Implementation of Predictive Models Making Models Come Alive

Implementation of Predictive Models – Making Models Come Alive John Lucker – Principal – Deloitte Consulting LLP Michele Yeagley, ACAS, MAAA – Asst. Vice President - Harleysville Insurance Casualty Actuarial Society - Predictive Modeling Special Interest Seminar September 2005

Discussion Themes • Predictive Modeling provides the toolset to assist with a variety of critical business operations • The market is softening and better risk assessment capabilities are needed to avoid inappropriate soft market dynamics and naïve reactions • Predictive Models can be implemented in ways that do not require complex, expensive efforts • The financial benefits that can be realized from predictive models are very significant – phased and rapid implementation can help fund additional future analytics and more complex implementations • The three most important parts of any predictive modeling project are (1) implementation; (2) implementation; and (3) implementation – building a great model should be a given Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 1

Some Predictive Modeling Basics

Some Areas that Insurance Predictive Modeling Can Address Customer Management Risk Profitability Assessment • Risk Attraction • Cross-Sell • Risk Retention • Client • Risk Avoidance • Geo View / Market Expansion • Non-Renewals • Book • Right-Pricing • Proactive • Audit Claims Management • Claim • Fraud • Retro Propensity Biz Scoring Placement/Retention • Pre-Deal Production Assessment • Post-Deal Retention • Recruitment Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. & Billing Options M&A Producer Management • Production Call Centers • Assumed Reduction • Profitable Rollover/Transfer Reinsurance Triage • Duration Success Rates 3 Remediation

Before Insurance Predictive Modeling – Class Underwriting Numbers are Illustrative Workers’ Compensation Commercial Auto CMP / BOP Property General Liability Private Passenger Auto Homeowners Overall Loss Ratio of 75% Roofers Youthful Drivers Florists Middle Aged Drivers 60% 70% 63% 80% 65% 110% 75% 90% 68% 100% 72% Below average Above average Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 4 115% 78% 125% 82% 135% 87% 140% 90%

A Predictive Modeling Approach Building and deploying predictive models requires a specialized combination of skills covering data management, data cleansing, data mart construction, actuarial and statistical analysis, data mining and modeling, and insurance operational and business processing and technologies External Data Score For Each Policy Data Aggregation & Data Cleansing 60% 70% 80% 160% 140% 120% 100%110% 90% 180% Predicted loss ratio Internal Data Evaluate and Create Variables Business Rules Engine Synthetic Variables Develop Predictive Model Score Driven Business Applications Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 5

With Insurance Predictive Modeling – Individual Risk Scoring Y = A + B(Var 1) + C(Var 2) + D(Var 3) + E(Var 4) + F(Var 5) … Internal / External Data Predicted Loss Ratio Bob’s Flower Shop = 821 120 Numbers are Illustrative Predicted Loss Ratio 90 Linda’s Flower Shop = 324 82 78 74 66 70 62 58 50 Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 6 Overall L. R. 75%

Laying the Groundwork for an End-To-End Implementation

What The Process IS NOT – What it IS NOT What it IS • A Black Box approach • Scoring drivers are known / understood • Stock delivery • Collaborative throughout the business • Replacement for underwriters • Additional underwriting toolset • Score used to communicate decision • Reason codes / messages are developed • Score drives results • Implementation drives results • A single variable magic bullet • Relationship among variables is power • Actuarial and/or systems project • Business initiative • Class plan underwriting • Efficient segmentation of policyholders Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 8

Using the Lift Curve for Business Applications – Renewal Business Highly Profitable • Risk Attraction Best 25% • Retention Priority • Less Loss Control Renewal Business Underwriter Workflow • Less Premium Audit • More Pricing Flexibility Highly Unprofitable • Risk Avoidance Worst 25% • Repricing Priority • Non-Renewals • Loss Control • Premium Audit Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 9

Using the Lift Curve for Business Applications – New Business Highly Profitable • Risk Attraction Best 25% • Retention Priority • Less Loss Control New Business Underwriter Workflow • Less Premium Audit • More Pricing Flexibility Highly Unprofitable • Risk Avoidance Worst 25% • Repricing Priority • Non-Renewals • Loss Control • Premium Audit Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 10

