as monthly unique isitors or a gien isitor segment.
Usage
1he resulting calculation will help you to understand whether your retained isitors ,or any
isitor segment or that matter, are returning to your \eb site once or seeral times a month
on aerage. Depending on your business model you may desire that retained isitors return
seeral times a month, perhaps to consider additional purchases or to interact with a isitor
community.
Interpretation
1he resulting calculation will tell you essentially the number o days in a month the aerage
retained isitor is returning to your \eb site. Say, or example, that the sum o daily isitors
or the retained segment was 1,000 and the measurement or the monthly isitors or the
same segment was 250. 1his tells you that 250 unique isitors came to the site in the month,
on aerage, our times ,1,000 daily isitors , 250 monthly isitors ~ 4 days,month,.

Again, based on your particular business model, higher or lower numbers are better.
Depending on how important isitor retention is to you, you may want to consider making
this measurement part o your retention perormance indicators and watching or dramatic
changes in the number.
MeusurIng ReLenLIon 1q

Related Metrics
1his measurement is related to the generalized requency measurement discussed in this
chapter but, whereas requency is generally reported directly by \eb analytics packages, you
should keep in mind that this metric is a calculation o the aerage requency o return oer
a shorter timerame.

Web Analytics Tip: How to Retain Customers
Ater your online business has gone through all the trouble o getting a customer to make
the irst purchase you then need to actually retain that customer and get them to come back
and purchase again. 1his chapter contains all the great metrics you can use to measure
retention but doesn`t really address how to get them to return and purchase again. \ou need
to keep in mind that the most critical actor in making a subsequent purchase is the act o
making the irst purchase.

In my opinion, the decision to make a irst purchase depends on a handul o both tangible
and intangible actors. 1hese actors include:

low your prices compare to those o your competition
low well you market your \eb site on and o the Internet
low well your company or brand is known in the marketplace
low good a job you do at actually moing isitors to the irst purchase
\hat others say about you ,BizRate, ePinions, Pricegrabber, and so on,

As you can see, most o these actors are within your control but some are more malleable
than others. \ou can probably lower your prices, but most proitable online businesses
would rather not compete on price. \ou can spend more money on marketing, but
marketing budgets are typically inite and your competition is oten able to spend more.
\ou can hae a perectly designed purchase process, but neer moe eery isitor who starts
a shopping cart through to purchase. \ou can do the best job possible at getting products to
people, and still some people will hae problems and complain. It neer ends.

Proided, then, that you are able to get a isitor to make the irst purchase, you now hae a
great opportunity. I you are able to absolutely delight your new customer you will hae
likely gained a customer or lie.

Some o the things that delight customers include:

Products that are promptly shipped and that arrie on time
Products that show up as ordered`
lollow-up mailings or emails to make sure that the product was shipped properly
and that the customer is satisied
Coupons or uture purchases that magically appear shortly ater the irst purchase
Courteous and prompt customer serice i there is anything wrong with the order,
which is ollowed up by quick resolution to the problem
o CIupLer 1q


All o these items are essentially Duh! O course we do that .` items, but the reality is that
this is not always the case. I eery online business did eerything right there would be no
need or BizRate to help potential customers ensure that you`re a merchant they want to do
business with ,www.bizrate.com,.

Regarding the eents that occur ater the isitor has clicked the process my order` button
and committed to making the purchase, it is my strong opinion that you cannot oerdo a
commitment to excellence. Making sure that the product will ship on time and contacting
the customer proactiely i it will not, proiding your customers the ability to track their
purchase ater it has been shipped, making sure you hae riendly, knowledgeable support
sta in case your customers hae questions about the product and gladly taking any returns
without question should be the bare minimum. Going the extra mile and sending coupons
or discounts, ollowing up to ensure that the product was well-receied and is eerything the
customer hoped and more-actiities like this will only reinorce in your customer`s mind
that they made the right decision.

\ou may be saying to yoursel, \eah, that`s easy or you to say but much harder to do.` I
agree, not all businesses will be able to delight their customers ater the irst purchase has
been made, and I simply assume that oer time these businesses will ail, clearing the
competitie landscape or those companies that are willing to work to satisy and retain their
customers. Now you may be saying, larsh. 1his guy is harsh!` And you`re right, I am
harsh. So is your competition and they are always only a click away.

Key Performance Indicators Recommended for
Measuring Retention
Not eeryone reading !eb .vat,tic. Dev,.tifiea is necessarily ocused on isitor and customer
retention in a traditional sense. Some business models don`t necessarily expect a customer
to make a second purchase in the near uture ,car dealers, real-estate \eb sites, some
subscription-based content sites,. Still, ery ew \eb sites are designed to only hae a isitor
come to the \eb site a single time, neer to return.

\ith that in mind, key retention indicators that should be tracked regardless o your
particular business model include:

1he ratio o daily to monthly returning isitors, as a quick measurement o the
aerage requency o return or all isitors. I you are tracking retained` isitors
you should then track this KPI or the retained isitor segment as well.
Percent returning isitors to ensure that your new,returning isitor mix is in line
with your new,retention marketing spend.
Loyalty measurements or groups o returning isitors such as 2
nd
through 5
th

return eer` to monitor or large changes in isitor loyalty.
Actiities o retained` isitors ,dierentiated rom returning` isitors using
isitor segmentation tools, including number o page iews, number o isits and
MeusurIng ReLenLIon 1

number o isitors. It is also worthwhile to track the aerage time these retained
isitors spend on your site.
Retained isitor conersion rate, which is important or all o the reasons that
conersion rates are important.
Customer retention rate, to keep track o how many customers or isitors you are
attracting back to your site.

CIupLer 1

CHAPTIR 1g
BRINGING IT ALL
TOGITHIR LSING KIY
PIRIORMANCI
INICATORS


Without clear goals, there's no need to measure anything, without measuring
there's no way to know if the work you're doing is helping to achieve your goals.
Jim Sterne in Advanced Lmail Marketing


Perhaps the most common complaint about \eb analytics data and the applications that
proide said data is that there is simply too much inormation`, too many graphs, too many
charts, too many options, too many ariables-too much or the aerage user to understand
and make use o. It is surprisingly likely that you will hear someone complain, I hae 400
reports to choose rom in my analytics package and I`m not sure which to use so I just don`t
use any o them .`

Counter this complaint with the plight o the \eb analytics endor and the need to present
att o the latest and greatest inormation to the customer because o the competitie nature
o business. Vendors that only hae a small number o well-presented data are oten
marginalized as niche players` in a larger market. \hile it would be ideal to hae small,
useul modules o inormation, perhaps compartmentalized to inorm well about a speciic
actiity o interest, none o the endors known to the author hae adopted this approach.

So, in order to ensure that you are able to make the most use o the inormation proided in
!eb .vat,tic. Dev,.tifiea, you should do two things. lirst, make good use o the continuous
improement process as described in Chapter 2. I you are able to ocus your goals or
change on your \eb site around a small number o changes, the metrics that support those
changes become much more reasonable and easy to manage. Instead o looking at
eerything, you can ocus on a speciic conersion rate or the process you are trying to
improe.

1he second strategy the author recommends is to make good use o key perormance
indicators ,KPIs,. Realistically, key perormance indicators are simply metrics or ratios, all
o which hae been described in the last ew chapters o this book. Conceptually, these
metrics are powerul and easy-to-understand indicators o A, how well your \eb site is
doing and B, whether anything has changed dramatically and needs additional attention.
BrIngIng L AII TogeLIer UsIng Key PerIormunce ndIcuLors

Key perormance indictors are usually built with comparisons o preious reporting periods
and percent change indicators to highlight any monitored metrics that hae signiicant
moement up or down, applying warnings i necessary ,ligure 59,.






Iigure S9: Samples of Key Performance Indicator (KPI) reports.
How Do I Build a KPI Worksheet?
lor the most part, all o your KPI monitoring can be done in a simple spreadsheet
application, such as Microsot Lxcel. Lxcel oers the opportunity to easily manipulate
numbers to create the necessary ratios, create graphs, generate warnings and is arguably the
world`s most popular analytics application. 1his popularity imparts a amiliarity and comort
to data presented in Lxcel spreadsheets and this comort oten promotes action-i people
understand the data and the ormat they are less likely to simply discard the data as
unreadable.` Lxcel also oers the ability to easily annotate data ia text or the notes`
unctionality, allowing you to improe communication about what the data means. 1he
author strongly recommends building your KPI reporting around Microsot Lxcel, or a
similar application.

1he basic idea you want to keep in mind when you are building a KPI worksheet is this:

1be vvvber. ov tbeir orv are te.. ivtere.tivg ava te.. ivforvatire tbav cbavge. iv tbe vvvber. orer tive.

Raw numbers are great and they can say a lot about the success o one`s \eb site. Len
some o the complex ratios presented throughout Chapters 11 through 14 o !eb .vat,tic.
Dev,.tifiea are interesting as simple indicators o success or a lack thereo. loweer, a key
perormance indicator worksheet is most aluable to the online business when it is
prompting action, and the greatest drier or action online is seeing that you are doing much
better or much worse than you were yesterday. I you are doing much better, you need to
understand what it was that you did that created that success, i you are doing much worse
you need to igure out what is going wrong and work to correct the problem.
q CIupLer 1


See ligure 60 or a general example o how the author builds KPI worksheets, emphasizing
the relationship between data rom this period` and last period.`



Iigure 60: General strategy for building KPI worksheets, highlighting data collected
in this period, last period and the percent change between the two. 1he data in
yellow is the raw data used to generate the KPI.

