Five Ways To Lower Costs With Predictive Analytics PDF Free Download

1y ago

23 Views

1 Downloads

5.97 MB

56 Pages

Report/dmca

Download PDF

Transcription

Five Ways to Lower Costswith Predictive AnalyticsPredictive Analytics WorldOctober 20, 2009 Copyright 2009. All Rights Reserved.Eric Siegel, Ph.D.Program ChairPredictive Analytics WorldPredictive Analytics World

Instructor: Eric Siegel, Ph.D.Program Chair: Predictive Analytics WorldPresident: Prediction Impact, Inc.pawcon.comInstructor: predictive analytics trainingFebruary 16-17, 2010 Former computer science professor:Columbia UniversitySan Francisco, CAEric Siegel, Ph.D.Program ChairCofounded softwarecompanies:Predictive Analytics WorldUser profiling and data mining Copyright 2009. All Rights Reserved.1Predictive Analytics World

Agenda What is predictive analytics? How does it provide value? 5 ways to lower costs with predictiveanalytics Hot methods in predictive modeling Copyright 2009. All Rights Reserved.5Predictive Analytics World

Predictive Analytics:Your organization learnsfrom its collective experience.Customerdata Copyright 2009. All Rights Reserved.PredictivePredictive modelingmodeling7PredictivemodelPredictive Analytics World7

IMPACT OF INDIVIDUAL DECISIONMany Small Decisions Add UpHighvalueStrategicTacticalOperationalLowvalueLow volumeDecision automationvia predictive analyticsHigh volumeDECISION VOLUMEFrom Smart (Enough) Systems by J. Taylor and N. Raden Copyright 2009. All Rights Reserved.8Predictive Analytics World

Business Applicationsof Predictive Analytics Response modelingCustomer retention with churn modelingProduct recommendationsBehavior-based advertisingApplication processingInsurance pricingCredit scoringLeads scoringForecastingFraud detection Copyright 2009. All Rights Reserved.9Predictive Analytics World

Success StoriesuseR Meeting:DC R Users Group AflacAmwayThe Coca-Cola CompanyCitizens BankFinancial TimesHewlett-PackardInfinity InsuranceIRSLifeline ScreeningThe NRAThe New York Times Copyright 2009. All Rights Reserved. Optus (Australian telecom)PREMIER Bankcard7:30pm-10:00pmReed ElsevierMagnolia roomSprint-NextelSunrise Communications (Switzerland)TargetUS BankU.S. Department of DefenseWalden UniversityZurichPlus: Busch, Disney, HP, HSBC, Pfizer10Predictive Analytics World

Objectives Strengthen the business impactdelivered by predictive analytics Establish new opportunities withpredictive analytics30% Managers and Directors - Analytics & BI25% Managers and Directors - Other16% CXOs, Presidents, VPs, Principals14% Analysts12% Practitioners, statisticians, scientists Copyright 2009. All Rights Reserved.13Predictive Analytics World

Lowering Costs with Predictive Analytics1.2.3.4.5.Response modelingResponse uplift modelingChurn modelingChurn uplift modelingRisk modelingCOST CUTTERS:Lower operational spend Copyright 2009. All Rights Reserved.15Brand nameanecdotesPredictive Analytics World

LiftLifeline Screening:Response up 38%, cost down 20%,62k more customers annuallyCOST CUTTER #1:Don't contact thosewho won’t respond. Copyright 2009. All Rights Reserved.PREMIER Bankcard:Direct mail response up 3-5%17Predictive Analytics World

1A. Content SelectionSelect the message, product, creative thecustomer's most likely to respond to –“Dynamic AB Selection”Target:15-20% lift over traditionaldirect mail targetingABUS Bank:Costs down 40%, lift up 2 times, andcross-sell ROI up 5 times?COST CUTTER #1A:Don't send “A” if thecustomer's more likely torespond to “B”.C Copyright 2009. All Rights Reserved.18Predictive Analytics World

