barry senensky fsa fcia maaa overview of claim scoring november 6, 2008
TRANSCRIPT
Barry Senensky FSA FCIA MAAA
www.claimanalytics.com
Overview of Claim ScoringOverview of Claim ScoringNovember 6, 2008November 6, 2008
On the AgendaOn the Agenda
About us
What is Claim Scoring?
Predictive Modeling
Building a Claim Scoring Model
Using Claim Scoring
Summary
Questions & Answers
• Founded in 2001 by two actuaries to apply predictive modeling techniques to insurance questions
• Clients in Canada and U.S.
• Several products
About UsAbout Us
What is claims scoring?
LTD/STD claims scored from 1 to 10, based on likelihood of recovery within a given timeframe
Scores are objective and accurate
Scores calibrated to probability of recovery
WhatWhat is claims scoring?is claims scoring?
J. Spratt Score: 4/6# 452135
P. Can Score: 3/9# 451156
J. Loe Score: 5/7# 452009
Predictive Modeling
Computer PerformanceComputer Performance
Measure IBM 7094
c. 1967
Laptop
c. 2004
Change
Processor Speed (MIPS)
.25 2,000 8,000-fold increase
Main Memory
144 KB 256,000 KB 1,778-fold increase
Approx. Cost ($2003)
$11,000,000 $2,000 5,500-fold decrease
What is a Predictive ModelWhat is a Predictive Model• A Predictive Model is a model which is
created or chosen to try to best predict the probability of an outcome
• Have been around for 40+ years
• Harnesses power of modern computers to find hidden patterns in data
• Used extensively in industry
• Many possible uses in insurance:
About Predictive ModelsAbout Predictive Models
May be parametric…
• apply numerical methods to optimize parameters
• E.g., gradient descent, competitive learning
Or non-parametric
• often have a decision tree form
• typically optimized using exhaustive search
Predictive Modeling ToolsPredictive Modeling Tools
Some common techniques
• Generalized linear models
• Neural networks
• Genetic algorithms
• Random forests
• Stochastic gradient boosted trees
• Support vector machines
Why aren’t Insurance Why aren’t Insurance Companies building more Companies building more predictive models?predictive models?• Life & Health Insurance Industry is conservative
and can be slow to change• Not a traditional actuarial tool• The times are changing!
– Especially P&C Insurers• Its only a matter of time!
– It just makes too much sense! – Innumerable applications to help solve
insurance problems
Building a Claim Scoring Model
Start with a data extract:
- Age - EP- Gender - Diagnosis- 2nd diagnosis - Income- Benefit - Occupation- Region - Own occ period- Industry - and more
Building the ModelBuilding the Model
Building the ModelBuilding the Model
1. Model presented with your historic claim data, including known outcomes.
2. Model begins making predictions on cases in the sample…
3. …compares predictions to real outcomes, and begins to detect patterns…
Initial predictions are rough…
But… model continues to learn
After millions of iterations and millions of comparisons… the model learns to predict accurately
And builds a complex algorithm that fits your data
Model ValidationModel Validation
• Critical test of model’s accuracy
• Outcomes of 10% of historical data withheld by client
• Once model declared complete, this data is used to test model — compare model predictions to actual outcomes
Model Validation ResultsModel Validation Results
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Reco
very
%
Pred Rec % 5% 15% 25% 35% 45% 55% 65% 75% 85% 95%
Recovery % 7% 12% 19% 32% 43% 57% 69% 76% 81% 92%
1 2 3 4 5 6 7 8 9 10
Model’s Predicted Recovery
Actual Recovery Rate
1. Scores can be calculated for all in-force claims
2. New claims can be scored weekly or even sooner
Claim Scoring ProcessClaim Scoring Process
Claim # Name EP Diagnosis Sex Age Benefit 6m Score
24m Score
12798 P.Can 119 Torn Medial Meniscus
M 52 1,250
12804 J.Loe 180 Fibromyalgia
F 46 2,500
12846 J.Spratt 364 Fibromyalgia
F 46 2,900
ReportingReporting
Note: actual reporting includes more fields than shown here.
Claim # Name EP Diagnosis Sex Age Benefit 6m Score
24m Score
12798 P.Can 119 Torn Medial Meniscus
M 52 1,250 3 9
12804 J.Loe 180 Fibromyalgia
F 46 2,500 5 7
12846 J.Spratt 364 Fibromyalgia
F 46 2,900 4 6
Using claim scoring
Are you kidding me?Are you kidding me?
&
Objective Triage
Facilitate early interventionFacilitate early intervention
Review of Old In-force Review of Old In-force ClaimsClaims
• High scores — opportunities for recovery
• Low scores — opportunities for expense savings
Discover new opportunities
Workload AllocationWorkload Allocation• Claims can be allocated by degree of challenge
• 4s to 7s more difficult, time-intensive — more experienced and expert claims handlers
• 1s to 3s, 8s to 10s simpler — newer / less experienced claims handlers
• Equalize workload of claims personnel
Smooth the workload
Prioritize TimePrioritize Time• Can be used by claims handlers to prioritize their
time
Snapshot of your workload
Social Security Social Security // Other Offsets Other Offsets
• 1s to 3s are good candidates to review
• Even better to build a model specific to determining which claims to send to Social security and when…
Learn when to reach out
Decision Support ToolDecision Support Tool
• Rehab
• IME’s
• Surveillance
• Other forms of intervention
• Settlements
Optimize resource $
MeasurMeasuree performance performance
• Scores represent expected recovery rates
• Can be used to measure actual to expected (A/E) recoveries
What you can measure, you can improve
Actual Recovery
%
Predicted Recovery
%
A / E
Regional Office 1
30 27 111
Regional Office 2
50 61 82
Planning/ForecastingPlanning/Forecasting
• Scores indicative of future recovery experience
• Use to develop financial projections for group business unit.
Don’t get blindsided!
Reporting: trend Reporting: trend identificationidentification
Average 24 Month Score
3.00
4.00
5.00
6.00
7.00
Score 5.78 5.68 5.63 5.55 5.47 5.37 5.34 5.23 5.15 5.07 5.02 4.94
May 03 Jun 03 Jul 03 Aug 03 Sep 03 Oct 03 Nov 03 Dec 03 Jan 04 Feb 04 Mar 04 Apr 04
BenchmarkingBenchmarking
• Accurately, objectively compare claim practices with other companies
How are we really doing?
Sample Predicted Recovery RateSample Predicted Recovery Rate
Predicted Recovery Rate:Age 45, Male, Displaced Disk
0%
20%
40%
60%
80%
100%
Predicted 81% 65% 74% 73%
Company A Company B Your Company Average
SummarySummary
Fast. Accurate. Objective.
Optimize resources.
Facilitate early action.
Improve results.
Opportunities in approach. Opportunities in approach.
AfterBefore
Questions?Questions?