WIXLIB

Predictive Analytics Applied:Marketing and WebBrought to you byPrediction Impact andWorld Organization ofWebmasters (WOW) Copyright 2008. All Rights Reserved.Eric Siegel, Ph.D.Prediction Impact, Inc.eric@predictionimpact.com(415) 683 - 1146Prediction Impact, Inc.

Training Program Outline1.2.3.4.5.IntroductionHow Predictive Analytics WorksApp: Customer RetentionApp: Targeting AdsSummary and take-aways Copyright 2008. All Rights Reserved.Prediction Impact, Inc.1

About the speaker: Eric Siegel, Ph.D.– President of Prediction Impact, Inc.– Training and services in predictive analytics– Former computer science professor atColumbia University– Before Prediction Impact, cofounded twosoftware companiesEric Siegel, Ph.D.Prediction Impact, Inc.eric@predictionimpact.com(415) 683 - 1146 Copyright 2008. All Rights Reserved.Prediction Impact, Inc.

Predictive Analytics:Business intelligence technology thatproduces a predictive score for eachcustomer or prospect3412 Copyright 2008. All Rights Reserved.27794259Prediction Impact, Inc.

Optimizing Business Processes“At a time whencompanies in manyindustries offer similarproducts and usecomparable technology,high-performancebusiness processes areamong the last remainingpoints of differentiation.”- Tom Davenport Copyright 2008. All Rights Reserved.Prediction Impact, Inc.4

Learn from Organizational ExperienceData is a core strategic asset –it encodes your business’ collective experienceIt is imperative to:Learn from your data.Learn as much as possible about your customers.Learn how to treat each customer individually. Copyright 2008. All Rights Reserved.Prediction Impact, Inc.5

Predictive Analytics:Your organization learnsfrom its collective experienceCollective experience: Sales records, customer profiles, Learn: Discover strategic insights and business intelligenceDiscover something that makes business decisionsautomaticallyDiscover business rules and puts this knowledge to action. Copyright 2008. All Rights Reserved.Prediction Impact, Inc.6

How Predictive Analytics WorksBuilding and Deploying Predictive ModelsEric Siegel, Ph.D.Prediction Impact, Inc.eric@predictionimpact.com(415) 683 - 1146Predictive Analytics for Business, Marketing and Web Copyright 2008. All Rights Reserved.Prediction Impact, Inc.

Wisdom Gained: A Predictive Model is ct logs Copyright 2008. All Rights Reserved.ModelingtoolPredictivemodel Performed by modeling software Usually offlinePrediction Impact, Inc.8

Applying a Model to Score a vior Copyright 2008. All Rights Reserved. Often performed bymodeling software Possibly onlinePredictedresponsePrediction Impact, Inc.9

Deploy to Take Business ActionBusinesslogicPredictedresponse Copyright 2008. All Rights Reserved. Usually not performed bymodeling software Possibly onlineBusiness actions,such as: Mail asolicitation Suggest across-sell option Retain with apromotionPrediction Impact, Inc.10

Fine-Tuned Models for Your BusinessPredictive models are created automatically from your dataSo, they’re generated according to your:––––ProductPrediction goalBusiness modelCustomer baseThis means customer intelligence specialized for your business– Customized business rules– Unique, proprietary mailing lists– Insights only your organization could possibly gain Copyright 2008. All Rights Reserved.Prediction Impact, Inc.11

Predictors: Building Blocks for Models recency – How recent was the last purchase? personal income – How much to spend?Combine predictors for better rankings:recency personal income2 recency personal income Copyright 2008. All Rights Reserved.Prediction Impact, Inc.12

Training Data Is a Flat kelyto es46glovesFlowpianoThis is the required form for training data; one record per customer. Copyright 2008. All Rights Reserved.Prediction Impact, Inc.13

Why Not Memorize The Training Examples?To learn from history is to remember history.So, we could just keep a lookup table and compare newcases to old ones.Answer: There are too many possible rows of data.That is, each customer is unique! Copyright 2008. All Rights Reserved.Prediction Impact, Inc.14