1B. Improve Paid Search AdsPredict ad bounce rate,based on: Ad text Landing pageCOST CUTTER #1B:Don't pay for clicksthat will bounce.Google:Improved bounce rateprediction 5-10%For more:http://www.bayardo.org/ps/kdd2009.pdf Copyright 2009. All Rights Reserved.19Predictive Analytics World

Uplift Modeling Some customers will buy anyway.A.K.A.: Net lift modeling Incremental modeling Impact modeling Differential responsemodelingAnalytical methods:Model on the residualDual recursive partitioning Copyright 2009. All Rights Reserved.21Predictive Analytics World

Buy if doreceive an offer2. Response Uplift ModelingNoDo-Not-DisturbsLost CausesPersuadablesYesSure ThingsYesNoBuy if don’treceive an offerCOST CUTTER #2:Don't contact thosewho'd respond anyway. Copyright 2009. All Rights Reserved.Leading financial institution:Incremental conversion up 0.02% to 0.43%Revenue per contact up by over 20 timesTarget:Challenges in upliftPredictivemodeling22Analytics World

3. Retention with Churn Modeling“A customer saved is a customer earned”Acquisition:New customersCOST CUTTER #3:Don't waste expensiveretention offers on thosewho’ll stay anyway.PREMIER Bankcard: 8 million est. retained Copyright 2009. All Rights Reserved.CustomerbaseDefection:Lost customersReed Elsevier's Caterer & HotelkeeperReduced churn by 16%Retention ROI up by 10%Optus (Australian telecom):Doubled churn model performancewith social dataOrange Labs (France Telecom):24Predictive Analytics WorldRapid scoring24

Fast Scoring on Large Telecom Dataset What's lift got to do with it?Bad, Bad Lee-ROI Brown99 Luft ModelsI am the Very Model of a Modern Major GeneralElectric Data Mining. Can You Dig It? Copyright 2009. All Rights Reserved.Predictive Analytics World

Leave if doreceive an offer4. Churn Uplift ModelingYesSleeping DogsSleeping DogsLost CausesPersuadablesNoSure ThingsNoYesLeave if don’treceive an offer"Leave well enough alone.""If it ain't broke, don't fix it.""Do not disturb!""Let sleeping dogs lie." Copyright 2009. All Rights Reserved.From FORRESTER whitepaper:Optimizing Customer Retention Programs,by Suresh Vittal26Predictive Analytics World

COST CUTTER #4:Telenor:Reduced churn 36%Cost-of-contact down 40%Campaign ROI up 11-foldDon't trigger thosewho'd otherwise stay.US Bank:Costs down 40%, lift up 2 times, andcross-sell ROI up 5 times Copyright 2009. All Rights Reserved.27Predictive Analytics World

5. Risk ManagementInsurance:Credit:Don't charge too little forhigh-risk applicants.Don't give credit to alikely defaulter.Zurich Insurance:Loss ratio down 1/2 pointSavings of almost 50 millCOST CUTTER #5:Don't acquire “losscustomers”. Copyright 2009. All Rights Reserved.30Predictive Analytics World

6. Leads ScoringHigh TechConsumerSun MicrosystemsDoubled number of leadsper phone callBella PicturesScores the bridesWalden UniversityScores the studentsCOST CUTTER #6:Don't expend salesresources on leads thatwon’t convert. Copyright 2009. All Rights Reserved.Every bride needsa mail solicitor.31Predictive Analytics World

7. Fraud DetectionTransactions include:Defense Finance andAccounting Service:InvoicesDetected 97% of knownfraud casesCredit card purchasesOnline activitiesCitizens Bank:Tax returnsLoss prevention up 20%or Insurance claims Prevention -staffdown 30%Telecom (calls)ChecksClicks (paid ads)COST CUTTER #7A:Don't let thieves getaway with it. Copyright 2009. All Rights Reserved.32COST CUTTER #7B:Expend fewer staffresources on fraudPredictive Analytics World