1he percent change` calculation is perhaps the most critical on any KPI worksheet as it
proides a isual representation o the change in the alues rom period to period. 1his
calculation is simply:

(THIS PERIOD LAST PERIOD)/LAST PERIOD = PERCENT CHANGE

Additionally, because Microsot Lxcel oers the ability to apply conditional ormatting to
alues, the author oten applies colors to the percent change alues depending on whether
they are moing in a positie or negatie direction. Keep in mind that negatie alues are
not always bad and should not, by deault, be colored red. In many instances the desired
change in a KPI is a decrease, hence a negatie alue, and so these instances should be
colored as appropriate.

lere are some speciics that the author recommends or building a KPI worksheet or your
online business:
Use the Right Measurements
As will be discussed in the next section, there is likely a core group o indicators that make
the most sense or your online business to be watching on a daily or weekly basis. \ou want
to try and ocus your energy on understanding and being able to act upon this group o
KPIs. 1hese numbers are intended to be driers, prompting you to drill down more deeply
into the primary data, not be end-all-be-all comprehensie reporting. Be careul about what
you put in these reports and make sure that the data is meaningul but not dense.
Meaningul data prompts people to ask related questions about the inormation contained in
the KPI, dense data prompts people to skip oer the KPI and not consider the inormation
it may hae contained.
BrIngIng L AII TogeLIer UsIng Key PerIormunce ndIcuLors g

Think about Presentation
Remember that in many instances the presentation is as, or more, important than the data
you present. 1his statement sounds odd but this is proen true oer and oer again, poorly
presented data will be oerlooked or mistrusted. \hen you build your KPI report the irst
thing you want to consider is what data will be included and what that data will be able to tell
you about your \eb site. 1he ery next thing you want to consider is how should we
present this data in a isually pleasant way that will not turn the intended audience o`
Simple things like headers that hae your company logo or colors, the use o lines and cell
color and shading, consistent use o onts and ont sizes are all important to how a report is
perceied by an audience.
Allow for Annotation
\hile it is an excellent idea to automate the deliery o KPI reports throughout the
organization, oten the numbers and their meaning can be conusing to users less inested`
in the data. 1he worst-case scenario is where a marked decrease is presented in the report,
causing top-to-bottom panic within the organization, when the decrease was expected by a
small group o indiiduals who simply orgot` to tell eeryone else.

I you proide a space in your reports, ideally at the top, or annotation and interpretation o
the data you will likely aoid panic situations. By routing the KPI report through a small
group o indiiduals who are ested in understanding the data and able to drill down on any
KPI that warrants more inestigation, the report can be annotated prior to wider deliery,
making the report more eicient when it is receied.
Automate Delivery
Key perormance indicator reporting or the online business is ery important and, in the
author`s opinion, the next big thing` in \eb analytics that will truly get businesses to act
upon the data that they are already collecting-as long as these reports can be generated in a
timely ashion and shared out to the organization. Oten the answer to the question, \hy
don`t you proide regular reporting to your entire organization on \eb analytic data` is, It
just took entirely too much time to copy and paste all the data rom the application into the
spreadsheet and we gae up.`

lortunately \eb analytics endors are waking up to the need or automated, customized
reporting in a ormat other than their particular application, and allowing or automated
reporting. Some endors oer customized l1ML reports that can be generated and
emailed out on a scheduled basis, others allow data to be exported directly to a SQL
enironment, where reports can be generated through internal tools and processes. linally,
at least one endor has created an application that allows data to be directly imported into
Microsot Lxcel or report generation, also allowing generated reports to be scheduled or
automation ia email. Regardless o which endor you use, i KPI reporting is important to
you ,and it certainly should be,, you should ask your endor how they enable this type o
reporting and how customizable and lexible data export ormats are.
6 CIupLer 1


Once you hae decided that you are going to build a KPI report and automate its deliery to
your organization, the next step is to decide which metrics listed in !eb .vat,tic. Dev,.tifiea
to include in your report. \hile the author is unable to personally help you decide exactly
which metrics to include, experience shows that you will be successul i you start with KPIs
that are most releant to you particular business model.
The Four Broad Business Models
\hile the author recognizes that eery online business is dierent, or at least eery online
business desires to be recognized as being dierent, experience shows that there are our
undamental online business models: Online commerce, adertising, lead generation and
customer support.
Online Commerce
1he online commerce model is perhaps one o the most well understood o the our
business models, simply because this is where the majority o the money is spent and made.
Any \eb site that sells products, serices or inormation its into this model and each o
these sites hae similar challenges. \hile the speciics o how online commerce works
diers slightly between sites selling products ,such as consumer electronics or books,, those
selling serices ,such as airline tickets or hotel rooms, and those selling inormation ,such as
news or stock inormation,, undamentally the challenges are the same-attract isitors to
the \eb site, show them the alue proposition and get them to conert. 1he online
commerce model is typiied by a ocus on all our stages o the customer lie cycle ,Chapter
10,, new customers need to be reached and acquired, the act o purchasing is the conersion
and obiously it is strongly desired that these customers are retained.
Advertising
1he adertising business model is one typiied by attracting isitors to a \eb site who will
read content and at the same time, theoretically, iew and respond to adertising. News and
sports \eb sites, such as CBSNews.com, CNN.com and LSPN.com all hae this business
model. 1he basis or making money on adertising sites is the ability to sell electronic
adertising in cost-per-thousand ,CPM,, cost-per-click ,CPC,, cost-per-acquisition ,CPA, or
sponsorship models.

In the cost-per-thousand ,CPM, rom cost-per-milli, Latin or thousand`, model, the
adertiser agrees to pay the \eb site or eery 1,000 impressions o the ad sered. Rates or
CPM adertising ary widely depending on the type o site sering the adertising and the
qualiication o their audience in relation to the adertisers needs. Most news and sports
\eb sites sell adertising in a CPM model.

In the cost-per-click ,CPC, model, the adertiser agrees to pay the \eb site or each click on
an adertising unit, not or the indiidual impression. 1he CPC model is relatiely new and
BrIngIng L AII TogeLIer UsIng Key PerIormunce ndIcuLors ,

took o slowly until Oerture and Google started selling adertising on search \eb sites
using this model. Now, anyone ,een the author o this book, can adertise on a number o
dierent \eb sites and only pay or the traic that is actually acquired.

In the cost-per-acquisition ,CPA, model, the adertiser works out an agreement with the
\eb site that speciies that they only pay or the isitors that they truly acquire,` which in
most instances means they hae been conerted. 1he CPA model can be more intensie to
set up initially, depending on the strategy used to determine that a customer has actually
been acquired but, i a strategy can be agreed upon, this is oten the most inancially
beneicial model or the adertiser because they are no longer playing the odds regarding
conersion rates, they hae actually acquired a customer.

Sponsorship is typically a customized payment model where an adertiser agrees to pay a
\eb site a ixed amount or preerred placement throughout the site. \hile the author has
neer seen a solid return on inestment case or the sponsorship model o adertising, there
are surely many out there as this model is based on saturation and is oten used to build
brand awareness as much as it is to reach and acquire isitors.

\hile it is ery important or adertising \eb sites to reach and acquire new isitors on a
near constant basis in order to hae an expanding audience to sere adertising to, the act o
conersion is typically less clear on these \eb sites, oten limited to the act o signing up or
a newsletter or email list. Perhaps the most important component o the customer lie cycle
or the adertising business model is retention, keeping audience members coming back time
and time again.
Lead Generation
1he lead generation business model is typiied by the presentation o inormation about a
product or a business in the attempt to gather contact inormation or the isitor. 1he
general idea is that, i the isitor is interested enough in the content the \eb site proides,
and i the alue proposition is clearly presented, the isitor is more likely to proide personal
inormation such as name, email address, and so on. Important perormance indicators or
the lead generation model mostly ocus on the reach, acquisition and conersion end o the
customer lie cycle as retention is usually drien ia external ,email, phone, contact ater the
lead is generated.
Customer Support
1he customer support business model is typiied by a \eb site designed to help isitors
answer questions about a product or serice without haing to pick up the telephone and
generate a costly phone call to a support department. lor the most part, customer support
is integrated into one o the other models already discussed-rare is a \eb site that ocuses
solely on customer support ater a product has been purchased without also trying to get the
isitor to purchase additional products ,online commerce model, or to generate some kind
o lead ,lead generation,. Still, because the customer support model ocuses mostly on the
retention phase o the customer lie cycle it is worth dierentiating. Also, while the other
business models are drien inancially by the idea that more page iews are usually better`
S CIupLer 1

the customer support model tends to be the opposite, the more quickly the isitor can ind
the inormation they are looking or the better.
Key Performance Indicators by Business Model
As preiously discussed, each online business will need to monitor dierent perormance
indicators depending on their speciic online goals. It is impossible or the author, in the
context o this book, to describe to you, the reader, exactly which metrics you should be
paying attention to. In the author`s experience, howeer, there are a handul o KPIs that
make sense or businesses to watch depending on their oerall business model. 1he
ollowing sections coer those KPIs and the rationale behind their use.
Online Commerce Key Performance Indicators
Because it is as important to retain customers as it is to acquire and conert new isitors, the
online commerce model draws perormance indicators rom each phase o the customer lie
cycle. Please see the description o the original metric in preious chapters or additional
inormation about the indicator`s calculation.

Metric Rationale for Inclusion
Ratio of New to Returning
Visitors
Page 116
1he ratio o new to returning isitors, called
acquisition mode` by the author, is a ast reerence to
describe the eect o your marketing mix on driing
isits.
Percent New Visitors
Page 118
Because the online commerce model depends on the
near-constant input o new isitors, monitoring the
percentage o new isitors to your site is extremely
important.
Page Stickiness
Page 138
lor key landing pages on your \eb site ,home page,
marketing response pages,, you want to ensure that
isitors are responding to your marketing materials but
getting no urther than the irst page.
Referring Domains
Page 158
Keeping track o the top 10 or so \eb sites that are
sending you traic will help you determine whether
online business partners are perorming as expected.
Search Keywords and Phrases
Page 166
Keeping track o the top search engine words and
phrases that are driing traic to you will help you
ensure that your eatured product mix is one that
inbound isitors are actually looking or.
Percent of Visits Under 90
Seconds
Page 149
Dramatic changes in this percentage can indicate either
much better or much worse reach in your marketing
campaigns. Lither should be responded to proactiely.
Key Conversion Rates
Page 14
Any and all key conersion rates, such as home to
purchase, search to purchase, shopping cart and
checkout completion should be monitored closely.

