Predictive Analytics Demystified
6 likes1,498 views
This document discusses predictive analytics and predictive modeling. It begins with an agenda that outlines an introduction to predictive models, why organizations build them, comprehensive analytics solutions, benefits of models, and a demonstration of IBM SPSS Modeler. Predictive models are empirically-derived algorithms used to predict future outcomes like customer actions and risk levels. Building predictive models allows organizations to better target customers, improve strategies, and increase profitability. Effective predictive analytics requires developing predictive models, segmenting customers, testing strategies, and establishing a feedback loop for continuous improvement.
Predictive Analytics Demystified
- 1. 1 Helping Companies Learn From the Past, Manage the Present and Shape the Future www.senturus.com Predictive Analytics Demystified
- 2. 2 This slide deck is part of a recorded webinar. To view the FREE recording of this entire presentation and download the slide deck, go to http://info.senturus.com/2013-05-30-Predictive-Analytics-Demystified.html The Senturus library has over 90 other recorded webinars, whitepapers, and demonstrations on assorted Cognos and BI topics which may interest you. Go to the recorded resources
- 3. 3 Agenda • Introduction • What are Predictive Models? • Why Build Predictive Models? • Comprehensive Analytics Solution • Reaping the Benefits of Models • Live Demonstration: IBM SPSS Modeler • Questions and Answers
- 4. 4 Presenters Today Greg Herrera Co-Founder/CEO Senturus Arik Killion Client Technical Professional IBM Business Analytics, SPSS Eric Zankman Analytics and BI Consultant Senturus, Inc 20-year track record of • applying data mining • predictive modeling • customer segmentation, • experimental design and optimization
- 5. 5 Who is Senturus ? • Consulting firm specializing in Corporate Performance Management – Business Intelligence – Predictive Analytics – San Francisco Business Times Hall of Fame -- Four consecutive years in Fast 100 list of fastest-growing private companies in the Bay Area • Experience – 13-year focus on performance management – More than 1,200 projects for 650+ clients • People – Business depth combined with technical expertise. Former CFOs, CIOs, Controllers, Directors... – DBAs with MBAs www.Senturus.com 888.601.6010 info@senturus.com
- 6. 6 A few of our 650+ Clients
- 7. 7 What are Predictive Models? Empirically-derived algorithms used to predict future outcomes In the context of customer analytics, a model: • predicts future customer actions • combines individual attributes that are strong predictors • produces an assessment score for each customer
- 8. 8 Uses of Predictive Models • Direct Marketing • Underwriting • Usage Stimulation • Cross-sell / Up-sell • Retention / Churn • Customer Value Acquire Grow Retain Predictive Customer Analytics Predictive Threat & Fraud Analytics Monitor Detect Control Predictive Operational Analytics Manage Maintain Maximize • Risk Management • Credit Policy Decisions • Channel Preference • Portfolio Management • Fraud Detection • Collections • Write-offs / Recoveries
- 9. 9 Predictive Modeling Timeline: Data snapshots from multiple points-in-time used to simulate forecast Forecast Observation Period Performance Period Observation Point Future Customer BehaviorPast Customer Behavior
- 10. 10 Model Objective Function Specifies the Customer Behavior to be Predicted • Who is Included in the Customer Sample? – Define all Inclusions and Exclusions • How is Customer Performance Defined? – Define “desirable” customers – Define “undesirable” customers – Define “indeterminate” customers • How Long is the Performance Period? – Define the Observation and Performance Dates
- 11. 11 Example Objective Function for Churn Customer Performance (Pertains only performance period) – Good: Non-churn – Bad: Voluntary Churn – Indeterminate: Involuntary Churn (customer terminated for non-payment) Forecast Observation Period = prior to 1/1/2013 Performance Period = 1/1/2013 to 4/30/2013 Observation Point = 1/1/2013 Future Customer BehaviorPast Customer Behavior Customer Sample – Include current customers as of 1/1/2013 – Exclude customers with no purchase/payment activity in the 24 months prior to 1/1/2013
- 12. 12 Example Predictive Model for Churn Account Age Score Less than 1 year 41 1 to less than 2 years 58 3 to less than 5 years 97 5 years or more 102 Average Balance in the Last 6 Months $0 - $75 80 $75.01 - $120 49 $120.01 or more 41 Roaming Charges in the Last 3 Months $0 - $20 80 $20.01 - $40 68 $40.01 - $75 40 $75.01 or more 29 Dropped Calls in the Last Month 0 110 1 57 2 41 3 or more 30 Number of Score Premium Services 0 35 1 - 2 57 3 or more 69 Number of Customer Service Calls in the Last 3 Months 0 85 1 - 2 38 3 - 4 29 5 or more 18 Number of Customer Disputes in the Last 12 Months 0 132 1 - 2 97 3 or more 40 Total Score for example customer = 421
- 13. 13 Why Build Predictive Models? Predictive models harness the knowledge within your data about customer behavior so you can treat different customers differently, tailoring the right treatments and offers to the right customers, thereby improving customer strategies and increasing profitability
- 14. 14 Evaluating Model Performance 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% ACTUALCustomerBadRateatTIME1 FORECASTED Customer Score (in deciles) at TIME 0 • Customers were scored at TIME 0 and rank-ordered by Customer Score • Bars show actual bad rate at TIME 1 Low Score High Score
- 15. 15 Translating Insights into Action 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% ACTUALCustomerBadRateatTIME1 FORECASTED Customer Score (in deciles) at TIME 0 Treat different customers differently Low Score High Score
