Fair Isaac Corporation · NYSE:FICO
Fair Isaac — Conference Presentation (December 2010)
44 slides from Fair Isaac's conference presentation, Dec 1, 2010
Slide 1: An Analytics Study: Achieving Greater Underwriting and Pricing Accuracy
An Analytics Study: Achieving Greater Underwriting and Pricing Accuracy Lamont D. Boyd, CPCU Scott Horwitz Gregory Marholin Insurance Market Director Senior Director, Senior Account Executive, Global Scoring Solutions Insurance Industry Practice Insurance Industry Practice FICO FICO FICO December 1,
Slide 2: Agenda
Agenda » Overview » Insurance Scores: CBIS and Property Risk scores » Custom Models » Premium leakage and pre-claims fraudulent detection at underwriting » Q&A 2 © 2010 Fair Isaac Corporation. © 2010 Fair Isaac Corporation. Confidential.
Slide 3: Overview: Achieving Greater Underwriting and Pricing Accuracy
Overview: Achieving Greater Underwriting and Pricing Accuracy » Credit-Based Insurance Score (CBIS) models were developed and introduced by FICO in 1993 » CBIS scores currently used by ~90-95% of all auto and home insurers » CBIS scores are monitored for effectiveness and amended as required by regu
Slide 4: Agenda: Insurance Scores: CBIS and Property Risk
Agenda » Insurance scores: CBIS and property risk scores » Custom models: GAM versus GLM Insurance Scores: CBIS and » Premium leakage andpre-claims Property Risk fraudulent detection at underwriting detection activity at underwriting 4 © 2010 Fair Isaac©Corporation. 2010 Fair Isaac Corporation. Conf
Slide 5: CBIS Expanded Usage
CBIS Expanded Usage » Expanded use provides consistent link in all areas that require strategic decisioning Decision Engine Command Center Credit-Based Insurance Score Customer Acquisition CRM Underwriter Management New Business Underwriting Cross-Selling Field Office Management Tier Placement Billi
Slide 6: Lifecycle: Acquisition – Client Benefits
Lifecycle: Acquisition – Client Benefits » New Customer Acquisition Strategy » Focused application » Cost-effective acquisition strategies » Targeting most likely responders » Response modeling » Seeking greatest profit potential » Limited utilization currently » Few regulatory restrictions 6 © 2010
Slide 7: Lifecycle: Underwriting and Pricing – Client Benefits
Lifecycle: Underwriting and Pricing – Client Benefits » New Business Underwriting Strategy » Earliest CBIS application » Consistent and objective new business decisions » Quickly accepting applicants with greatest profit potential » Dedicating resources to appropriate risks » Minimizing exposure to
Slide 8: New Business Underwriting Strategy: Loss Ratio
New Business Underwriting Strategy Decline Referral Automated Approval 140 130 120 110 100 Loss Ratio 90 80 70 Average 60 50 40 0 1 2 3 4 5 6 7 8 9 10 Population by deciles 8 © 2010 Fair Isaac Corporation.
Slide 9: Lifecycle: Renewal Underwriting – Client Benefits
Lifecycle: Renewal Underwriting – Client Benefits » Renewal Underwriting Strategy » Secondary CBIS application » Updating CBIS scores at renewal where allowed » Identifying renewal policyholders for swift processing » Highlighting renewal policyholders requiring greater attention » Renewal tier-plac
Slide 10: Renewal Business Underwriting Strategy: Loss Ratio
Renewal Business Underwriting Strategy Refer to N/R Underwriting Automated Approval 140 130 120 110 100 Loss Ratio 90 80 70 Average 60 50 40 0 1 2 3 4 5 6 7 8 9 10 Population by deciles 10 © 2010 Fair Isaac Corporation.