BrIngIng L AII TogeLIer UsIng Key PerIormunce ndIcuLors q

Average Order Value
Page 190
\atching or signiicant moement in your aerage
order alue can identiy changes in either your
customer or aailable product mix, both o which
should be responded to proactiely. Also monitor
AOV or new and returning customers ,xxx,.
New and Repeat Customer
Conversion Rates
Page 194
Again, here you are watching or dramatic changes that
may indicate changes in your customer base or product
and marketing mix.
Sales per Visitor
Page 195
A good point o comparison with your conersion rate,
helping you ensure that as your incoming traic
increases your site-wide conersion rate stays roughly
the same.
Percent Returning Visitors
Page 205
Much like monitoring the percent o new isitors to the
site, watching how the percent o returning isitors
changes will help you keep track o isitor retention.
Customer Retention Rate
Page 21
Again, similar to keeping track o your returning
isitors, monitoring the percentage o returning
customers is useul in determining how eectiely your
retain customers.
1able S: Recommended key performance indicators for the online commerce
business model.

\ou will probably ind it worthwhile to hae a separate worksheet in your KPI reporting or
campaign-related metrics, one that allows you to see the important eects o campaigns in a
single, easy-to-read iew. I you hae too many campaigns to monitor easily, consider
haing indiidual reports or top campaigns by response rate` and top campaigns by
conersion rates,` as these are the undamental measurements o campaign success.

Metric (Page in Book) Rationale for Inclusion
Open Rates
Page 12
As you work to reach isitors, making sure that open
rates or l1ML mailings stay high will help you
determine whether your message is being well receied.
Response Rates
Page 152
\ou want to make sure that you know at any gien time
what campaigns are working hardest or you to attract
isitors. \hile response rate can be slightly more
inoled to calculate, monitoring response rate will help
you control marketing spend.
Cost-per-Acquisition
Page 154
Monitoring your acquisition costs or campaigns is
perhaps the best way to control your marketing spend
on campaigns.
Campaign Conversion Rates
Page 182
lor your top campaigns, keeping track o their
conersion rate is again critical to ensuring that your
marketing eorts are ocused on your most successul
campaigns.
1able 6: Recommended key performance indicators for tracking campaigns within
the online commerce business model.
o CIupLer 1


It is important to keep in mind that these perormance indicators are only a subset o all
aailable metrics discussed in this book. \our online business may hae dierent speciic
reporting needs and you should work to include those indicators that are most useul to you.
Advertising Key Performance Indicators
1he adertising model or online businesses dictates that new isitors are acquired and that
they iew as many dierent pages and pieces o content as possible in order to generate
adertising iews or the \eb site`s customers.` 1he recommended perormance
indicators relect this ocus. Please see the description o the original metric in preious
chapters or additional inormation about the indicator`s calculation.

Metric (Page in Book) Rationale for Inclusion
Overall 1raffic Volumes
Page 110
Because adertising iews are so closely tied to page
iews it is worthwhile to keep track o changes in
oerall traic olume to the \eb site.
Number of Visits
Page 112
Similar to oerall traic olumes, watching or changes
in the number o isits to the \eb site can help predict
whether more or ewer adertising iews will be
aailable.
Ratio of New to Returning
Visitors
Page 116
1he ratio o new to returning isitors, called
acquisition mode` by the author, is a ast reerence to
describe the eect your marketing mix has on driing
isits.
Percent New Visitors
Page 118
Because the adertising model depends on the near-
constant input o new isitors, monitoring the
percentage o new isitors to your site is extremely
important.
Percent Returning Visitors
Page 205
Much like monitoring the percent o new isitors to the
site, watching how the percent o returning isitors
changes will help you keep track o isitor retention.
Page Stickiness
Page 138
lor key landing pages on your \eb site ,home page,
marketing response pages, you want to ensure that
isitors are responding to your marketing materials but
getting no urther than the irst page.
Referring Domains
Page 158
Keeping track o the top 10 or so \eb sites that are
sending you traic will help you determine whether
online business partners are perorming as expected.
Search Keywords and Phrases
Page 166
Keeping track o the top search engine words and
phrases that are driing traic to you will help you
ensure that your content is what inbound isitors are
actually looking or.
Percent of Visits Under 90
Seconds
Page 149
Dramatic changes in this percentage can indicate either
much better or much worse reach in your marketing
campaigns. Lither should be responded to proactiely.
BrIngIng L AII TogeLIer UsIng Key PerIormunce ndIcuLors 1

Average Pages Viewed per
Visitor
Page 13
Knowing how many pages, and adertising iews, you
can expect the aerage isitor to see is helpul when
selling adertising units. Monitoring or changes in this
number is important to ensure adertising deliery.
Average Number of Visits per
Visitor
Page 134
Important or understanding how attracted your
isitors are to your content and worth monitoring or
large changes.
Lntry Pages and Content
Page 120
Keeping track o the top 10 pages and content areas
that isitors are most likely to begin their isit will help
you identiy key pages adertisers may be interested in.
Heavy User Share
Page 143
Depending on how you deine heay user` this can be
an important KPI to keep track o the percentage o
your online audience that is likely to see a larger
number o adertising units.
Average 1ime Spent on Site
Page 150
Longer isits typically mean greater interest in content
and more adertising iews, so large changes in aerage
time spent should be monitored.
Key Conversion Rates
Page 14
Any conersion rate on your \eb site that will lead to
increased retention and loyalty should be monitored
closely. Lxamples include newsletter sign-up, alert
eatures and chat or newsgroup application sign-up.
Activity of Retained Visitors
Page 214
I you hae the ability to determine the number o page
iews, isits, isitors and aerage time spent or
returning isitors it is worthwhile to separate that
reporting out to watch or changes in the actiity o
your retained audience.
1able 7: Recommended key performance indicators for the online advertising
business model.

Additionally, i you are actiely spending money on campaigns to drie isitors to your \eb
site you should consider monitoring campaign metrics as described in 1able .
Lead Generation Key Performance Indicators
1he lead generation business model is ocused on haing isitors iew inormation that
prompts them to submit some kind o personal inormation ,the lead`,. 1he
recommended perormance indicators relect this ocus. Please see the description o the
original metric in preious chapters or additional inormation about the indicator`s
calculation.

Metric (Page in Book) Rationale for Inclusion
Overall 1raffic Volumes
Page 110
Because the likelihood o generating a lead is tied to the
number o page iews it is worthwhile to keep track o
changes in oerall traic olume to the \eb site.




CIupLer 1

Ratio of New to Returning
Visitors
Page 116
1he ratio o new to returning isitors, called
acquisition mode` by the author, is a ast reerence to
describe the eect your marketing mix has on driing
isits.
Percent New Visitors
Page 118
Because the lead generation model depends on the
near-constant input o new isitors, monitoring the
percentage o new isitors to your site is extremely
important.
Percent Returning Visitors
Page 205
Much like monitoring the percent o new isitors to the
site, watching how the percent o returning isitors
changes will help you keep track o the olume o
isitors who hae learned to use your support systems
online.
Page Stickiness
Page 138
lor key landing pages on your \eb site ,home page,
marketing response pages, you want to ensure that
isitors are responding to your marketing materials but
getting no urther than the irst page.
Referring Domains
Page 158
Keeping track o the top 10 or so \eb sites that are
sending you traic will help you determine whether
online business partners are perorming as expected.
Search Keywords and Phrases
Page 166
Keeping track o the top search engine words and
phrases that are driing traic to you will help you
ensure that your content is what inbound isitors are
actually looking or.
Lntry Pages and Content
Page 120
Keeping track o the top 10 pages and content areas
that isitors are most likely to begin their isit will help
you identiy the type o inormation isitors are most
interested in.
Lead Conversion Rate
Page 16
1he lead conersion rate is likely the most important
KPI you will track. \ou should always work to
understand why this number is increasing or decreasing.
1op Pages and Content
Requested by New Visitors
Page 144
Knowing what irst-time eer isitors are looking or
on your \eb site can help you to tailor content and
drie isitors towards the lead generation process more
quickly. Changes in these patterns should be
responded to proactiely.
1able 8: Recommended key performance indicators for the lead generation business
model.

Additionally, i you are actiely spending money on campaigns to drie isitors to your \eb
site you should consider monitoring campaign metrics as described in 1able 8.
Customer Support Key Performance Indicators
1he customer support business model is mostly ocused on a isitor`s ability to ind speciic
kinds o inormation, with ewer clicks being better as this likely indicates that isitors are
BrIngIng L AII TogeLIer UsIng Key PerIormunce ndIcuLors

inding the inormation are looking or quickly. 1he recommended perormance indicators
relect this ocus. Please see the description o the original metric in preious chapters or
additional inormation about the indicator`s calculation.