More Cost-CuttingInfinity Insurance:1100% increase infast tracking claimsApplication processing: benefits, jobs, insurance, .8. Don't spend as much time on likely denials.Forecasting9. Don't acquire/produce too many.Sunrise Communications:Reduced printing costs 20%Human resources10. Don't hire unreliable employees. Copyright 2009. All Rights Reserved.33Predictive Analytics World

Earn More Money"A penny earned is a penny earned"a)b)c)d)e)f)g)Increase responseAcquire better customersRetain moreCross-sellMore clicksCharge moreCollectionsWORKSHOP:Predictive Modeling Methods& Common Data Mining MistakesJohn Elder, Ph.D.h) Improve lift on current deployments Copyright 2009. All Rights Reserved.34Predictive Analytics World

Ensemble ModelsBagging, boosting, bootstrapping,random forests, meta-learning, blendingDoesn’t this violateK.I.S.S.?Somehow, they’re more robust! Copyright 2009. All Rights Reserved.35Predictive Analytics World

Dean Abbott:How to Improve Customer Acquisition Modelswith EnsemblesProphetNon-profitCase study: Non-prophet organizationNational Rifle Association Copyright 2009. All Rights Reserved.36Predictive Analytics World

Netflix Prize: Product RecommendationsBellKor's Pragmatic ChaosGrand PrizeMeta-learningBellKor in BigChaosProgress PrizePragmaticTheoryMeta-learningBellKor Copyright 2009. All Rights Reserved.BigChaos37Predictive Analytics World

Lowering Costswith Predictive AnalyticsResponse modelingTarget campaignsA) Content selectionSend the right messageB) Ad bounce predictionPay for effective clicksResponse uplift modeling Suppress “sure things”Churn modelingTarget retentionChurn uplift modelingLet sleeping dogs lieRisk modelingAvert riskLead scoringTarget sales resourcesFraud detectionAvert theft41 Copyright 2009. All Rights Reserved.Predictive Analytics World

Business Applicationsof Predictive AnalyticsBusiness application:What is predicted:Driven business decision:Direct marketingcustomer responseExpend cost of contact?Customer retentioncustomer churn/attritionExpend retention offer?Product recommendationswhat each customer wantsWhich product to recommend?Content/ad selectionwhich one they’ll respond toWhich content to display?Lifetime Valuehow profitable will beHow much to invest?Fraud detectionwhether transaction is legitWhether to investigate?Credit scoringdebtor riskWhether to grant loan/creditFundraising for nonprofitsdonation amountExpend cost of solicitation?Application processingapplication approvalWhether to fast-track/deny?Leads scoringconversionExpend sales resources/time?Insurance pricing & selectioninsured riskWhether to cover/how to price?Forecastingaggregate profit(Strategic decisions) Copyright 2009. All Rights Reserved.Predictive Analytics World

Business Applicationsof Predictive AnalyticsBusiness application:What is predicted:Case studies PAW Oct-09Direct marketingcustomer responseLifeline Screening, NRA,PREMIER BankcardCustomer retentioncustomer churn/attritionAflac, Optus (telecom), Orange,PREMIER, Reed ElsevierProduct recommendationswhat each customer wantsUS BankContent/ad selectionwhich one they’ll respond toTargetLifetime Valuehow profitable will beAmwayFraud detectionwhether transaction is legitCitizens Bank, DoD, HP, IRSCredit scoringdebtor riskFundraising for nonprofitsdonation amountApplication processingapplication approvalInfinity InsuranceLeads scoringconversionWalden UniversityInsurance pricing & selectioninsured riskZurich InsuranceForecastingaggregate profitCoke, Financial Times, NewYork Times, Sprint-Nextel Copyright 2009. All Rights Reserved.Predictive Analytics World

