Upload
phungnguyet
View
216
Download
0
Embed Size (px)
Citation preview
Predictive Analytics in Life Insurance
Advances in Predictive Analytics Conference, University of Waterloo
December 1, 2017
Format of this session
Speakers:Jean-Yves Rioux - DeloitteKevin Pledge – Claim AnalyticsIan Bancroft – Sun LifeEugene Wen – Manulife
Discussion and Questions
PredictiveAnalyticsinlifeinsurance
Jean-YvesRiouxAdvancesinPredictiveAnalyticsConference,UniversityofWaterloo
December1,2017
Beyonddescriptivestatistics
Evolutionoftheanalyticsprocess
• Predictiveanalyticsistheuseofdatatogeneratepredictiveinsightsinordertomakesmarterdecisionsthatimproveperformanceofbusinessesanddrivestrategytooutlastthecompetition.Analyticsshouldgobeyonddescriptionofthepastandshouldprovideactionableinsightsaboutthefuture.Thisdiscussionwitheducateyouonthefullpotentialoftheapplicationoftechniques.
4
Hindsight
Insight
Foresight
Simulationandmodelling
Quantitativeanalyses
Advancedforecasting
Exceptionsandalerts
Sliceanddicequeriesanddrill-downs
Managementreporting
Optimizationalgorithms
Predictiveandprescriptive
Role-basedperformancemetrics
Descriptive
Enterprisedatamanagement
…andwhyresearching/understandinganalyticsiscomplex
Overallmethodology
5
Phase UnderstandingBusiness Problem
Data Collection DataCleansing AnalysisofData ModelSelection
VariableSelection
ParameterizationandOptimization
ResultsPresentationandImplementation
Description Identify thekeyquestionstoanswer,problemstosolve,andbusinessconstraints
Collectandunderstanddataandvariablerepresentations
Validate,standardize,andtransformdata
Analyzevariables,distributions,correlations,andclustering
Identifyappropriatemodellingoptionsbasedonbusinessproblemanddataavailable
Identifypreliminaryvariablesthatappeartobesignificantpredictors
Traindifferentmodelsandvariablecombinationstodeterminetheoptimalmodelandparameters
Presentresultsinamannerthatcanbeunderstoodbyseniormanagementofvarioustechnicalbackgrounds.Embedthepredictivemodelintobusinessprocesses
Predictiveanalyticsapplications
Otherapplications• ProvideinsightsusingHR/workforceAnalytics
SalesandMarketing• Buildmoreeffective,targetedmarketingcampaigns• Identifycustomerslikelytopurchaseproducts• Provideappropriateproductrecommendations
AgencyManagement• Recruitadvisorsmostlikelytobecomesuccessful• Optimizeagentretentionefforts• Monitorperformanceofagents
Underwriting• Identifybestrisksandstrategicallyprioritizeunderwritingefforts• Improvesimplifiedunderwritingandstreamlineunderwritingprocesses• Expandunderwritinginformationusingnewsources
InforceManagement• Profileandsegmentcustomers• Identifyandretainpolicyholderslikelytosurrender• Designretentionstrategiesandofferadditionalproductstocurrentcustomers• Analyzecustomersatisfaction
Pricing,Reserving,andExperienceStudies• Developaccurate,competitivepricingandpricingstructure• Identifynewexperiencedriversusingaugmenteddata• Improvereservingaccuracy• Improveunderstandingofexperiencedrivers
ClaimsManagementandFraudDetection• Analyzeclaimfrequencyandseverity• Prioritizeclaimsresources• Identifylikelyfraudulentorrescindedclaims
6
7
Applications1. Cross-selling/ Up-selling2. Risk-Based Segment Targeting3. Distribution Partner/ Client Matching4. New Business Application Triage/ Simplified UW5. Underwriting Process Improvement6. Underwriting Application Simplification7. Customer Retention/ Lifetime Value8. Pricing Risk Score9. Experience Studies Policyholder Behavior10. Experience Studies Mortality11. Fraud Detection12. Workforce Analytics