Metric (Page in Book) Rationale for Inclusion
Overall 1raffic Volumes
Page 110
Because isitor desire or support is tied to the number
o page iews it is worthwhile to keep track o changes
in oerall traic olume to the \eb site.
Percent New Visitors
Page 133
New isitors are those who may struggle more to ind
inormation as they learn your inormation architecture
and support system. 1he greater the olume o new
isitors, the more likely you are to hae inbound
support calls i isitors are unable to ind inormation
online.
Page Stickiness
Page 138
lor key landing pages on your \eb site ,home page,
product support pages, you want to ensure that isitors
are not becoming conused and simply leaing the \eb
site.
Search Keywords and Phrases
Page 166
Keeping track o the top search engine words and
phrases that are driing traic to you will proide
insight into what support issues isitors may be
experiencing most commonly.
Percent of Visits Under 90
Seconds
Page 149
It is unlikely that a isitor can do much on a customer
support site in less than 90 seconds other than ind a
phone number to call. 1his indicator should be
watched and correlated to call olume ,i possible,.
Lntry Pages and Content
Page 120
Keeping track o the top 10 pages and content areas
that isitors are most likely to begin their isit will help
you identiy the type o inormation isitors are most
interested in.
Information Iind
Conversion Rate
Page 1
I you track the process o isitors moing rom your
home page to speciic inormation you can generate an
inormation ind` conersion rate. \ou should always
work to understand why this number is increasing or
decreasing.
1op Pages and Content
Requested by New Visitors
Page 144
Knowing what irst-time eer isitors are looking or
on your \eb site can help you to tailor content and
drie isitors towards the most common support
topics. Changes in these patterns should be responded
to proactiely.
1able 9: Recommended key performance indicators for the customer support
business model.

Do keep in mind that each o the perormance indicators listed aboe or each business
model are simply recommendations and that you may need to ollow more, or dierent,
metrics depending on your speciic needs. 1he most important thing to consider when
deciding which metrics to watch as KPIs is the answer to the question, Is the inormation
q CIupLer 1

actionable` I the KPI does not hae the potential to drie action and make you look more
deeply into your analytics reports or make some critical change to your \eb site or
marketing programs, the metric is likely vot a good key perormance indicator or you.
Advunced TopIcs g

CHAPTIR 16
AVANCI TOPICS

Leveraging information and capabilities throughout the corporation, these
integrated Web analytics and e-Business applications can deliver sustainable
competitive advantage by helping corporations present a unifiedrather than
fragmentedfront to the customer.
Guy Creese and Mark Veytsel, the Aberdeen Group


Congratulations! I you are still reading and you hae made it this ar you now understand
more about \eb analytics than most people know, want to know, or een know that there is
to know. Lerything contained in !eb .vat,tic. Dev,.tifiea in the preceding chapters was
about core unctionality that is, or should be, proided by \eb analytics applications-the
basics that eeryone should understand in order to make the best use o whateer analytics
application they hae purchased to help them run their online business. I you hae read the
preious chapters and they hae made complete sense to you then you are likely well on your
way to a signiicant upgrade in your \eb analytics reporting-which will, hopeully, translate
into a better isitor experience online, more proits, less abandonment, and so on.

Still, when the author sat down and planned !eb .vat,tic. Dev,.tifiea there were a small
number o topics that were important, but outside o the concept o a Marketers Guide to
\eb analytics.` Data integration-or the combining o the aorementioned \eb analytics
data with related data about the customer, the perormance o the \eb site, inormation
about adertising, and so on, is a topic o some import to online businesses that are a bit
urther along in their understanding o the \eb Analytic process. Conersely, the use o
personas when architecting \eb sites and \eb-based applications can be a powerul tool in
making sure that your design is consistent with both your site`s business objecties and the
goals o your most important customers. 1hese topics are treated in this inal Chapter o
!eb .vat,tic. Dev,.tifiea.
Data Integration
\hile eery metric presented in the preceding chapters o !eb .vat,tic. Dev,.tifiea should be
aailable rom your analytics application regardless o price or complexity, some online
businesses will ineitably be urther along and need additional inormation about their
isitors. One o the irst topics to arise when businesses begin to moe towards a more
adanced \eb analytics program is that o data integration.

Data integration is built around the idea that the \eb traic data you are collecting and
analyzing in your analytic program is only a subset o the data that you actually hae about
your online isitor or customer. 1here are typically a number o dierent but related data
6 CIupLer 16

sources that are components that the author considers inormation about the customer`-
especially or businesses doing some type o online commerce- including:

Customer speciic data, such as the customer`s name, address and business title
Marketing inormation, such as the campaigns you hae exposed the customer to in
the past and those they hae aorably responded to
Surey inormation, such as data collected rom exit sureys or external sources
such as BizRate.com
Customer complaints and questions, logged in a customer support system

In addition, many online businesses that are part o larger, multi-channel enterprises hae the
desire to integrate data rom other parts o the enterprise into their analytics reporting to
create a single-iew enironment o how the oerall business is perorming. Data oten
integrated when businesses are looking to create enterprise-wide iews that the author
considers inormation about sales` include:

Order speciic data, such as the customer`s order history, requency o returns,
types o purchase
Point-o-sale order inormation rom indiidual stores
Order inormation taken rom kiosks in stores
Order inormation taken rom customer serice representaties ,CSR, ia
telephone
Orders originating rom catalogs and oline adertising, including teleision, radio,
newspaper and direct mail

linally, some online businesses want to hae the tightest integration possible and combine
inormation about online sales with enterprise resource planning ,LRP, inormation, which
the author considers inormation about the business,` including:

System-wide aailability o products
Cost and margin inormation about products
Shipping inormation
Oerall health-o-business as measured against online sales or traic

Larly in the history o \eb analytics there was a sense that eentually this work will lead to a
360-degree iew o the customer,` meaning you would be able to see eerything they did,
beginning to end, as they interacted with your business online and oline. Back-end data
integration was ultimately going to be the conduit through which this holy grail o customer
knowledge would be enabled because inormation about the customer was truly disparate.
Because dierent systems were used to track point-o-sale, telesales, catalog sales, online
sales and customer support, a single customer identiier was unlikely to be ound. \ithout a
single customer identiier, the integration o these disparate, but related, data appeared to be
unlikely at best.

Seeral years hae now passed, giing sotware engineers plenty o time to work on this
problem but the author is disappointed to report that little progress has been made in this
area. On the contrary, most analysts and endors are proclaiming that the 360-degree iew
Advunced TopIcs ,

o the customer` was simply psychobabble and is unlikely to eer be aailable, no matter the
price. \hile the author tends towards the pessimistic side regarding promises made by the
endors and analyst community, he does beliee that there is alue in attempting to build a
wider iew o the customer, but that it should be done as makes sense and where the data is
easily combinable.

1o this end, some o the endors are currently establishing partnerships with companies that
hae dierent, but related, iews o the customer. Some notable examples include:

Coremetrics partnerships with email proiders like CheetahMail and Digital Impact,
enabling direct email marketing integration, as well as their partnerships with
Belree or ailiate marketing integration and loreSee Results and BizRate.com or
customer surey integration
\ebSideStory`s partnership with Saleslorce.com enabling one-to-one online
marketing integration
Omniture`s partnership with Speedera to proide integrated application
perormance measurement
\eb1rends,NetIQ`s partnership with lyperion to enable OLAP integration o
multiple data sources

1he general trend you can see aboe is that endors are orging strategic partnerships around
their indiidual technologies to allow isitors and customers to be tied together using a
common customer identiier. \hile these particular partnerships only help you i you hae
no analytics solution in place-or are willing to scrap what you hae, choose new endors
and start oer-it is worthwhile to know that analytics endors are working on soling the
problem o data integration.

Depending on the particular stage your online business is in regarding analytics and
reporting, the integration o disparate data will be either absolutely irreleant or absolutely
critical. I data integration is a priority or you there are two broad strategies or making this
happen, custom data collection and back-end data integration.
Custom Data Collection
One o the quickly emerging solutions or data integration is the idea o custom data
collection and reporting ia the existing analytics application interace. Instead o limiting
what can be tracked and reported on to the core \eb analytics data points ,iews, isits,
isitors, reerrers, and so on,, many \eb analytics applications are opening up their data
collection architectures to allow any data point to be collected and analyzed. By allowing
you, the customer, to deine speciically what data you want tracked and reported on, light`
data integration is made aailable. \hile the author beliees that more and more \eb
analytics endors will be proiding custom data collection in the uture, this strategy is not
without its adantages and disadantages.
S CIupLer 16

Advantages of Custom Data Collection
1he major adantage o using custom data collection to enable light` data integration and
reporting is that most known endors oer reporting on custom data directly rom within
their existing interace. \hy this is an adantage will become more clear ater reading about
back-end data integration, but suice to say being able to easily tie custom data directly to
the more traditional \eb analytic data discussed in this book is desired. Custom data
collection can acilitate many o the most common data integration needs expressed by
online businesses, especially or endors who hae soled the problem o being able to tie
normal analytic data directly to custom ields using some type o cross-tabulation or drill-
down unctionality.

A related adantage o custom data collection is that, because this collection and reporting
happens within the existing analytics interace it typically costs less. Again, this will be
clearer in the context o back-end data integration, but unless your endor supports direct
import o non-standard data into their reporting module there will likely be some reporting
costs incurred internally by your business.
Disadvantages of Custom Data Collection
1he major disadantage o custom data collection is the limitation on what types o data can
be collected and how the data is tied to other data within the interace. In this regard, some
endors are able to collect custom ariables in the orm o strings and simply report back on
the number o times each string was collected ,analogous to a page iew,. Other endors are
able to collect complex, two-dimensional strings in a parent-children` relationship and
report back on the number o iews generated or each. \et other endors are able to tie
custom-collected data back to other eents in a isit, such as a purchase or an internal
search. Depending on what kind o custom inormation you are trying to glean about your
isitors you will need dierent reporting abilities rom your analytics application. Beore you
commit to using custom data collection to satisy your data integration reporting needs you
should contact your analytics endor to determine what can be collected and how this data is
tied to other data within the application.
Back-End Data Integration
Conceptually, back-end data integration inoles tying together disparate, but related, data
sources ater the act in order to deelop a more clear or reined picture o what
happened.` 1he most common case o this type o data integration is the who was that`
model, tying \eb Analytic data to your company`s customer relationship database in such a
way you can stop saying 10 isitors looked at our pricing page yesterday` and start to say
Bob and 1ed and Alice and Ramona looked at our pricing page yesterday and here are each
o their email addresses so we can ollow up with them indiidually.` Marketers much
preer the latter, to the point that at least one endor in the analytics marketplace has made
considerable inestment in making this type o data integration easy and accessible.