Predictive Analytics World ConferenceFebruary 16-17, 2010: San FranciscoBigger wins!Strengthen the business impactdelivered by predictive analytics.www.PredictionImpact.comwww.pawcon.com CopyrightCopyright 2009.2006. All2008.All RightsRights Reserved.Reserved "Predictive Analytics World was probably thebest analytics conference I have attended.turned into my new must-go-to conference."Dennis MortensenDirector of Data InsightsYahoo!February 2009 drew 200 attendees from 13countries, and included case studies from:Acxiom, Amazon.com, Bella Pictures,Charles Schwab, ClickForensics, Google,youAssociation,predict it, Pinnacolyou own it.The NationalIfRifleAssurance, Reed Elsevier, SunMicrosystems, TaxBrain,Telenor,Wells Inc.PredictiveAnalyticsWorld ImpactPredictionPredictionImpact,Fargo, Yahoo! and several more.46

Training program:Predictive Analytics for Business, Marketing and WebNovember 11-12, 2009San Francisco, CAbusinessprediction.com98% of attendees since Oct. 2008 rated this course Excellent or Very Good Copyright 2009. All Rights Reserved.47Predictive Analytics World

Predictive Analytics and Data MiningServices: Defining analytical goals & sourcing data Developing predictive models Designing and architecting solutions for model deployment "Quick hit" proof-of-concept pilot projectsTraining programs: Public seminars: Two days, in San Francisco, Washington DC, and other locations On-site training options: Flexible, specialized Instructor: Eric Siegel, Ph.D., President, 16 years of data mining, experiencedconsultant, award-winning Columbia professor Prior attendees: Boeing, Corporate Express, Compass Bank, Hewlett-Packard, LibertyMutual, Merck, MITRE, Monster.com, NASA, Qwest, SAS, U.S. Census Bureau, Yahoo!www.PredictionImpact.com CopyrightCopyright 2009.2006. All2009.All RightsRights Reserved.Reserved If you predict it, you own it.48PredictiveAnalyticsWorld ImpactPredictionPredictionImpact,Inc.

Predictive Analytics and Data MiningApplications: Response modeling for direct marketing Product recommendations Dynamic content, email and ad selection Customer retention Strategic segmentation Security–––– Verticals: Online business: Social networks,entertainment, retail, dating, job hunting Telecommunications Financial organizations A fortune 100 technology company Non-profits High-tech startups Direct marketing, catalogue retailFraud discoveryIntrusion detectionRisk mitigationMalicious user behavior identificationCutting-edge research forgroundbreaking data mining initiativeswww.PredictionImpact.com CopyrightCopyright 2009.2006. All2009.All RightsRights Reserved.Reserved If you predict it, you own it.49PredictiveAnalyticsWorld ImpactPredictionPredictionImpact,Inc.

Predictive Analytics and Data MiningTeam of several senior consultants: Experts in predictive modeling forbusiness and marketing Relevant graduate-level degrees Communication in business terms Complementary analyticalspecialties and client verticals Published in research journals andindustrialsExtended network of many more: Closely collaborating partner firms East coast coverage CopyrightCopyright 2009.2006. All2009.All RightsRights Reserved.Reserved Eric Siegel, Ph.D., PresidentPrediction Impact, Inc.San Francisco, Californiaeric@predictionimpact.com(415) 683 - 1146For our bi-annual newsletter, click “subscribe”:www.PredictionImpact.comTo receive notifications of training seminars:training@predictionimpact.comIf you predict it, you own it.50PredictiveAnalyticsWorld ImpactPredictionPredictionImpact,Inc.

Eric Siegel, Ph.D. Program Chair Predictive Analytics World Program Chair: Predictive Analytics World President: Prediction Impact, Inc. Instructor: predictive analytics training Former computer science professor: Columbia University Cofounded software companies: User profiling and data mining pawcon.com February 16-17, 2010 San Francisco, CA .