You are on page 1of 4

HOW CMOS CAN MEASURE RETURN ON AD SPEND WITH BETTER MODELING AND CONVERSION ATTRIBUTION

by Doug Bryan, Vice President, Analytics, iCrossing

ICROSSING POV:

CMOs are getting bad information on the value of their digital ads. Here's why: people online are usually exposed to more than one marketing message before their exploration of your brand converts to a transaction but most reporting systems give all the credit for the transaction to the last message that the consumer saw. As a result, marketers have a hard time knowing how to wisely allocate their digital advertising resources because the value of some ads is overstated while the value of others is overlooked.
But CMOs have reason to hope: marketing analytics specialists are trying to determine the true value of advertising through media mix modeling and conversion attribution. It's important that CMOs understand those techniques and align their customer value key performance indicators across channels. Media mix modeling uses statistical methods to allocate promotion resources to advertising media channels such as TV, Internet, newspapers, radio, and magazines. By contrast, conversion attribution allocates channel resources to individual ads, like the paid search keyword "chefs knives" at Google.com. The following table illustrates these levels of granularity:

MARKETING MIX
Product Price Placement Promotion

MEDIA MIX
TV Internet Print Radio

CONVERSION ATTRIBUTION
"chefs knives" paid search broad match Yahoo.com 300 x 250 display ad "Wsthof knives" paid search exact match

The key differences between media mix modeling and conversion attribution are the amount of data used and, consequently, the statistical techniques applied.

MEDIA MIX MODELING


Marketers use media mix modeling to measure the relationship between each channel's ad spend and their value generated (such as revenue) by analyzing spend per week per channel and comparing that spend to the overall value generated. Usually value increases as spend increases. But the ratio of value to spend can vary greatly by channel and by spend amount. For example, a certain display ad network might generate 500 percent value/spend while paid search marketing generates 800 percent. But if paid search spend doubles, then the law of diminishing returns takes effect, and paid search drops to 400 percent value/spend.

ICROSSING. ALL RIGHTS RESERVED.

HOW CMOS CAN MEASURE RETURN ON AD SPEND WITH BETTER MODELING AND CONVERSION ATTRIBUTION

SEPTEMBER 2011

Marketers can use advanced media mix models to go a step further and find the interactions between channels. TV ads, for example, can directly affect search engine queries for an advertiser's brand name, increasing the volume of searches for a particular term, and, in turn, increasing the paid search channel's cost. In fact, 60 percent of TV viewers now use the Internet while watching TV. Viewers can and do use Google and Bing during TV commercials.1 Advanced media mix modeling measures both the direct return on ad spend (ROAS) of such TV ads and the indirect ROAS of the additional paid search traffic. Internet advertisers can learn from media mix models since there are strong interactions between the Internet channels themselves; perhaps the strongest interaction is between display and paid search. Studies have shown that display ads can increase brand name searches by 52 percent2 and paid search conversion rates by 56 percent.3 And display ads together with paid search can increase online sales 50 percent and offline sales 33 percent compared to paid search alone.4 Although media mix modeling provides lessons on channel relationships, it fails to leverage one of the Internet's main competitive advantages: granular data. Conversion attribution relies on that wealth of data and moves the analysis from channels to individual ad placements.

CONVERSION ATTRIBUTION
Media mix focuses on channels by week. Conversion attribution analyzes events by customer to understand the value of individual ad placements. Consider a multi-channel example: + + + A customer types in "chefs knives" at Google.com and clicks on a paid search ad. The customer browses the advertiser's website for a while but doesn't buy anything The customer sees a dozen display ads for the advertiser over the next week, finally clicks one, and browses awhile but again does not buy The next day, the customer goes to Yahoo.com, enters the advertiser's name, clicks the top natural search result, and makes a purchase

How much credit for the conversion should the knives' paid search ad, the display ad, and the natural search result get? It's a tough problem, but not a new one. Banks and telcommunications companies addressed this issue 10 years ago with multi-touch campaigns using automated phone callers, direct mail, bill inserts, out-bound call centers, in-bound call centers, and in-store or in-branch cross-selling. Marketers can apply three levels of sophistication to unravel the attribution problem: 1. 2. 3. winner-take-all even split increment lift

Winner-take-all is commonly used today and gives all the credit to the first click, first impression, last click or last impression. It's nice and simple, and doesn't require much data. Last century, Einstein advised us to "make things as simple as possible but not simpler." So is winner-take-all too simple? Yes; we can do better and improve ROAS because winner-take-all tends to over-value natural and paid search brand keywords and under-value display ads. The even split approach makes few assumptions and simply splits credit equally between all the ad placements involved in a conversion. In the example cited about chefs' knives, the "chefs knives" paid search ad gets 1/14th the credit, each of the 12 display ads would get 1/14th, and the natural search result gets 1/14th. Credit doesn't change with temporal order since using order wouldn't be as simple as possible, and there's little evidence that temporal sequence matters.5

1 2 3 4 5

Edmund Lee (2011) Among media, TV is still on top, Advertising Age, March 29. Source: Nielsen comScore (2008) Whither the Click? comScore Brand Metrix norms prove 'View-Thru' value of online advertising, November 17 John Lovett (2009) A Framework for Multicampaign Attribution Measurement, Forrester Research Magid Abraham (2008) The Off-Line Impact of Online Ads, Harvard Business Review, April Demetrios Vakratsas and Tim Ambler (1999) "How advertising works: What do we really know?" Journal of Marketing, 63(1):2643

ICROSSING. ALL RIGHTS RESERVED.