1he undamental challenges in this type o data integration are, \hat application will be
used to analyze the data` and, low will the records be tied together` lor the irst
Advunced TopIcs q

challenge, there are a ariety o tools-business intelligence tools, OLAP applications, data
warehouses, databases and een the \eb analytics endors` own toolsets-that are able to
bring the data together in a common enironment, prerequisite or analysis. 1he best
application to use or data integration really depends on what kinds o questions you are
trying to answer. Regarding the second challenge, this is the >64,000 question, how to tie
these data sources together in a meaningul way.
What Application Should Be Used to Integrate Data?
1he answer to this question is, unortunately, It depends.` It depends on what kinds o
questions you are trying to answer. It depends on your internal expertise with databases and
data warehouses. It depends on whether your \eb analytics application supports the direct
import o non-traic data. Lach o the dierent kinds o applications listed aboe will likely
create as many problems as they are able to sole when attempting to integrate data, so the
best guidance that the author can gie you in this regard is as ollows:
Have Well-Defined Questions
1he more deined your questions are the more clear the strategy or combining data will
become. \e want a list o the names o known isitors who saw pricing and demo pages
in the month o August,` will be easier to determine than \e want to know eerything
about all the isitors who hae eer been to the \eb site.`
Integrate Slowly, if Possible
By keeping the scope o indiidual integration projects reasonable and then building on your
knowledge you are more likely to be successul than i you just jump in, head irst. Keep in
mind that the continuous improement process applies eerywhere, not just to making
changes to your \eb site.
Dont Become Frustrated
Remember, hindsight is 20,20 and nobody is perect. I you really get inoled in data
integration projects you will quickly run into questions that cannot be answered or any
number o reasons. Some data is not aailable, some data sets are not combinable, some
data sets are too large to be analyzed in a reasonable amount o time. 1he best practice here
is to make note o the questions you wanted answers or but could not get and reisit them
next time you are shopping or a \eb analytics application.
Garbage In Means Garbage Out!
Another uniersal truth: crappy data combined with good data yields crappy data, not the
other way around. \hen you hae worked out a strategy or data integration you should
consider spending signiicant time making sure that the data sets are as clean as possible
prior to their inal integration or analysis. Vigorous scrubbing is necessary to ensure that
the combined data will be meaningul, especially when attempting to integrate CRM data
collected primarily ia the Internet.
qo CIupLer 16

Choose a Data Integration Tool Carefully
Because data integration is relatiely new it is not uncommon that people will use new
applications or applications they are unamiliar with to attempt to combine data. 1he
problem with this is that you may ery well end up banging your head against a wall to sole
a problem that would be easily soled with enough inormation about the application. By
integrating in an enironment that you or someone in your company knows well or by
integrating in a well-known enironment ,such as SQL, which is well-supported on and o
the Internet, you increase the chances that the integration will go smoothly. Alternatiely
you could outsource the integration through your \eb analytics endor, proided they oer
integration serices, and rely on their expertise.
How Should Different Data Sets Be Tied Together?
Once you hae decided on an application or process or integrating the data you still hae to
work out how the data sets should be related to each other. Depending on the kinds o
questions you are trying to answer this can be more or less diicult. Say, or example, you
are simply trying to tie \eb traic to the olume o inbound calls to your customer support
center on a daily basis. 1his integration will likely be ery easy since the key that ties the two
data sets together is simply the date,time stamp on the records in each set. Likely you hae
granularity to at least the leel o the minute, allowing you to slice-and-dice the data rom
each set and answer questions like:

\hen call olumes go up, was there a related increase in traic to our support site
immediately prior, suggesting that isitors are not making eectie use o online
support materials
Do we see a correlation between calls on a certain subject and traic to that
subject`s support area online

But what i you want to know the exact pages that a isitor was looking at in your customer
support section just prior to making a call to your customer support center laing this
inormation would be great or your support sta, no doubt! 1hey could ask insightul
questions o the caller about the inormation they had recently iewed and perhaps sae the
caller time in diagnosing any problem they may be haing. Unortunately, this type o
integration has a strong dependence on A, a unique user identiier ,UUID, that is common
to both your \eb enironment and your customer support management system and B, real-
time, integrated reporting based on the UUID.

Let the author assure you that while A` aboe is quite possible gien a great deal o thought
when designing your \eb site, the combination o A` and B` is quite diicult to attain,
perhaps impossible ,or the time being,.

Again, the best the author can do is to proide the ollowing guidance when attempting to
tie data sets together:
Advunced TopIcs q1

Consider the UUID
I you are able to pass in a UUID o your choosing, either ia a ariable in your page tag or
the query string to your log iles, you will be much more likely to be able to tie your \eb
data back to internal inormation. Because eery system generates its own unique ID or a
customer, isitor, and so on. your best bet is to tie the \eb data to another system`s UUID
at the time the data is generated. 1his way, proiding the analytics application is able to
proide you some type o data export, you will already hae a UUID that is yours`
contained in the data.
Carefully Research Your Data Export Options
Unless you hae knowingly purchased a \eb analytics application that allows you to import
data rom other sources you will most likely end up exporting data or analysis in some other
enironment. I this is the case you want to understand as much as possible about export
ormats in order to determine what ields may be aailable to acilitate data combination.
\hile most applications support some type o export to Microsot \ord or Microsot
Lxcel these ormats are typically too limiting or data integration projects. A comma- or
tab-delimited ile ,CSV, is getting better but the ideal ormats are direct exports to a database
or XML. Lach o the latter two ormats will come with an established schema describing
the relationship between \eb analytics data.
Choose Your Vendor Carefully
Not all \eb analytics endors hae built the same support into their applications or data
import and export. As o the time o this book`s writing, some endors proided robust
support or importing and exporting data, either directly or ia OLAP tools proided by
business partners. Other endors had worked quite hard on establishing partnerships with
companies that hae requently integrated data sets ,email, adertising, marketing, CRM,
perormance data,. Still other endors proide little or no support or data integrations
projects.

It is best to ask your existing endor how they A, currently support data integration projects
and B, plan on doing so in the uture. I they look at you with a blank stare when you ask
this question, buy them a copy o !eb .vat,tic. Dev,.tifiea and get a new analytics endor.
Advantages of Back-End Data Integration
1he major adantage o back-end data integration is that, when properly done, the
inormation quality o \eb analytics data is signiicantly increased. 1he dierence between
knowing that 100 isitors` did something, and knowing who each o those 100 isitors`
were and what they had done on preious isits to your \eb site is, well, like the dierence
between /vorivg abovt money and actually barivg money. 1he more you know about your
online isitors and customers the better o you are. Being able to directly market to
prospects based on their online habits is surely part o the uture o \eb analytics.
q CIupLer 16

Disadvantages of Back-End Data Integration
Aside rom the diiculty inoled in setting up a back-end data integration, the major
disadantages o back-end data integration reole around priacy concerns. 1here hae
been a ew cases o \eb analytics endors getting in trouble with priacy groups or passing
inormation deemed to be priate` using only semi-secure protocols ,l11PS is semi-
secure, at best,. \ou should be careul to A, protect your customer`s data as much as
possible and B, hae a clearly posted priacy policy stating exactly what you are doing and
why you are doing it, especially when outsourcing the integration o data, which typically
inoled passing o inormation ia the Internet. \hile it is amazing to this author what the
popular media and general public will become agitated about, cookies being the most
popular subject or alarm, woe is the online business that attracts the wrath o the priacy
concerned` on the Internet. More than any other aspect o \eb analytics measurement, the
combination o related but separate data sources is the most likely to attract unwanted
attention and so should be monitored closely and administered closely.
Use of Personas in Design and Architecture
A strategy that is becoming more common in the \eb design process is the use o personas
and persuasie architecture. Personas, irst described by Alan Cooper in his 1999 book, 1be
vvate. .re Rvvvivg tbe ..,tvv, are a tool used to ocus designers and design decision makers
on the real needs o their customers. 1hese personas are essentially ictitious people with
real goals that they want to accomplish on your \eb site. 1he use o personas allow
designers and decision makers to easily choose to include or drop pages and content based
on the answer to the question, \ould this inormation help one o our personas complete
their task on our \eb site` I the answer is no,` or probably not,` then the page can be
let out o the inal design. 1hese tools, when used properly, can proide your design team a
great Ocham`s razor to simpliy your \eb design.
How Personas Are Used
larley Manning o lorrester Research does a good job o describing how personas are best
deined in his 2003 report, 1be Porer of De.igv Per.ova.. In this report Manning cites that
studying indiidual users in the business` target market deelops eectie personas. 1he
descriptions are then used to tell a story about a real person, their motiations, goals and
behaiors. An example persona that BackcountryStore.com may hae used in their design
process may be described as such:
Advunced TopIcs q


Name: Bobby Backcountry
Goals: Buying high-end outdoor gear at the best price, eeling like he`s
a contributing member o an elite community.
Motivations: Bobby is motiated by a need to be perceied as an elite
outdoor athlete, one that can aord the ery best gear and
knows how to use it. Bobby wants to shop on a \eb site that
also has a community, because he beliees himsel to be a
knowledgeable gear head` and he wants to share that
knowledge. Bobby is always looking or the best deal on gear
because his inancial situation does not allow or exorbitant
spending.
Description:


Bobby is in his early 30s and earns less than >30K per year
working an inormation technology job that aords him the
opportunity to escape into the backcountry at a moment`s
notice. Bobby has always made due` with whateer gear he
could aord or ind but since turning 30 he yearns to be
perceied o as successul. Bobby is not married and does not
hae a girlriend right now, he inds companionship both
among other outdoor enthusiasts and also, because he is a
technocrat, online at \eb sites requented by backcountry
experts.
Actions: Research gear, Purchase sale items, Register, Post to
community orums.
Irequency of Visits: lrequent - up to once per week
Irequency of Purchase: Inrequently - typically around sales actiity
Keywords: ,All high-end brands,, sale, sale items, specials, discounts, on
sale, saings, cost saings.
1able J0: A fictitious persona that may have been useful for the owners of
BackcountryStore.com when designing their Web site. It is worth noting that the
addition of motivations to a persona is a patent-pending invention of Iuture Now,
Inc.