Slide 11: CBIS Expanded Uses – Client Benefits
CBIS Expanded Uses – Client Benefits » CBIS enhances decision-making for the enterprise » Book management » Managing books of business at varying level for profitability » Underwriting management » Managing underwriter responsibilities to match risk complexity and varying levels of production and pr
Slide 12: An Update: FICO CBIS Scores in Current Economic Environment
An Update: FICO CBIS Scores in Current Economic Environment Have FICO CBIS score distributions changed over time? » Ongoing analysis » FICO CBIS score distributions remain essentially the same today as during the past 5 years » Some consumers are seeing lowering scores » Economic climate-driven fact
Slide 13: Regulatory Environment Forecast
Regulatory Environment Forecast » Regulatory/legislative questions will continue » Individual state actions » NAIC committees/task forces » Risk classification studies » Proposed vendor regulation model law » FTC Study—Auto (2007) and Home (TBD) » Education is the key » Education to help consumers u
Slide 14: Property Risk Score
Property Risk Score » Property PredictR® Insurance Scores were launched in 2007 to help insurers harness predictive value from property inspections performed by FICO’s partner – Millennium Information Services » Snapshot of risk based on property inspection data » Rank-orders properties by loss rati
Slide 15: Property PredictR Score Retro Analysis: Loss Ratio Relativity
Property PredictR® Score Retro Analysis Loss Ratio Relativity vs Score Results 1.6 1.53 1.4 1.26 Loss Ratio Relativity 1.2 1.10 1.0 0.92 0.86 0.76 0.8 0.64 0.6 0.4 0.2 0.0 Below 180 180-201 202-222 223-244 245-272 273-293 294 or more 11.7% 9.1% 19.3% 17.6% 20.7% 11.1% 10.5% Score by Percent Premium
Slide 16: CBIS with Property PredictR Loss Ratio Relativity Matrix
CBIS with Property PredictR® Loss Ratio Relativity Matrix Property Property PredictR®Score PredictR™ Score SCORE RANGE Medium Medium Lowest Low Medium High Highest SCORE RANGE 20% 20% 20% 20% 20% Total Credit-Based Insurance Score Lowest Scoring 20% Consistently Poor Risks 1.48 Medium Low 20% 1.83 0
Slide 17: Agenda: Custom Models
Agenda » Insurance scores: CBIS and property risk scores » Custom models: GAM versus GLM Custom Models » Premium leakage and pre-claims fraudulent detection at underwriting detection activity at underwriting 17 © 2010 Fair Isaac©Corporation. 2010 Fair Isaac Corporation. Confidential. © 2010 Fair Isa
Slide 18: Analytics Spectrum
Analytics Spectrum Different decisions lead to different methodologies Predictive Multi-Dimensional Profiling & Decision Models Trade-Off Segmentation Optimization or ―Scores‖ Assessment HIGH X X X X X X X X X X X X X X X X X X X X XX X X X X X X XX XXX X X X X X XXXX X X XX X XX XX X XX X X X X XXX
Slide 19: Modeling
Modeling A model that will improve decisions should be … » Flexible » Can capture complex data relationships » Palatable » Interpretable » Satisfies operational constraints » Legal concerns » Implementation friendly » Able to be computed quickly and seamlessly deployed » Robust into the future » Gra
Slide 20: Project design: Understanding business challenges
Project design Understanding business challenges » Data Mining - ―Try multiple algorithms and see which one best matches the business challenge‖ Data Mining Prepare Build Build Forecast Deploy Access Visualise Modelling Standard Scorecard Predictive Analytics Data & Explore Datasets Models Model
Slide 21: Model Development: Initial data analysis
Model Development Initial data analysis Fine Binning » What are we trying to predict? » Metrics » How do we know the model is working? » What data sources are available? » What are costs vs. benefits? Coarse Binning » Common data issues » Incomplete and/or inaccurate data » Outlier anomalies can ske
Slide 22: Model Development: Segmentation
Model Development Segmentation .080 .068 Lowest Claim Frequency .066 Total Applicant Population .064 .062 Segment 1 Segment 2 .060 .058 .056 .054 .052 .050 Scorecard 1 Scorecard 2 Scorecard 3 Scorecard 4 0 1 2 3 4 5 6 ... Segment 1.1 Segment 1.2 Segment 2.1 Segment 2.2 # Scorecards … how many scorec
Slide 23: Model Development: Score Engineering
Model Development Score Engineering » Imposing constraints on the score formula to achieve various objectives » Objectives » Palatability » Legal requirements » Robustness over changing times » Adjustments for known sample biases » Engineering techniques » Choose the ―right‖ objective function – pos
Slide 24: Forecast model performance in production
Forecast model performance in production 24 © 2010 Fair Isaac Corporation.
Slide 25: Model Development: Rank Ordering of Performance vs. Score
Model Development Rank Ordering of Performance vs. Score Metric by Binned Score 58.1% New 54.1% Renewal 44.3% 43.3% Metric 34.0% 33.6% 27.7% 25.9% 25.5% 24.2% w orst 20% 20% - 40% 40% - 60% 60% - 80% Best 20% Score Range Binned by % Premium $508 Metric by Binned Score $475 New $382 Renewal $366 Metr
Slide 26: Implementation
Implementation Can I make it work in practice? Modeling Process Variables Scoring X Data Z Preparation Y Stats Modeling Data Modeling Stage Compare Repository Stats x’ Data Runtime y’z’ Mapping Deploy Package Variables Scoring (Java) Production Data IT Model Deployment Process 26 © 2010 Fair Isaac C
Slide 27: Agenda: Premium leakage and fraud detection at underwriting
Agenda » Insurance scores: CBIS and property risk scores » Custom models: Premium leakage GAM versus and fraud GLM detection at underwriting » Premium leakage and pre-claims fraudulent detection at underwriting detection activity at underwriting 27 © 2010 Fair Isaac©Corporation. 2010 Fair Isaac Corp
Slide 28: Why We're Here
Why We’re Here ―Auto insurance fraud (including premium leakage) accounts for $0.20–$0.30 of every premium dollar‖ A Top 10 Personal Lines Auto Insurer Suspicious staged accident claims soar in Florida DES PLAINES, Ill., June 28 /PRNewswire-USNewswire Insuring a vehicle you do not own Is there any r
Slide 29: Defining Fraud—It's an Amoeba
Defining Fraud—It’s an Amoeba » Fraudsters constantly adapt to stay below the radar… 0–60 Days 90–120 Days then 9-10 Months 29 © 2010 Fair Isaac Corporation.