Decision Tree for Cross SellDecisionTreesBought shoes?Female?HatGlovesHigh Income?Shoes Non-linear Specialized PrecisePianoIf they bought shoes and they’re not female then sell gloves.If they didn’t buy shoes and they’re high income then sell shoes. Copyright 2008. All Rights Reserved.Prediction Impact, Inc.15

Response Modeling for Direct MarketingLower campaign costs and increase response rates Make campaigns more targeted and more selectiveIdentify segments five or more times as responsiveAchieve 80% of responses with just 40% of mailingIncrease campaign ROI & profitability Copyright 2008. All Rights Reserved.Prediction Impact, Inc.16

Predictive Models Score Each Customer Each customer is scored with a response probability The customers are listed in order of prediction score The highest-scoring customers are targeted firstID number:429528256674Name:E. SiegelD. LeongG. ClooneyT. Mitchel Copyright 2008. All Rights Reserved.Score:35%31%22%14%Response:Yes "No !Yes "No !Prediction Impact, Inc.17

300,000200,000100,0000ProfitCampaign Profit 00-600,000-700,0000% Copyright 2008. All Rights Reserved.25%50%75%100%Percent of Customers ContactedCustomers ranked with predictive analyticsCustomers not rankedPrediction Impact, Inc.18

Lift Chart30.643.551.170.180.2100% 0800100% Population 50,000 customers 3 times as likely to buy 200,000 customers 2 times as likely to buy Copyright 2008. All Rights Reserved.Prediction Impact, Inc.19

Example Business RuleNew customers who come to the websiteoff organic search results, buy more than 150 on their first transaction, are male,and have an email address that ends with“.net” are three times as likely to bereturn customers. Copyright 2008. All Rights Reserved.Prediction Impact, Inc.20

App: Customer RetentionRetain Customers by Predicting Who Will DefectEric Siegel, Ph.D.Prediction Impact, Inc.eric@predictionimpact.com(415) 683 - 1146Predictive Analytics for Business, Marketing and Web Copyright 2008. All Rights Reserved.Prediction Impact, Inc.

Growth Acquisition - DefectionAcquisition:New customersCustomerbaseDefection:Lost customersWhich is the best way toincrease growth?1. Increase acquisition2. Decrease defection Copyright 2008. All Rights Reserved.Prediction Impact, Inc.22

Increase retention by 3%and boost growth by 12%3% Copyright 2008. All Rights Reserved.Prediction Impact, Inc.23

Predictor Variables Age of membershipExternal acquisition sourceU.S. state of residenceWilling to receive postal mailing (opt-in)Num days since last failed login attemptNum days between logins Copyright 2008. All Rights Reserved.Prediction Impact, Inc.24

At-Risk Segment SOURCE COBRAND chatSUBSCR AGE 237.819LOGIN DAYSSINCELAST F 1.85344LOGIN DAYSSINCELAST F 22.6578Churn rate: 33.5% (vs. 20% average) Copyright 2008. All Rights Reserved.Prediction Impact, Inc.25

Loyal Segment STATE is one of many, including ME, NE, NH, NS, QC, SK, WYSOURCE COBRAND chatSUBSCR AGE 237.819TIME SINCE JOINED 535.456LOGIN DAYSSINCELAST F 99.6539LOGIN DAYSSINCEFIRST F 112.003LOGIN DAYSSINCELAST S 131.45Churn rate: 6.5% (vs. 20% average)Very loyalSmall segment Copyright 2008. All Rights Reserved.Prediction Impact, Inc.26

Forecasted Profit of Retention Campaign Copyright 2008. All Rights Reserved.Prediction Impact, Inc.Cost per mailing: 25, ave profit: 100, num subscribers: 100,00027

Loyal Segment: Online Retail ITEM QTY 2.5 TAX SHIP 8.78 19% will return (versus 10% average) Copyright 2008. All Rights Reserved.Prediction Impact, Inc.28

App: Targeting AdsPredictively Modeling Which PromotionEach Customer Will AcceptEric Siegel, Ph.D.Prediction Impact, Inc.eric@predictionimpact.com(415) 683 - 1146 Copyright 2008. All Rights Reserved.Prediction Impact, Inc.