1his description o Bobby`s persona will help designers greatly in making design decisions.
Imagine that Bobby is BackcountryStore.com`s primary persona-the key user type that
BackcountryStore.com is trying to sere. I designers were aced with the ollowing design
and marketing decisions, understanding Bobby would make the decision process ery easy.

1. Do we eature our most high-end gear, regardless o the price
2. Do we hide our clearance and discount gear since the margins are much lower on
these items
3. Do we de-emphasize the role o community, assuming that anyone who comes to
BackcountryStore.com already has enough riends to hang out with

Clearly, with Bobby in mind the answer to each o these questions would be no.` Bobby is
looking or deals on gear and wants to spend time on a \eb site that creates and osters a
sense o community. \ith a small number o Bobby`s` used careully to ocus design
qq CIupLer 16

decisions, \eb site owners can streamline the process o haing to decide which pages
should and should not be included in the inal ersion o the site.
Personas Used in Persuasion Architecture
Bryan Lisenberg and the smart people at luture Now, Inc., hae incorporated personas into
their design ramework, one that they hae termed persuasion architecture.` 1he main
tenet o persuasion architecture is that the goal o any commercial \eb site is to inorm and
er.vaae-persuade the isitor to proide a lead, make a purchase, click more deeply into the
site or not pick up the phone and place a costly telephone call. 1he way you persuade a
isitor to do something is to create a series o micro-actions that, when considered
collectiely, constitute a macro-action. 1he macro-action is the business objectie that we
discussed in Chapter 5-make a purchase, submit a lead, and so on. 1he micro-actions are
the indiidual steps in the process o completing the business objectie, steps that can be
measured using any number o the tools and metrics described in this book.

Persuasion architecture essentially asks these three critical questions o eery page a isitor
would see in a macro-action:

1. \hat action needs to be taken at this point
2. \ho, or which persona, needs to be taking that action
3. low do we persuade the person or persona to take the action

By answering these important questions you are again able to use personas in a ery aluable
way. 1ypically the answer to the irst question is something like, Get the isitor to moe to
the next page in the process.` By understanding which o your personas is likely to be
engaged in the macro-action, something that would hae been clearly established at the
persona-deinition stage, you will hae the answer to question two. 1he answer to question
two proides you the answer to question three-simply consult the motiations and
keywords or each o your personas and write marketing language that relects that
inormation.

Again, as an example, say that Bobby was looking at a particular product on
BackcountryStore.com, one that was already being oered at a discount. \hen we are trying
to persuade Bobby to add that item to his shopping cart, based on his persona, the language
we are going to use is that describing the quality o the particular product and how the
product is preerred by other gear experts. Bobby will hae already done his homework and
realized that the price is among the best he will get. \e know he is a loyal isitor, one likely
to come back week ater week looking or the best deals. Now all we hae to do is make
sure that he understands that the product in question is among the best aailable and Bobby
will conert.
Why Is This Relevant?
1he author chose to include this discussion on personas and persuasion architecture because
it is extremely common or \eb site design projects to be run with complete disregard to
how the inal product will be measured. \eb sites are commonly thrown together with a
Advunced TopIcs qg

great deal o thought and discussion in the early stages and then most o that work gets
thrown out in the eleenth hour when the project is in danger o being deliered late. \hile
this is not always the case, in the author`s experience it is more common than not. \hen
eeryone is rushing to build out templates, inish marketing copy and optimize images the
last thing on anyone`s mind is, low will the eect o all this work be measured`

Obiously the author sees this as a huge problem, one that needs to be addressed.

1he continuous improement process described in Chapter 2 does not only need to be
applied to a inished product undergoing reision. 1he design` and measure` steps are
easily applied to any step in the \eb design process, regardless o the state o
completeness` o the \eb site. luture Now`s persuasion architecture blends nicely with
the continuous improement process because it explicitly asks the question, \hat should
the isitor do next` 1his micro-action can be measured in nearly eery instance-more
oten than not you are looking or a simple conersion rate o isitors moing rom one
page to the next. Bubble this back up to the macro-action and you are looking at a single,
measurable step in an action. Continue to bubble this up and you are measuring the eect
o a single action on an oerall business objectie.

\ell thought-out design is undamental to taking adantage o most o the metrics described
in this book. I you build without measurement in mind you are likely to back yoursel into
a corner and hae little or no idea how good or bad a job you did with your design. In this
regard, the author`s recommendation regarding design is as ollows:

1. Clearly deine the business objecties or your \eb site, irst and oremost.
2. Make use o personas to proide ocus to your designs when important decisions
need to be made regarding a page`s releance.
3. Use persuasion architecture to derie the greatest alue rom your personas and
identiy the micro-actions that isitors need to accomplish as they eectiely
accomplish your business objecties.
4. Build measurement in during the design stages, going so ar as to establish key
perormance indicators or each micro- and macro-action.
5. Apply the continuous improement process wheneer possible with all o the
aboe in mind, constantly looking or changes in key personas and how those
personas are completing key business objecties.

q6 BIbIIogrupIy

BIBLIOGRAPHY
Books and Electronic Documents
Berners-Lee, 1im. 2000. \eaing the \eb: 1he Original Design and Ultimate Destiny o
the \orld \ide \eb. larperBusiness, New \ork, New \ork.

Creese, Guy and Alex Veytsel. 2002. !eb .vat,tic.: Ma/ivg v.ive.. ev.e of Ovtive ebarior.
1he Aberdeen Group, Inc., Boston, Massachusetts.

Lisenberg, Bryan. 2003. !bat Covrert. earcb vgive 1raffic. luture Now, Inc., New \ork,
New \ork. ,www.caeshops.com,uturenowinc.826183,

Lisenberg, Bryan, Jim Noo and John L. Shreee. 2002. 1be Cviae to !eb .vat,tic.: or to
|vaer.tava ava |.e Yovr !eb 1reva. to Maivie Re.vtt.. luture Now, Inc., New \ork, New
\ork.

lletcher, Peter, Alex Poon, Ben Pearce and Peter Comber. Practicat !eb 1raffic .vat,.i.:
tavaara., Prirac,, 1ecbviqve., Re.vtt.. glasshaus, Birmingham, United Kingdom.

Inan, lurol. 2002. Mea.vrivg tbe vcce.. of ,ovr !eb.ite: . Cv.toverCevtric .roacb to !eb.ite
Mavagevevt. Pearson Lducation Australia, lrenchs lorest, New South \ales, Australia.

Krug, Stee. 2000. Don`t Make Me 1hink: A Common Sense Approach to \eb Usability.
QUL, Indianapolis, Indiana.

Manning, larley with Bruce D. 1emkin and Nicole Belanger. 2003. 1be Porer of De.igv
Per.ova.. lorrester Research, Inc. Cambridge, Massachusets.

Neilsen, Jakob. 2000. !eb |.abitit,: 1be Practice of ivticit,. New Riders Publishing,
Indianapolis, Indiana.

Neilsen, Jakob. 1994. |.abitit, vgiveerivg. Morgan Kauman, San lrancisco, Caliornia.

Silerstein, Michael, larold Sirkin and Peter Stanger. 2003. Retaitivg Ovtive: Covivg of .ge.
Llectronic document published and distributed by 1he Boston Consulting Group
,www.bcg.com,.

Sterne, Jim. 2002. !eb Metric.: Prorev Metboa. for Mea.vrivg !eb ite vcce... John \iley &
Sons, Inc. New \ork, New \ork.

Sterne, Jim. 2003. .aravcea vait Mar/etivg: |.ivg vait to .cbiere ate. ava Mar/etivg Coat..
Lyris 1echnologies, Inc. Berkeley, Caliornia.

BIbIIogrupIy q,

Sterne, Jim and Matt Cutler. 2000. Metric.: v.ive.. Metric. for tbe ^er covov,. Llectronic
document published and distributed by NetGenesis ,www.netgen.com,.
Web Sites Referenced
Because o the eer-changing nature o the Internet the author has opted to only proide
base URLs or inormation cited in !eb .vat,tic. Dev,.tifiea and would encourage the reader
to search the \eb site or speciic inormation cited.