End-To-End Implementation – Making Models Come Alive Biz & Technical Planning Predictive Modeling Technical Developmt Business Process Redesign Biz & Systems Integration Organize Change Mgmt Perform Metrics & Reporting • Predictive Models must be effectively implemented to derive their benefit potential • The financial benefits can be so significant that urgency should drive the pace of the project • Create a benefit analysis and use the benefits to drive the project – a complex process (PIF counts, LR management, retention, not written, etc) • Competitive jockeying should also drive project pace – first adopter advantages • A best practice is to create a continuum of implementation solutions and phases • Initial implementation should focus on extracting value from models before automation • Tactical implementation can be achieved in 2 -4 months • Planning, planning, and then some more planning Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 11

End-To-End Implementation – Making Models Come Alive Biz & Technical Planning Predictive Modeling Technical Developmt Business Process Redesign Biz & Systems Integration Organize Change Mgmt Perform Metrics & Reporting • Steering Committee and Project Committee Structure • Phased structure and focus on 80: 20 Rule • Development of End-State-Vision & Project Planning Document – some key questions are: • How will predictive modeling guide decision making, pricing, and tier placement? • How will predictive models impact existing business processes (e. g. by line / account)? • How will predictive models be blended into the field and agency management process? • What key performance measures must be achieved? • How will underwriters/raters/other personnel’s compliance be measured? • What level of automation is desired for various business processes? Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 12

End-To-End Implementation – Making Models Come Alive Biz & Technical Planning Predictive Modeling Technical Developmt Business Process Redesign Biz & Systems Integration Organize Change Mgmt • Extract, Transform, Load process for Internal, External, and Synthetic Data • Management of external data vendors and external data acquisition • Data quality and cleansing issues • Construction of Scoring Engine(s) (real time, batch, manual, etc) • Construction of Business Rules Engine or similar process • Construction of operational data mart • Design of technical architecture, data flow, messaging, data management, etc. • Model maintenance Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 13 Perform Metrics & Reporting

End-To-End Implementation – Making Models Come Alive Biz & Technical Planning Predictive Modeling Technical Developmt Business Process Redesign Biz & Systems Integration Organize Change Mgmt Perform Metrics & Reporting • Underwriting workflow (renewal business, new business, touch level) • How model scores will be used in the U/W process (scores, reason codes, action thresholds, risk avoidance, risk acquisition, retention management, account vs. monoline, etc) • How will models be used as risks proceed from new to renewals (disruption issues)? • Use of model scores for downstream processes • Relationship of model usage to field and producer management • Business rule creation, optimization and maintenance Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 14

End-To-End Implementation – Making Models Come Alive Biz & Technical Planning Predictive Modeling Technical Developmt Business Process Redesign Biz & Systems Integration Organize Change Mgmt • What systems modifications are required to accommodate the process? • What will different people in different roles see throughout the process? Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 15 Perform Metrics & Reporting

End-To-End Implementation – Making Models Come Alive Biz & Technical Planning • Predictive Modeling Technical Developmt Business Process Redesign Biz & Systems Integration Organize Change Mgmt Perform Metrics & Reporting Identification of all new process stakeholders (Underwriting, actuarial, systems, executive, legal/regulatory, claims, field, agency, training, project management, etc) • How will predictive modeling be described internally and externally to all stakeholders? • Manage any legal/regulatory issues and concerns • Once communicated, how to deal with questions, concerns, issues, etc. from Underwriters, field personnel, agents, market analysts, etc. • Development of change management, communication, and training activities and materials • Development of necessary implementation materials for all stakeholders • Development of feedback mechanisms using objective and subjective criteria Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 16

End-To-End Implementation – Making Models Come Alive Biz & Technical Planning Predictive Modeling Technical Developmt Business Process Redesign Biz & Systems Integration Organize Change Mgmt Perform Metrics & Reporting • Creation of management reports and metrics measurement processes including dashboards • Communication of model usage, results tracking, and management metrics at all process points to all constituencies • Loop back processes to manage compliance or deviation of model usage business plan Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 17

Contact Information John Lucker Principal Deloitte Consulting 860 -543 -7322 JLucker@Deloitte. com Michele Yeagley Asst. Vice President Harleysville Insurance 215 -256 -5403 MYeagley@Harleysville. Group. com Copyright 2005 – © 2005 – Deloitte Development LLC & Harleysville Insurance. 18