Learn More From AB Testing:“Dynamic AB Selection”Target content––––Landing pageFeatured productColor of product displayedPromotion oradvertisementAB?C Copyright 2008. All Rights Reserved.Based on:– Customer behavior profile Browsers vs. hunters––––Time of dayGeographical locationPagesWhere they came from Ad that clicked them throughSite they came fromSearch query they were onProfile, when availablePrediction Impact, Inc.30

Dynamic AB Selectionoffline onlineTrials of AModelingtoolPredictivemodel for APredictive score:What’s the chanceMary responds to A?Mary’sprofileTrials of B Copyright 2008. All Rights Reserved.ModelingtoolPredictivemodel for BPredictive score:What’s the chanceMary responds to B?Prediction Impact, Inc.31

Selecting Between 291 Sponsored PromotionsClient: A leading student grant and scholarship search serviceSponsors include:– Universities– Student loans– Military recruitment– Other misc. Copyright 2008. All Rights Reserved.Prediction Impact, Inc.32

The Business: Great Potential Gain High bounties, up to 25 per acceptance High acceptance rates, up to 5%, due togeneral relevancy of the promotions Big Data: Wide and Long– Rich user profiles volunteered for funds eligibility– Over 50 million training cases: 01/05 - 09/05 Copyright 2008. All Rights Reserved.Prediction Impact, Inc.33

Already seen? N/YEmail opt-in Y/NPrior solicitation of this programOpt in %YN3.00%3.00%2.00%2.00%1.00%1.00%0.00%0.00%Region: S, NE, W, M-WFraternity/None/SororitySority or FraternityRegion of current 0%southnortheastwestmidwest 4.00%2.00%3.00%fratnonesoro2.00%1.00% Copyright 2008. All Rights Reserved.0.00%1.00%0.00%Prediction Impact, Inc.34

Highly Responsive Segment for "Art Institute" AdSegment definition included: Still in high school Expected college graduation date in 2008 or earlier Certain military interest Never saw this ad beforeSegment probability of accepting ad: 13.5%Average overall probability of accepting ad: 2.7% Copyright 2008. All Rights Reserved.Prediction Impact, Inc.35

Highly Responsive Segment for a "Navy" AdSegment definition included: Has opted in for emails Has not seen this ad before Is in college SAT verb-to-math ratio is not too low, nor too high SAT written is over 480 ACT score is over 15 No high school name specifiedSegment probability of accepting ad: 2.6%Average overall probability of accepting ad: 1.6% Copyright 2008. All Rights Reserved.Prediction Impact, Inc.36

Money-Making Model:Improved Resulting RevenueA-B test deployment to compare:A. Legacy system based on acceptance rates across usersB. Model-based ad selectionResult:25% increased acceptance rate3.6% increased revenue observed; 5% later reported by the clientThis comes to almost 1 million per year in additional revenue,given the existing 1.5 million monthly revenue. Copyright 2008. All Rights Reserved.Prediction Impact, Inc.37

ConclusionsSummary and take-awaysEric Siegel, Ph.D.Prediction Impact, Inc.eric@predictionimpact.com(415) 683 - 1146Predictive Analytics for Business, Marketing and Web Copyright 2008. All Rights Reserved.Prediction Impact, Inc.

Predictive Analytics Initiatives Per-customer predictions: Unlimitedrange of business objectives Management: An organizational processensures predictions are actionable, drivenby business needs; multiple roles andskills Deployment: Mitigate risk and trackperformance Copyright 2008. All Rights Reserved.Prediction Impact, Inc.39

Predictive Analytics Technology Data preparation: An intensivebottleneck, critical to success Modeling methods and tools: No onemethod that is always best; comparemultiple methods Copyright 2008. All Rights Reserved.Prediction Impact, Inc.40