Slide 30: The Severity of the Problem: Soft Fraud—Premium Leakage
The Severity of the Problem … Soft Fraud—Premium Leakage Type of Premium Leakage Range of Lost Annual Premium Undisclosed Youthful Driver Impact = $800–$1,200 Undisclosed Non-Youthful Driver Impact = $500–$800 Incorrect Territory Impact = $500–$1,200 Garaging Location Impact = $500–$1,200 Commercial
Slide 31: The Severity of the Problem: Hard Fraud
The Severity of the Problem … Hard Fraud—Premeditated Claims Fraud » 5% to 8% of all claims are fraudulent with average costs in excess of $7,000 » Many factors independently or in combination link to hard fraud including: » Invalid SSN (if used) » Non-Residential Address » Non-Existent Address » VI
Slide 32: The Severity of the Problem: Number of Inconsistencies
The Severity of the Problem … The Number of Inconsistencies » 10%–14% Undisclosed Non-Youthful » 9%–12% Undisclosed Youthful » 7%–10% Garaging Location » 7%–10% Territory Mismatch » 7%–9% DLN Format » 2%–3% VIN Error » 1%–2% SSN » >2%–Various Other Data Points Source: Feedback from various carriers
Slide 33: Leveraging the Power of Predictive Analytics
Leveraging the Power of Predictive Analytics » Measures consumer forthrightness and other fraudulent behavior specific to your book of business » Ability to verify and score: » Consumer Identity Input » Presence and Relation of Undisclosed Drivers » Vehicles (title, use and operators) » Payment Type
Slide 34: Categories of Data to Consider
Categories of Data to Consider » Examines key categories of risk independently and collectively to best separate good risks from bad » Consumer Identity » Vehicle Ownership » Garaging » Unreported Vehicle Damage » Address Legitimacy » Financial Momentum » Presence and Relation » Suspicious and Fraud
Slide 35: Modeling 101: Data Exploration
Modeling 101: Data Exploration » Each book of business is unique » The data defines the solutions » Exploratory Tools » Regression » Clustering » Link Analysis » Client Provides: » Historical Policy and Performance » Recaptured Premium leakage (known) » Known Fraud » Other Pertinent Information 35 ©
Slide 36: Our Modeling Process—Framework
Our Modeling Process—Framework » Custom Models » Pre-Claims Fraud Model » Premium Leakage or Report Ordering Model » Sophisticated Decision Tree » Business Rules Management (BRM) Tool 36 © 2010 Fair Isaac Corporation.
Slide 37: Fraud and Premium Leakage Scoring Model Results
Fraud and Premium Leakage Scoring Model Results » Proof of Concept (POC) analysis demonstrating in excess of a 10:1 Return-on-Investment » Generalized Additive Model (GAM) –Potential Solver » Expanded version of GLM » Transparent to business users (Reason Codes) » Score engineering providing flexibi
Slide 38: Sample Output: Scored Applicants with Reason Codes
Sample Output Scored Applicants with Reason Codes Name Policy # Score Relationship John Smith 554333221 1650 Named Insured Verify the Consumer's Identity Jane Smith Listed Driver—Spouse Undisclosed driver(s) under 25 Undisclosed driver(s) have same surname Issue Details Undisclosed driver(s) differe
Slide 39: Benefits of fraud and premium leakage detection at underwriting
Benefits of fraud and premium leakage detection at underwriting » Generates more premium based on client tolerances. » Provides fewer false positives through predictive analytics » Actionable results defined by client » Mitigates claims fraud » Configurable by business users » Ability to use data fr
Slide 40: Agenda: Summary
Agenda Summary 40 © 2010 Fair Isaac©Corporation. 2010 Fair Isaac Corporation. Confidential. © 2010 Fair Isaac Corporation. Confidential.
Slide 41: Overview: Achieving Greater Underwriting and Pricing Accuracy
Overview: Achieving Greater Underwriting and Pricing Accuracy » Credit-Based Insurance Scores (CBIS) continue to be extremely effective in predicting risk » Property PredictR® Insurance Scores are very effective risk segmentation tools » Custom Models should be targeted toward specific business oper
Slide 42: Agenda: Polling Questions
Agenda Polling Questions 42 © 2010 Fair Isaac©Corporation. 2010 Fair Isaac Corporation. Confidential. © 2010 Fair Isaac Corporation. Confidential.
Slide 43: Polling Questions
Polling Questions » I am interested in credit based insurance scores. Yes/No » I am interested in property risk scores. Yes/No » I am interested in further discussions on custom models. Yes/No » I am interested in detecting fraud and premium leakage at underwriting. Yes/No » I am interested in optim
Slide 44: Q&A
Q&A Gerald Fahner Senior Director, Research & Development FICO GeraldFahner@fico.com Lamont D. Boyd, CPCU Scott Horwitz Gregory Marholin Insurance Market Director, Senior Director, Senior Account Executive, Global Scoring Solutions Insurance Industry Practice Insurance Industry Practice FICO FICO FI