Predictive Analytics Explained and IBM SPSS Demo

May 30, 2013

Predictive Analytics

Predictive Models: What They Are, How to Use Them and Demo of SPSS

Many organizations have made big investments in business intelligence, deploying sophisticated reporting systems and performance dashboards. Yet for most companies, translating “interesting insights” into quantifiable business benefits is more the exception than the rule. Predictive analytics can be the next logical step in the evolution toward achieving dramatic improvements to the bottom line. In this webinar recording, Eric Zankman demystifies predictive analytics by explaining what predictive models are, how to develop them and how to apply them within a customer management framework to create measurable ROI.

Components of a comprehensive analytics approach are explored:

  • Creating a customer data mart with predictive analytics in mind
  • Building predictive models
  • Segmenting customers
  • Developing champion/challenger strategy tests
  • Establishing processes for continuous learning and improvement
  • Reaping the benefits of predictive models

The second half includes a demo of IBM SPSS Modeler.

Business Context

Predictive analytics can greatly improve profitability when part of a comprehensive analytics solution

  • A well-designed data mart is the first step toward effective predictive analytics
  • Organizations must be committed to ongoing strategy testing to maximize their benefits

Predictive Analytics, IBM SPSS Modeler


Business executives, managers and analysts who are responsible for devising and implementing a range of strategies that would be applied to different customer, product or other market segments to improve profitability.


Eric Zankman

Mr. Zankman is an analytics and business intelligence consultant with a 20-year track record of improving profitability for some of the world’s largest firms by applying data mining, predictive modeling, customer segmentation, experimental design, and optimization.

Mr. Zankman heads the predictive analytics practice for Senturus. He has founded Zankman Solutions, a provider of analytics consulting services, and served as Analytics Practice Leader, Coordinator of CRM Knowledge Center, and Senior Subject Matter Expert at American Management Systems (AMS), a global business and IT consulting firm.

Mr. Zankman has given numerous presentations at major industry conferences related to customer analytics. He has authored several white papers and thought leadership articles, including a Gartner publication on customer loyalty.

Mr. Zankman earned a BS degree from the Massachusetts Institute of Technology and an MBA from the Haas School of Business at UC Berkeley.

Arik Killion

Client Technical Professional, IBM Business Analytics, SPSS



General Definition

  • Empirically-derived algorithms used to predict future outcomes

Customer Analytics Definition

  • Predicts future customer actions
  • Combines individual attributes that are strong predictors
  • Produces an assessment score for each customer

Model Uses

  • Customer Analytics
    • Direct Marketing
    • Underwriting
    • Usage Stimulation
    • Cross-sell / Up-sell
    • Retention / Churn
    • Customer Value
  • Operational Analytics
    • Risk Management
    • Credit Policy Decisions
    • Channel Preference
    • Portfolio Management
  • Threat and Fraud Analytics
    • Fraud Detection
    • Collections
    • Write-offs / Recoveries
  • Predictive Model Timeline
    • Observation Period / Observation Point
    • Performance Period / Forecasts

Part of a Comprehensive Analytics Solution

  1. Customer Data Mart
  2. Predictive Models
  3. Customer Segmentation
  4. Champion / Challenger Strategy Tests
  5. Model, Segmentation and Strategy Execution
  6. Strategy Test Evaluation
  7. New Champion Strategy
  8. Back to Step 4 (continuous learning and improvement)

Predictive Model Development Methodology

  1. Define business goals
  2. Specify model objective function
  3. Design/build modeling database
  4. Partition modeling data
  5. Derive potential predictors
  6. Analyze predictor strength
  7. Perform sub-population analysis
  8. Build model algorithms
  9. Evaluate model performance

10. Deploy model

Champion / Challenger Strategy Improvement Methodology

  1. Develop a “champion” strategy and “challenger” strategies for each segment
  2. Execute strategy tests and analyze results after a defined test period
  3. Perform model validation (controlling for treatment)
  4. Deploy new champion strategy with quantifiable business improvements
  5. Create new round of promising challengers

Example Application: Customer Churn Reduction

  • Develop predictive models for customer value and churn
  • Identify customers with high value and/or high churn propensity for tailored treatments (e.g., special retention campaigns, VIP service, liberal fee-reversal policies)
  • Conduct champion/challenger tests to identify the best treatments for each segment
  • Implement new champion and develop next set of challengers

Demonstration: IBM SPSS Modeler

▼ Q&A


  • Groups and Point Assignments
    • For scoring models, there are tools within SPSS that help define the number of groups for an attribute (e.g., the breadth of each group: Customer Age groups of <1 yr; 2 to 3 yrs; 3 to 5 yrs; 6 or more) and the number of points that should be assigned to each group for scoring purposes.
  • Consulting Validation for Data Readiness
    • Senturus does offer this “discovery” call or visit as a service and it usually takes a few hours.
  • Modeling Technique Choice Criteria
    • Statistical Regression and Point Scoring models work well for binary outcome situations (e.g., a customer cancels an account or doesn’t; a customer defaults or doesn’t; etc.)
    • Decision Trees work well for multi-outcome situations
    • Neural Nets work well for a resulting list of suspects (as for fraud) as opposed to an off/on or yes/no evaluation for each customer
  • Customer Involvement
    • Usually significant over the 2 to 3 month implementation period for domain expertise; model validation; segmentation implementation; strategy determination, testing and strategy changes
  • Model Implementation Challenges
    • Business Use and Customer treatment (for example)
  • Model Validation Frequency: at least quarterly is recommended
  • How is SPSS software implemented: hosted or on-premise?
    • Both (see links in the slide deck)
  • Does SPSS require a data warehouse?
    • Not necessarily because the observation data can come from any source, but a customer DW – for example – provides other reporting and analysis benefits and is an excellent source for SPSS for it supports changing the combination and population of attributes that SPSS analyses.
  • Are the SPSS demonstrated charts and graphs available in Cognos 10 BI suites?
    • No. The SPSS data would have to be brought into Cognos BI, some calculations added, and the native Cognos charts and graphs would then display the SPSS data.
  • Is in-data base analytics recommended for SPSS?
    • Yes, this is one way SPSS has been architected. Most data bases have internal algorithms that can be turned on and utilized by SPSS for analysis. (As opposed to the older, perhaps “standard” approach of extracting data from the data bases and handing that data to SPSS to store and analyze.)
  • Can data marts and warehouses support predictive modeling?

Yes, but there are analysis and sampling requirements for making observations and getting attribute values that usually require multiple snapshots and date stamping them so that values can be obtained for a specific point in time. Such as the values on a day that a customer defaults on a payment, or the values on the day a customer cancels an account, or the first day of each month to obtain trended historical values in the observation period.