Web Site URL Referenced In
Aberdeen www.aberdeen.com Chapter 1
Gartner Group www.gartnergroup.com Chapter 1
PC \ebopedia www.pcwebopedia.com Chapter 1
litBox \eb Doctor www.hitboxcentral.com Chapter 2
Amazon www.amazon.com Chapter 2
\al-mart www.wal-mart.com Chapter 2
Microsot Network www.msn.com Chapter 2
\ahoo! www.yahoo.com Chapter 2
LSPN www.espn.com Chapter 2
Keynote Systems www.keynote.com Chapter 2
NetMechanic www.netmechanic.com Chapter 2
OpinionLab www.opinionlab.com Chapter 2
\eb1rends,NetIQ www.webtrends.com Chapter 2
\ebSideStory www.websidestory.com Chapter 2
\usage www.wusage.com Chapter 2
Andromedia ,Macromedia, www.macromedia.com Chapter 3
NetGenesis,SPSS www.spss.com Chapter 3
Accrue ,Datanautics, www.datanautics.com Chapter 3
IBM Suraid www.ibm.com Chapter 3
Reenue Sciences www.reenuesciences.com Chapter 3
Apache www.apache.org Chapter 3
Microsot www.microsot.com Chapter 3
\orld \ide \eb
Consortium ,\3C,
www.w3c.org Chapter 3
Coremetrics www.coremetrics.com Chapter 3
Omniture www.omniture.com Chapter 3
Statmarket www.statmarket.com Chapter 3
NetIQ www.netiq.com Chapter 4
Best Buy www.bestbuy.com Chapter 5
Apple Computers www.apple.com Chapter 5
Akamai www.akamai.com Chapter 11
Nielsen,NetRatings www.neilsen-netratings.com Chapter 11
\orld \ide \eb
Consortium ,\3C, Library
www.w3.org Chapter 12
SearchLngine\atch searchenginewatch.com Chapter 12
CyberAtlas cyberatlas.internet.com Chapter 12
qS BIbIIogrupIy

Oerture www.oerture.com Chapter 12
SLO Consultants www.seoconsultants.com Chapter 12
ClickZ www.clickz.com Chapter 14
Jim Noo www.jimnoo.com Chapter 14
CheetahMail www.cheetahmail.com Chapter 16
Digital Impact www.digitalimpact.com Chapter 16
Belree www.beree.com Chapter 16
loreSee Results www.oreseeresults.com Chapter 16
BizRate www.bizrate.com Chapter 16
Saleslorce www.salesorce.com Chapter 16
Speedera www.speedera.com Chapter 16
lyperion www.hyperion.com Chapter 16

Please note that these URLs are only current and actie as o the time o this book`s
publication and may hae changed. Please consult the author`s \eb site,
www.webanalyticsdemystiied.com or an up-to-date list.
List of Tables and Figures
ligure 1: BackcountryStore.com home page, a shining example o how to drie isitors to
purchase online. ........................................................................................................................... 4
ligure 2: Lxample reerring URLs driing traic to the author`s \eb site,
www.webanalyticsdemystiied.com........................................................................................ 52
ligure 3: Lxample o how abandonment and conersion are oten measured using some
type o conersion unnel or process measurement tool................................................... 54
ligure 4: 1he pyramid model o \eb analytics data. .................................................................... 5
ligure 5: Sample content groups and page names rom the author`s \eb site,
www.webanalyticsdemystiied.com........................................................................................ 61
ligure 6: Lxample o top-line naigation on Circuit City \eb site ,www.circuitcity.com,.... 64
ligure : Lxample o top-line naigation on eBay site ,www.eBay.com,. ................................. 64
ligure 8: Lxample o top-line naigation on lotels.com site ,www.hotels.com,.................... 64
ligure 9: Business objectie content groupings or BackcountryStore.com. ............................ 6
ligure 10: Drill-down to Men`s Ski Jackets` in Sell Products` on BackcountryStore.com.
...................................................................................................................................................... 6
ligure 11: An example o how one endor helps their customers isualize the abandonment
and conersion through an online actiity............................................................................ 1
ligure 12: Required registration step on BackcountryStore.com rom early 2002................... 4
ligure 13: Process measurement tools employed by BackcountryStore.com. .......................... 5
ligure 14: New method or logging in as a registered user on BackcountryStore.com. ......... 5
ligure 15: Visitor segments being tracked by BackcountryStore.com. ...................................... 84
ligure 16: Visitor segment conersion obsered by BackcountryStore.com to support a
streamlined checkout process. ................................................................................................ 84
ligure 1: Lxample o how one endor helps their customers measure response, conersion,
conersion rates and campaign alues or pay-per-click adertising online. .................. 8
ligure 18: Lxample o how the author`s pay-per-click terms show up in Google Ad\ords
search results. ............................................................................................................................. 92
BIbIIogrupIy qq

ligure 19: Responses, conersions, conersion rate and campaign alue or a sample o
keywords purchased at Google by this book`s author. .......................................................92
ligure 20: Campaign perormance comparing two dierent timerames. .................................93
ligure 21: Commerce data showing the distribution o orders, reenue, and so on, rom
reerring traic partners. O note is the browse-to-buy ratio ,B,B,, which is good
or comparing the quality o traic sent by each partner................................................ 100
ligure 22: Drill-down into the products that BizRate.com helped BackcountryStore.com sell
in this example. ....................................................................................................................... 101
ligure 23: 1raic pattern measured or a single day on BackcountryStore.com`s \eb site.112
ligure 24: Graph o isits to BackcountryStore.com in December 2003. 1he yellow line
represents the period aerage, which BackcountryStore.com clearly exceeded right up
until the last ew days beore the holiday. .......................................................................... 114
ligure 25: Steady, continuous growth in the raw number o new isitors to the
BackcountryStore.com \eb site in Q4, 2003. \hile this growth is clearly good news,
a better analysis would be to compare this growth to the total number o isitors to
the \eb site to determine whether the percentage o new isitors is also growing... 116
ligure 26: Percent new isitors monitored on BackcountryStore.com ia regular KPI
reporting. BackcountryStore.com has optimized its marketing mix to achiee a
roughly 5 to 85 percent mix o new to returning isitors. ............................................ 119
ligure 2: 1op ten entry pages by percentage o entry page iews. ......................................... 121
ligure 28: Visitor cities by percent o isitors isiting BackcountryStore.com. ..................... 123
ligure 29: Interested and committed isitor share measured or BackcountryStore.com. .. 125
ligure 30: \eekly reporting on the aerage number o isits per isitor or
BackcountryStore.com, calculated using weekly` unique isitors. .............................. 135
ligure 31: Page iews per isit, part o BackcountryStore.com`s regular site perormance
reporting. .................................................................................................................................. 13
ligure 32: Page stickiness` reporting or BackcountryStore.com, part o their regular,
automated KPI report. Note that \inter Packs` alls well short o a 60 percent
threshold or the percent o isitors who click at least one-page urther, highlighted in
bold. Also note that \omen`s lleece Jackets` are o some concern with only 43.1
percent o isitors going any urther. .................................................................................. 141
ligure 33: Sample code or setting up a cookie-based isit` tracking system, written in
JaaScript by the author in as inelegant a ashion as humanly possible. ...................... 14
ligure 34: Percent o isits under 90 seconds included with interested` and committed`
isitor shares in BackcountryStore.com`s regular and automated KPI reporting. ...... 150
ligure 35: Sample data rom the author`s \eb site showing actie campaigns and number o
responses, broken down by new respondents and repeat responses. ........................... 153
ligure 36: Response rates ,called click-through-rate` in Google-ese, or a handul o
keywords that the author purchased in the last hal o 2003 ,data proided by Google
Ad\ords,................................................................................................................................. 154
ligure 3: Cost-per-acquisition ,cost-per-conersion` in Google-ese, which is technically
correct, although the author treated conersion dierently in this campaign, or a
handul o keywords that the author was purchasing in the last hal o 2003 ,data
proided by Google Ad\ords,. .......................................................................................... 15
ligure 38: Domains reerring traic to BackcountryStore.com and the percent change rom
comparable timerames. \ebSideStory, BackcountryStore.com`s \eb analytics
endor, reers to isitors who hae typed the URL directly into the browser as haing
come rom Bookmarks or directly reerred URLs.` ...................................................... 160
ligure 39: Sample o traic rom major search engines to BackcountryStore.com.............. 161
go BIbIIogrupIy

ligure 40: Lxample reerring URLs that are the basis or the reerring search engines shown
in ligure 33...............................................................................................................................161
ligure 41: Lxample o naigation path rom reerring source rom the
BackcountryStore.com \eb site. .........................................................................................164
ligure 42: 1raic to BackcountryStore.com rom some o the major search engines showing
the Change` in isits rom each. ...................................................................................166
ligure 43: Search engine traic to BackcountryStore.com that dries isitors to purchase
,data proided by litBox Commerce,. 1he B,B` column is the browse-to-
buy` ratio, an indicator o how likely isitors rom each search engine are to purchase.
1he Reenue` and Visits` columns hae been blurred to protect
BackcountryStore.com`s priacy. .........................................................................................166
ligure 44: Search terms rom BackcountryStore.com`s internal search that suggested that
isitors were interested in buying paddle boats and gear rom them online. ...............169
ligure 45: BackcountryStore.com`s paddleboat category, prompted by reerring search
keyword and internal search actiity. ...................................................................................169
ligure 46: Campaign conersion rates rom the author`s \eb site. 1he act o conersion at
the time was the download o a preiew copy o \eb Analytics Demystiied............184
ligure 4: Cost-per-conersion inormation or a sample o keywords the author purchased
to drie leads to www.webanalyticsdemystiied.com ,data proided by Google
Ad\ords,. ................................................................................................................................186
ligure 48: Segment conersion report or BackcountryStore.com. O interest is the
dierence in conersions rom Unregistered ~ Customer` and Registered ~
Customer.` ...............................................................................................................................190
ligure 49: Key perormance indicator worksheet or an anonymous online business showing
AOV or all, new and repeat isiting customers................................................................192
ligure 50: Searches yielding results ,aboe, and one yielding no results ,below, rom
BackcountryStore.com. ..........................................................................................................196
ligure 51: Amazon.com`s method or starting a checkout or new isitors. All that is
required is an email address and you are told that eentually they`ll want you to add a
password. ..................................................................................................................................199
ligure 52: Lxample buttons rom an anonymous Luropean \eb site ,let, and
BackcountryStore.com ,right,. ..............................................................................................200
ligure 53: \hole screen rom BackcountryStore.com, note the Add to Shopping Cart`
button on the right side o the page. Note also the Send Lmail` button near the
lower-right portion o the screen, consistent in look but less pronounced..................201
ligure 54: Progress indicators rom BackcountryStore.com ,top, and Amazon.com ,bottom,,
both o which are easy to read and are accurate in terms o describing the number o
pages you`ll hae to get though to complete your purchase............................................202
ligure 55: Monthly returning isitor pattern or BackcountryStore.com. Note the spike in
returning isitors mid-month, attributed to retention marketing actiities. .................206
ligure 56: Loyalty measurement taken rom BackcountryStore.com showing the number and
percentage o isitors who come to the \eb site n` times. ..........................................210
ligure 5: Lxample o aerage requency o return or BackcountryStore.com. ...................211
ligure 58: Page actiity or BackcountryStore.com`s retained customer` segment ,let, and
all isitors` segment ,right,..................................................................................................215
ligure 59: Samples o Key Perormance Indicator ,KPI, reports. ...........................................223
ligure 60: General strategy or building KPI worksheets, highlighting data collected in this
period,` last period` and the percent change between the two. 1he data in yellow is
the raw data used to generate the KPI. ...............................................................................224
ndex g1