HOW CMOS CAN MEASURE RETURN ON AD SPEND WITH BETTER MODELING AND CONVERSION ATTRIBUTION

SEPTEMBER 2011

Lastly the incremental lift approach attacks ROAS by measuring the value of ads relative to no ad rather than relative to other ads. To understand incremental lift, consider an illustrative if absurd story: Ms. Wilson owns an ice cream shop in a small town strip mall. She hires two summer interns, gives them each $1,000 to spend on advertising, and tells them that whoever drives the most sales will get a bonus. One intern immediately runs out the door to go price newspaper inserts, radio spots, and other channels, while the second intern hangs back and asks, "How will sales be measured?" Ms. Wilson, wanting to make things as simple as possible, says she'll just randomly ask some customers if they saw any of her ads and whichever intern's ads are mentioned most often wins. So the second intern rents a big billboard at the mouth of the only driveway to the strip mall, making it impossible to get to the ice cream shop without going right past his ad, and wins the contest. But what was the ROAS of the billboard? Did it increase sales or just increase impressions? Incremental lift modeling measures the sales the billboard generated compared to Ms. Wilson not renting the billboard at all.

MORE GRANULAR CUSTOMER VALUE


Both media mix modeling and conversion attribution require common value metrics across advertising channels. And since the Internet allows us to easily track many kinds of customer actions, savvy marketers are adopting those models as a catalyst for increasing the granularity of their customer value metrics. The following tables list value granularity examples from a few industries:
RETAIL 1. register 2. newsletter signup 3. wish list signup 4. gift registry signup 5. gift card purchase 6. catalog request 7. find a store 8. estimated revenue next 12 months 9. estimated lifetime value CREDIT CARDS 1. start an application 2. complete an application 3. approved accounts 4. balance transfer amount 5. spend and fees the first month 6. estimated spend and fees next 12 months 7. spend and fees next 12 months adjusted by credit risk 8. spend and fees next 12 months adjusted by estimated tenure SUBSCRIPTIONS 6 1. leads 2. start free trail 3. start paid subscription 4. estimated tenure CPG 1. view recipe 2. print recipe 3. tell a friend 4. print coupon 5. register 6. newsletter signup 7. redeem coupon B2B 1. sales lead 2. need identified 3. qualified prospect 4. proposal sent 5. negotiation 6. closing 7. closed 8. revenue 9. estimated lifetime value

This proliferation of metrics provides new opportunities to understand customer behavior. Which of these metrics matter most, and which are levers that can be adjusted to have meaningful impact on marketing programs, are open questions that each brand will have to answer for itself. Some leading marketers are doing just that working with the data, testing, experimenting, and analyzing. They are working to uncover the metrics that matter most to their business so that they can improve and optimize their programs and spend their marketing dollars more wisely.

Subscriptions are goods or services that are billed periodically such as a magazine, cell phone service, cable TV, Netflix and Hulu Plus.

ICROSSING. ALL RIGHTS RESERVED.

HOW CMOS CAN MEASURE RETURN ON AD SPEND WITH BETTER MODELING AND CONVERSION ATTRIBUTION

SEPTEMBER 2011

WHAT YOU CAN DO


Many advertisers do not yet have the data collection or business metrics in place to support media mix modeling and conversion attribution. To assess if you are ready, ask yourself: + + + What web conversion tracking systems do you use? Many websites use different systems per ad channels, making channel results difficult to compare. Which visitor actions does each channel track? For example, if paid search is tracking newsletter signups, are natural search and display ads also tracking it? What attribution schemes do the tracking systems use?

Business questions to ask include: + + + Is there agreement within your organization on what characterizes a good customer and a bad customer? How often do you compare ROAS across ad channels? How often do you measure the lifetime value of your customers?

Once tracking systems and business metrics are in place, Internet advertisers can apply media mix modeling and conversion attribution to increase their return-on-ad-spend and better compete with offline media. As this race heats up, marketers will reap the benefits. Rather than make big marketing guesses, or being inundated with data that tells them nothing, marketers will be equipped with the insights they need to increase their business results.

STAY CONNECTED
Find out more at www.icrossing.com Call us toll-free at 866.620.3780 Email us at findout@icrossing.com Follow us on twitter @icrossing and @thecontentlab Become a fan at facebook.com/icrossing

ICROSSING. ALL RIGHTS RESERVED.

You might also like