8
Applications1. Agency Management2. Claims Management3. Product Design4. Direct to Consumer Targeted Marketing5. Predictive Underwriting
9
• CIAPredictiveModelingCommittee’smissionistopromotetheapplicationoftheactuarialskillsettopredictivemodelingandtomarketactuariesasskilledexpertsinthisfield.Morespecificallythecommitteewill:
• a)Promotetheroleofactuariesinthepredictivemodelingfieldbothwithinandoutsidetheprofession;
• b)Identifyneedsfornewresearchinthefieldofpredictivemodelingthatwilldrawattentiontotheworkofactuaries;and
• c)Communicateexistingresearchandcasestudiesofworkinthefieldthroughvariousmedia
• d)InfluencetheEducationCurriculumandCPDofferingtoincluderelevantpredictivemodelingtechniques
• CIAPredictiveModelingCommittee’sinitiativesincluded:
9
Promotionwithin/outside Needsforresearch Communication/Dissemination Influencecurriculum Collaboration/RelationshipProfileSeriesine-bulletinofPredictiveModelers(storiesfromactuariesandnon-actuariesandtheirworkinPM,startwithcommitteemembersandthenexpand),target3withinfirstyear
Organizeawebcast GetonCPDsessions (Ongoing)Connectwithotherorganizationstocoordinate/collaborate
Bookletaboutwhyandapplications MonitorCPDopportunitiesandinformtheCIAforinclusionintheCPDopportunitiese-mail
ProvideinputintotheCIAcurriculuminsupportofthealignmentwiththecurriculumadoptedbyotherorganizations(CASandSOA)
OrganizeaCIAeventWeb-basedindexofstrongreferencedocumentsrelatingtoTechniquesandApplications[completed]
ExpandtheWeb-basedindextoincludelinkstotools[completed]
Theeffortsoftheprofession– CanadianInstituteofActuariesinitiatives
Predictive Analytics in Life Insurance
Kevin PledgeClaim Analytics
Advances in Predictive Analytics Conference University of WaterlooDecember 1, 2017
Predictive analytics success seen in P&C and other industries – life and health slow to adopt
Why was that?
Long term nature of the business
11
Claim Analytics c.2000 saw an opportunity in claim management for group disability, based on expected outcome:
12
Claim Scores
Highest potential for return to work (a good claim manager will have biggest impact)
PermanentResolve Quickly
Don’t spend expensive resources
Expense management
In Between
Return to Health
Highest potential for success with investment of proactive and solid CM strategies. Likely to have a longer duration and require more specialized resources and experienced CMs.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Prob
abilityofR
TW
ClaimScore
8
9
10
5
6
7
2
3
4
Low Touch
Require solid but generally straightforward CM strategies –often less complex claims with early RTW focused interventions.
1
Expense Mgmt
Alternate resolution strategies such as SSDI offsets and settlements should be considered as the likelihood of a solid outcome with investment of other resources is less likely.
Claim Scoring6 Months
• Claim Assignment• Early
opportunities
12 months
• Claim Assignment• Resource
allocation
18 months
• Resource allocation
• Transition
24 months• Settlements
Can this be applied to Short Term Disability?
Duration Management
Transition Management
Resource Allocation
0
20
40
60
80
100
0-7 8-14 15-21 22-28 29-35 36-42 43-49 50-56 57-63 64-70 71-77 78-84 85-91
Cum
ulat
ive
Term
inat
ions
Use predictive models to compare companies
Benchmarking
RehabClaim Management
Claim Performance
Related models?