INIX

A
A,B testing ......................................... 8-80
abandonment rate...................................180
Aberdeen Group............. 5, 18, 31, 38, 235
Accrue......................................................... 18
G2 product ........................................... 18
acquisition
deined.................................................106
Actie Serer Pages ........................... 48, 49
Adobe PDl ............................................... 3
adertising
business model, deined...................226
Akamai ............................................... 90, 122
Allthe\eb ................................................164
Amazon ........................................... 8, 16, 6
American Society or Inormation
Science and 1echnology .......................... 63
Andromedia ............................................... 18
AOL................................................... 25, 164
Apache........................................................ 19
ARIA........................................................... 18
Ask Jeees ................................................164
aerage order alue.................................190

B
BackcountryStore.com3, 4, 66, 6, 4, 5,
6, 83, 84, 92, 100, 101, 189
Belree.......................................................23
Best Buy......................................... 64, 65, 91
BizRate.com.................................... 219, 23
BroadVision............................................... 63
Browser caching........................................ 24
Bryan Lisenberg.... 8, 124, 141, 13, 195,
198, 21
Bytes Receied ................................... 20, 21
Bytes Sent............................................ 20, 21

C
Campaign response metrics ..................151
CGI ............................................................. 49
CheetahMail .............................................23
click-through
deined............................................ 4-48
ClickZ................................................. 12, 212
Client IP Address............................... 19, 21
Cold lusion................................................49
common log ormat..................................19
ComScore................................................ 10
continuous improement process..........12
practical application ............................16
conersion
deined............................................ 106-
conersion paths
"special oer" to purchase.............. 15
home to purchase ............................. 14
naigation to inormation ............... 1
search to purchase............................ 15
conersion rate ....................................... 19
analysis using process measurement
tools..................................................2
and its application to content analysis
tools..................................................88
campaign conersion rates.............. 182
deined............................................ 53-54
cookies ........................................................21
acceptance o........................................35
as a unique identiier ...........................35
setting up a isit-cookie tracking
system......................................145-4
Coremetrics....................................... 31, 23
page tags identiied..............................33
cost-per-acquisition ,CPA,
adertising model, deined.............. 22
use in measuring acquisition........... 154
cost-per-click ,CPC,
adertising model, deined.............. 226
use in measuring acquisition........... 154
cost-per-thousand ,CPM,
adertising model, deined.............. 226
customer lie cycle
deined............................................ 104-5
Customer support
business model, deined .................. 22


Datanautics.................................................18
Digital Impact ......................................... 23
g ndex

directly reerred traic........................... 160
DoubleClick
their use o 3rd party cookies ........... 44

I
error pages
usage in measuring reach................. 123
LSPN............................................................ 8
Lxcel .....................................................3, 38

I
lireclick...................................................... 11
page tags identiied ............................. 33
loreSee Results....................................... 23
lorrester Research ................................... 95
rames
conounding eects on measurement
.......................................................... 49
requency
deined .................................................. 56

G
Gomez........................................................ 3
Google. 8, 90, 92, 93, 113, 126, 153, 156,
164, 16, 168, 10, 11, 184, 198
Adwords program............154, 15, 186
Ad\ords program.. 153, 156, 185, 186
1oolbar application .......................... 198
Guy Creese ................................... 5, 38, 235

H
hit
and their application in the pyramid
model o \eb analytics data ....... 5
deined .................................................. 4
lit List ....................................................... 18
litBox.......................i, 4, 46, 66, 10, 24
litBox Commerce................................. 166
lotBot ..................................................... 164
l1ML .... , 19, 22, 3, 4, 48, 49, 63, 126
l11P .......................19, 20, 21, 28, 52, 159
l11P Status .......................................20, 21
l11PS ..................................................... 242
lurol Inan.................................. 5, 110, 133
lyperion.................................................. 23

I
I,PRO...................................................... 132
IBM SurAid
page tags identiied.............................. 33
internal traic ..........................................160
Internet Inormation Serer ............ 19, 23
IP address................ 19, 20, 21, 25, 2, 122

J
Jakob Neilsen .............................................. 9
Jason Burby................................................ 12
Jim Noo...... 141, 195, 213, 21, 246, 248
Jim Sterne.....................3, 1, 148, 212, 213
Jupiter Research......................................10

K
key perormance indicators
use to measure retention..................220
used in measuring conersion.... 202-3
used to measure acquisition ..... 11-2
used to measure reach......................131
Keynote Systems...................................8, 3

L
lead generation
business model, deined...................22
Log Analyzer .........................................2, 1
Lord Leerhulme........................................ 1
Louis Roseneld ........................................ 63
loyalty
deined................................................... 55
Lycos.........................................................164

M
Macromedia ............................................... 18
Mark Sakalosky............................... 212, 213
Mark Veytsel ............................................235
Marketing Sherpa....................................10
Marketwae................................................ 18
Matt Cutler...............................................148
Matthew Berk............................................ 1
Mercury Interactie .................................. 3
ML1A tag .................................................. 28
Microsot...................................8, 19, 20, 23
MSN................................................. 164, 16

N
NCSA................................................... 20, 43
NetGenesis ................................................ 18
NetIQ ............................................ 1, 18, 31
ndex g

new isitors ..................................... 116, 118
impact on loyalty measurements ....209
used to measure acquisition ............133
Nielsen,NetRatings ...................... 132, 24

O
Omniture........................................... 31, 23
page tags identiied.............................. 33
online commerce
business model, deined...................226
open rate
deined.................................................12
OpinonLab................................................. 10
Oerture10, 48, 8, 92, 126, 168, 10, 11

P
P3P.............................................................. 44
page "stickiness" .....................................139
page iew
and their application in the pyramid
model o \eb analytics data........ 5
and their use in isitor segmentation82
aerage page iews per isit.............135
aerage pages iewed per isitor ....13
deined............................................ 48-50
measured by entry pages..................120
use in measuring reach.....................111
page weights.............................................8, 9
pay-per-click
ee cost-per-click.................................. 93
PLRL ......................................................2, 49
priacy policy
use in describing the use o cookies. 45
Protocol Version................................ 20, 21
Proxy caching ............................................ 24

R
recency
deined................................................... 56
reerrer
business alue o tracking.................. 21
deined............................................ 52-53
request.................................................. 19, 23
request or proposal ...............................101
Resource requested............................ 19, 21
response rate
and commonly obsered click-through
dropo ............................................ 90
and its application to campaign
analysis tools ...................................88
response time.............................................19
retention
deined................................................ 108
return on inestment .. 2, 89, 99, 154, 155,
158, 10, 11, 14, 18, 186, 18
returning isitors .................................... 116
ratio o daily to monthly returning
isitors........................................... 20
ratio o returning isitors to all isitors
........................................................ 208
used in measuring retention............ 205
Reenue Sciences ............................... 18, 30

S
Saleslorce.com....................................... 23
search engine optimization...........160, 164
and cost-per-click, pay-per-click
models.....................................10-1
search engines
as they relate to reerring traic..... 160
SearchLngine\atch.......................164, 24
Serer IP.............................................. 20, 21
Serer Name ....................................... 20, 21
Serice Name...................................... 20, 21
session tracking..........................................99
Shop.org............................................. 95, 13
Six Sigma.....................................................15
Speedera.................................... 90, 122, 23
sponsorship
adertising model, deined.............. 22
SPSS.............................................................18
SQL .............................................................23
Stanord Publishing group.................... 184
Stee Krug.......................................9, 60, 0
SurAid................................................. 18, 30

T
1oshiba .......................................................63

L
Unileer.........................................................1
unique identiied isitors
and their application in the pyramid
model o \eb analytics data ........59
URI Query........................................... 20, 21
usability studies............................................9
gq ndex

User Agent...........................................19, 21
user surey
practical application............................ 11
UUID .......................................... 21, 59, 205

V
isit.............................................................. 62
and their application in the pyramid
model o \eb analytics data ....... 58
and their use in process measurement
tools .................................................
and their use in isitor segmentation82
aerage page iews per isit ............ 135
aerage isits per isitor................... 134
use in measuring reach..................... 112
isitor .......................................................... 62
and their application in the pyramid
model o \eb analytics data ....... 58
and their use in process measurement
tools .................................................
and their use in isitor segmentation82
aerage isits per isitor................... 134
deined ............................................51-52
use in measuring reach..................... 114
isitor retention rate............................... 21
VisitorVille
page tags identiied.............................. 33

W
\eb analytics...........................................5-6
\eb serer perormance data................... 6
\eb traic data........................................... 6
\eb transactional data............................... 6
\ebSideStory ..i, 3, 1, 31, 10, 23, 24
page tags identiied.............................. 33
their use o 3rd party cookies............ 44
\eb1rends ....2, 11, 1, 18, 31, 46, 4, 5,
23
page tags identiied.............................. 33
\in32 Status....................................... 20, 21
\usage........................................... 2, 11, 24

X
XML..........................................................241

Y
\ahoo!........................ 8, 164, 16, 10, 24

Z
ZAAZ......................................................... 12