Professional Documents
Culture Documents
AnswerTree 3.0
AnswerTree 3.0
Reveal segments
to your promotions
Use results to make a difference So you have information that empowers you to:
in your customer relationships ■ Develop better-targeted direct mail programs
AnswerTree finds segments and patterns in your data and gives ■ Offer specific products to the customer groups most likely
you information you can use. For example, AnswerTree analyzes to purchase them
your data to: ■ Create campaigns to increase customer retention
■ Discover which customer groups buy specific products ■ Improve student retention and graduation rates
■ Identify which customers will most likely defect ■ Develop more effective government programs and services
■ Predict which customers are likely to repay their loans on time ■ And more
ÂÊÊ
We developed (AnswerTree) models that enabled UNICEF Germany to identify optimal target
ÂÊÊ
and individual student performance without
having to be a world-class statistical genius.
– Chris Bowman,
Technology Manager,
Lafourche Parish School Board
Decision trees start with a root node and split into branches — from top to bottom, by order
of variable importance. In this example, AnswerTree analyzes data for a direct marketing
campaign to determine the target variable — which group of households will most likely
respond. The decision tree shows the prime candidates are in households in which: income
is >$49,999, a bankcard is present, children are present, the age of the household head is
>18-30 and there are >3 people.
Leverage your data across your organization
with scalability Build better models with
Because AnswerTree’s algorithms are built with scalability in unprecedented analytical
mind, you can work with your data more efficiently. But sometimes
power
Ê
you need more processing power than your desktop computer
provides — especially if you’re working with large datasets, such Different types of data work better with
as a catalog mailing list. If your data has outgrown your desktop
different algorithms, and your organization
computer, run AnswerTree on a server.
likely has many types of data — which
You can leverage your IT investment when using a larger server
change over time. Therefore, you need the
machine with a client-server version of AnswerTree installed. The
ability to try different types of decision-tree
distributed architecture means you process data where they can
algorithms with your data to find the best fit.
be handled more effectively. For example, if you have multiple
users, you can analyze data on the server rather than running AnswerTree gives you four powerful algo-
data on users’ individual machines. rithms that empower you to select the right
one for your specific data. For example, you
can choose one of the four algorithms and
Act on results quickly
build a model, then compare it against
See the information that enables you to act on results quickly — using
another algorithm to determine which one
AnswerTree’s unique evaluation graphs, including gains, response, lift,
works best for your dataset. Whether you’re
profit and ROI charts. These at-a-glance charts provide summaries of
selected segments — giving you a clear picture of your results. For analyzing purchase amounts, product cate-
example, looking at a gains chart, a telecommunications company could gories, demographics or satisfaction ratings,
determine what percentage of likely churners to target when creating a only AnswerTree gives you the widest range
customer retention campaign. of decision-tree algorithms available today:
Situation A regional branch of a national A financial services company A social services agency, which
insurance company faced high needed to maintain high customer is responsible for paying benefits,
churn rates. acquisition and retention rates — needed to ensure that people
all while keeping marketing who recently left its programs
costs low to maintain margins. remained self-sufficient.
Critical The company needed to under- In the past, the company mass- The agency needed to maximize
issue stand which customers left the mailed marketing materials to all the efficiency of its programs
organization and which groups types of customers — spending and services. It needed to
of customers were likely to defect. a lot of money mailing offers to understand how more effective
It also needed to understand the people who weren’t likely to program management can
factors that led to defection. ever respond. empower the agency to reach
its goals — ensuring the self-
sufficiency of people who
recently exited a program.
Solution The company used AnswerTree, The company selected The agency used AnswerTree,
which enabled it to: AnswerTree, which enabled it to: which enabled it to:
■ Build customer profiles ■ Uncover customer segments ■ Build profiles of participants
■ Understand the complexities within its current customer ■ Evaluate the programs people
Results ■ Better understood how premi- ■ Reduced key marketing costs ■ Isolated factors that lead
ums, tenure and group rates by 30 percent because it sent to program re-entry
affect churn fewer mailings to a more tar- ■ Assessed individual program
■ More accurately predicted geted audience success when it understood
churn rates among customer ■ Boosted campaign profitability which groups of people
groups and modified controlled due to a higher rate of return entered and re-entered
factors to reduce churn programs
Feature overview
Trees ■ Gains chart: identify segments by highest (and lowest) contribution and select
■ Display tree diagram, tree map, bar graphs and data nodes using this criteria
■ Build trees easily using a wizard that prompts you through the ■ Summary report: document analysis results as well as the criteria used to
model-building steps build trees
■ Choose from three tree-generating methods: automatic, interactive
target variables, tables or both in each node — Trees as Windows bitmap (BMP) or meta files
■ Collapse and expand branches without deleting the model itself — Gains charts and risk summary tables as tab-delimited text files
■ View and print trees horizontally or vertically — Rules and summaries as text files
■ Print large trees more easily using the print preview — Trees, gains charts and risk, rule and summary tables as HTML
■ Specify the exact percent you want to zoom in on models ■ Export decision rules that define selected segments in SQL to score databases
■ Re-run tree building using the production mode; generate scripts or SPSS syntax to score SPSS files
automatically from the user interface or edit models directly from the script ■ Export XML models for use with SmartScore, a software development kit
— CHAID by Kass (1980) centers and Web sites for automatic scoring
— Exhaustive CHAID by Biggs, de Ville and Suen (1991) — Customize and integrate into every point of your decision-making process
■ Methods for handling missing data: assign to a category or impute by surrogate ■ Get native access to database management systems including Oracle, SQL Server,
■ Automatic discretization of continuous variables according to the number of DB2; additional access to any ODBC-compliant sources using the ODBC Wizard
categories the user specifies
■ Partition data between training and test data to verify model accuracy System requirements
■ Cost complexity pruning for C&RT and QUEST AnswerTree client:
■ Random sampling of source data ■ Operating system: Windows 98, 2000 or Windows NT 4.0 with Service
■ Pruning: select subtree based on either standard error rule or minimum risk Pack 5 or higher
■ Stopping rules control the following settings: ■ Hardware: Pentium 90 or higher processor, SVGA monitor and CD-ROM drive
— Maximum tree depth by maximum number of levels or minimum number for installation
of cases ■ Minimum free drive space: 70MB for software
— C&RT: specify the minimum change in impurity ■ Minimum RAM: 32MB; 64MB is required for Windows 98 only
Scalability
■ Algorithms made more scalable to better handle large datasets AnswerTree server:
■ New server-tier processing added to increase scalability: ■ Windows NT Server, Windows 2000 Server or Windows 2000 Advanced Server
— Decrease time for analyses with larger datasets — Hardware: Pentium 90 or higher processor, SVGA monitor and CD-ROM drive
— Reduce network traffic by enabling multiple users to analyze data on — Minimum RAM: 64MB for the server
the server rather than bringing data to each user’s machine for analysis ■ Solaris 2.6, 7 and 8:
— Hardware: Ultra Sparc 2 (or better) and CD-ROM drive for installation
■ Interactive evaluation graphs enable visual representation of gains summary table: — Minimum RAM: 256 MB
risk estimates
SPSS is a registered trademark and the other SPSS products named are trademarks of SPSS Inc. All other names are trademarks of their respective owners. Printed in the U.S.A. © Copyright 2001 SPSS Inc. AT3BRO-0701W