- 16. 16 This slide deck is part of a recorded webinar. To view the FREE recording of this entire presentation and download the slide deck, go to http://info.senturus.com/2013-05-30-Predictive-Analytics-Demystified.html The Senturus library has over 90 other recorded webinars, whitepapers, and demonstrations on assorted Cognos and BI topics which may interest you. Go to the recorded resources
- 17. 17 Predictive Models: Part of a Comprehensive Analytics SolutionCustomer Data Mart Customer Segmentation Champion/Challenger Strategy Tests Model, Segmentation, and Strategy Execution New Champion Strategy Strategy Test Evaluation Predictive Models Continuous Learning and Improvement
- 18. 18 Solution – Customer Data Mart Store the right data for predictive modeling and analysis Customer-level Data Summary Billing Data Detailed Transaction Data Predictive Model Development Strategy Test Evaluation Ongoing Analysis and Model Validation Custom er Data Mart
- 19. 19 Solution – Predictive Models Apply Modeling Methodology • Define business goals • Specify model objective function • Design/build modeling database • Partition modeling data • Derive potential predictors • Analyze predictor strength • Perform sub-population analysis • Build model algorithms • Evaluate model performance • Deploy model
- 20. 20 Solution – Customer Segmentation Create customer groups to enable differentiated strategies Apply model scores and other criteria to segment customers All Customers
- 21. 21 Solution – Champion/Challenger Tests Employ test-and-learn methodologies to evolve strategies • Develop a “champion” strategy and “challenger” strategies for each segment • Execute strategy tests and analyze results after a defined test period • Perform model validation (controlling for treatment) • Deploy new champion strategy with quantifiable business improvements • Create new round of promising challengers Model Score Low Medium High Champion (85%) Treatment A1 Treatment B1 Treatment C1 Challenger 1 (5%) Treatment A2 Treatment A1 Treatment C2 Challenger 2 (5%) Treatment B1 Treatment C1 Treatment B1 Challenger 3 (5%) Treatment B2 Treatment B2 Treatment B2
- 22. 22 Solution – Test and Learn Feedback Loop Establish culture and capabilities for continuous strategy improvement
- 23. 23 Reaping the Benefits: Churn Management Customer Segmentation Champion/Challenger Strategy Tests Model, Segmentation and Strategy Execution Strategy Test Evaluation New Champion Strategy Value and Churn Predictive Models Customer Data Mart Continuous Learningand Improvement Solution 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 Performance Measures Customer value Churn rates Average Revenue per User (ARPU) Customer tenure Average number of products Retention costs Challenges Rising churn rates Declining revenue per customer High costs to acquire new customers
- 24. 24 Predictive Analytics - Summary • Predictive analytics can greatly improve profitability when part of a comprehensive 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
- 25. 25 This slide deck is part of a recorded webinar. To view the FREE recording of this entire presentation and download the slide deck, go to http://info.senturus.com/2013-05-30-Predictive-Analytics-Demystified.html The Senturus library has over 90 other recorded webinars, whitepapers, and demonstrations on assorted Cognos and BI topics which may interest you. Go to the recorded resources
- 26. 26 DEMONSTRATION: IBM SPSS MODELER
- 29. • Through gut instinct and hypothesis – this analyst concludes subscribers that are Single and pay by Credit Card have the largest attrition • Another analyst might arrive at a completely different conclusion
- 33. 33 Other Resources In-Person IBM/SPSS Events IBM Business Analytics Summit June 13, 2013 San Francisco • complimentary, one-day event • Emphasis on SPSS with many breakout sessions focused on predictive analytics • Click below to go to IBM's website and learn more about the event and register: http://www.senturus.com/continue.php?link=aHR0cHM6Ly93d3cuaWJtLmNvbS9ldmVudHMvd3dlL2dycC9ncnA wMDQubnNmL2FnZW5kYT9vcGVuZm9ybSZzZW1pbmFyPUIyRktSM0VTJmxvY2FsZT1lbl9VUyZTX1RBQ1Q9QlBfU 2VudHVydXM= SPSS Modeling Workshops • San Francisco, CA June 18 • Costa Mesa, CA June 19 • Minneapolis, MN June 20 Dates being finalized for workshops in July – December throughout the U.S. and Canada Email to follow with details.
- 34. 34 Other Resources: IBM Analytic Answers • Prepackaged solutions • Cloud based Analytic Answers for Student Retention http://public.dhe.ibm.com/common/ssi/ecm/en/ytd03292usen/YTD03292USEN.PDF Analytic Answers for Prioritized Collections http://public.dhe.ibm.com/common/ssi/ecm/en/ytd03291usen/YTD03291USEN.PDF Analytic Answers for Retail Purchase Analysis and Offer Targeting http://public.dhe.ibm.com/common/ssi/ecm/en/ytd03289usen/YTD03289USEN.PDF Analytic Answers for Insurance Renewals http://public.dhe.ibm.com/common/ssi/ecm/en/ytd03290usen/YTD03290USEN.PDF
- 35. 35 This slide deck is part of a recorded webinar. To view the FREE recording of this entire presentation and download the slide deck, go to http://info.senturus.com/2013-05-30-Predictive-Analytics-Demystified.html The Senturus library has over 90 other recorded webinars, whitepapers, and demonstrations on assorted Cognos and BI topics which may interest you. Go to the recorded resources
- 36. 36 Helping Companies Learn From the Past, Manage the Present and Shape the Futurewww.senturus.com 888-601-6010 info@senturus.com Copyright 2013 by Senturus, Inc. This entire presentation is copyrighted and may not be reused or distributed without the written consent of Senturus, Inc.
- 37. 37