15
1. Need to show value (initially quickly)2. Start simple and build from there3. Revisit & Refresh – always check in with the business
Summary & Final Thought
16
New businessUnderwriting
Don’t limit yourself
Fraud
DescriptionofSLFModelingTeamsTheQuestion….ShouldwecentralizeordecentralizethemodelingfunctionintotheBusinessUnits(BU’s)?ie developoneareaofexpertmodelers….butwithoutbusinessexpertise,orshouldwemovethemodelersclosertothebusiness?Similarquestiontohowtoorganizetheactuarialfunction
DescriptionofSLFModelingTeamsTheAnswer….We’veplacedmostofourmodelersintoteamsintheBU’s,andalsohaveasmallcentralizedteam….similartoouractuarialorganizationalstructureThemakeupofeachmodellingteamdependsontheneedsoftheirwork…..eg avaluationpredictivemodelingteamisverydifferentthanafraudteamWespendalotoftimeconnectingacrossthedifferentmodelingteamstopromotesynergiesandbestpractices
DiversityofSkillsRequiredforModelingWorkNoonepersoncanbeanexpertinallmodelingtechniques,dataengineering,businessknowledge,changemanagement ….thereisno“IvoryTower”formodelersRatherateamofpeopleisneededGoodcommunication,collaborationskillsareveryimportant
GrowingandBuildingModelingCommunitiesAsmodelingbecomesmore“mainstream”,creatingamodelingcommunityacrossyourorganizationbecomesincreasinglyimportantThereisnooneuniquesolution……thingswearedoinginclude◦ Annualanalytics1dayconferences◦Monthlydeepdiveswithrotatingspeakers◦ Biweeklystatusmeetings◦ Aninternalcompetition(similartoKaggle)◦ InternaltrainingcoursesWefrequentlyexperimentwithandadjustourprocessestoimproveoursolutions
PredictiveAnalyticsinLifeInsurance
EugeneWen,MD.DrPH.VP- GroupAdvancedAnalyticsManulife
AdvancesinPredictiveAnalyticsConferenceDecember1,2017UniversityofWaterloo
Beforetheterm“DataScientist”wascoined,ActuaryhadservedasTHE“datascientist”ofinsuranceindustryformorethantwohundredyears.
…since the first Actuary
• Society for Equitable Assurances on Lives and Survivorship in London in 1762
• The first actuary: Edward Rowe Mores• Role of actuary
• Business Analysts: underwriting, claim management, marketing, investment, finance, risk, strategy, IT, BI & reporting etc.
• Data Scientists
Advanced Analytics in Manulife/John Hancock (1/2)
26
CindyForbes,EVP,ChiefAnalyticsOfficer,
formerGlobalChiefActuary,
ChairofGovernors,UniversityofWaterloo
Advanced Analytics in Manulife/John Hancock (2/2)
27
§ Over 60 FTEs globally
§ Approximately 25 in Asia
§ Advanced Analytics teams consist of highly educated and specialized talent across the globe
§ 15 PhD’s among the team members
§ Programs include: Pattern Recognition and Machine Intelligence, Math & Statistics, Computational Physics, Machine Learning, Computational Neuroscience, Computer Science, Economics, Medicine
§ 20 Masters degrees among team members
§ Programs include: Systems & Engineering, Statistics, Management Analytics, Computer Science, Electrical Engineering, Finance and Information Systems
§ 7 MBAs
DATA SCIENCE
Modernized Underwriting
Text Analytics
Computer Vision
Buying a life insurance product canbe a lengthy and time-consumingprocess. How can predictiveanalytics improve the underwritingapproval process?
MODERNIZED UNDERWRITING
02
We have billions of customerservice transcripts. How can weuse this data to improve thecustomer experience?
TEXT ANALYTICS03
Our life insurance applicants mustprovide a lot of information to beunderwritten for a policy. Can weimprove this process by extractingfeatures from a photo?
COMPUTER VISION
04 Fraud Detection
01
Data Science Projects (1/2)
We have billions of transactionsannually. The challenge for the datascience team is to identify fraudulenttransactions in all of the data.
FRAUD DETECTION
Data Science Projects (2/2)
How can we identify those that are likely to be dishonest on their application form?
Predicting Smoker Status
Life experiences don’t occur often. How do we ensure our model is performing as expected going forward?
Monitoring Experiences
Can we develop a model that significantly reduces the
time to underwrite a policy for qualified applicants?
Predictive Underwriting
How can we improve the life insurance
purchasing process for our applicants?
Purchase Process
*MortalityStudy/Liability/Nextyearpayout*LTClapseafterpremiumincrease*Segfund lapse