Success in Implementing Workers Compensation Predictive Models at Zurich Financial Services

Transcription

1 Success in Implementing Workers Compensation Predictive Models at Zurich Financial Services Joel Appelbaum Zurich Financial Services Chief Risk Officer of North America Underwriting Transformation May

2 Agenda What is Predictive Modeling All About What will it do for us? Lessons Learned Examples of How we have Implemented 2

3 Zurich Global footprint - Five continents countries 3rd largest commercial insurer 27 straight quarters of profitability 3

4 4

5 Predictive modeling feud What are predictors of future workers compensation losses? Prior losses/loss history Financial Health Safety Program Industry/SIC Code Location/Demographics X X X X X X X Change mgmt experience SEDIMENT X X X X X X 2X LAYERS X X X X X X 2X DX X X X X X X 1X X X X X X X X X X 5

6 What is the Value Proposition related to Predictive Analytics? Value Drivers: Implementation Methods: Loss Ratio Improvement Design risk based pricing structure Focus retention effort on profitable risks Identify potential premium leakage Expense Reduction Implement straight through processing Optimize audit procedures, loss control, service offerings Enhance our ability to detect fraud Profitable Growth Focus sales on profitable niches Target new business Identify highly profitable agencies Customer Centricity Reduce application information collection Increase our Cross-selling / Product Density 6

7 Expense Reduction Putting Predictive Analytics in the Underwriting Process Current Future Non- Value Added Non- Value Added Review of deductibles Collection of Financials Calculation of Quick ratios Providing the Financial Ranking Lowering costs and Increasing Premiums Audited 7

8 Decile Retention* Projected Loss Ratio Commercial insurance - Avoiding Adverse Selection Potential for Adverse Development 125% 68.0% 100% 67.0% 67.0% 75% 50% 64.5% 65.0% 66.0% 64.0% 66.0% 65.0% 64.0% Best 30% Middle 50% Worst 20% Projected LR 25% 63.0% 0% 62.0% Keep all Lose best 5% Lose best 15% Lose best 25% Vary Retention Scenario Description 8

9 Commercial insurance Frequency vs. Severity Commercial Lines Predictive Modeling, Thomas R. Bayley, Emphasis magazine 9

10 Predictive Modeling Why Now? Predictive Modeling is becoming widespread in commercial lines. Failure to act will result in adverse selection! 10 Source: Towers and Perrin Survey

11 Top 10 Lessons Learned Analytical Transformation 1. Use Innovation and entrepreneurial principles 2. Be thought leaders 3. Reuse insight where you can Business Fundamentals Communicate 4. Apply project management competencies 5. Control scope successful bite size chunks 6. Sprint the last mile implementation is the only thing that counts 7. Measure outcomes against key performance indicators 8. Establish a communication plan upfront 9. Apply change management competencies to overcome adoption challenges 10. Share success 11

12 Data 12

13 Data - Cleansing / reconciling bad data elements is time consuming - Multiple data sources required - Commercial lines - Regulation - Legacy systems - Data Stewardship savvy leadership on what the data means 13

14 Keep it Simple! Reducing the Noise 14

15 IT Impact Predictive tools often bump into legacy systems which can be costly 15

16 The Last Mile the most important activity for the leadership team is to keep analytical initiatives on track and to monitor outcomes to ensure that anticipated benefits are achieved. Competing on Analytics, Thomas Davenport, Harvard School of Business,

17 Opportunities and Value of Predictive Modeling Supports a risk-based pricing structure Focuses retention efforts on profitable risks Identify potential premium leakage Increases profitability and decreases loss ratio by avoiding adverse selection Does not eliminate the need for underwriter judgment and accountability but creates a more analytically based framework in which judgments can be applied or evaluated The model results will be used in conjunction with your own underwriting insights and consistent with filed rating plans M/D/2009 Program Administrator Work Comp Modeling Training 17

18 How can you benefit from the Work Comp Model? The additional segmentation provided by the model scores: - Shows you the range of scores for all renewals in the month - Helps you target rate changes for your renewal policies - Provides additional facts and insight for more accurate risk selection The policy Reason Codes provide additional insight and may direct you to risk factors you may want to investigate M/D/2009 Program Administrator Work Comp Modeling Training 18

19 Price Change & LR Line of Business Countrywide Pricing, Loss ratio by Decile Bands Renewal Price Change & Loss Ratio (Retail) % % % 35.20% 35.30% 32.30% 29.90% 30.08% 30.00% 30.50% 29.80% 30.44% Best 3 RPC Avg 5 RPC % Worst 2 RPC % Best 3 LR % % 5.00% 3.90% 4.60% 6.40% 4.10% 4.50% 7.30% 5.20% 6.10% 9.20% Avg 5 LR Worst 2 LR 0.00% Q1 Q2 Q3 Month 19

20 Line of Business Countrywide Retention Results by Decile Bands Retention Percentage Retention (Retail) % % % % % % % 76.00% 77.00% 74.50% 75.10% 71.80% 70.85% 65.60% 63.40% 62.10% Best 3 Avg 5 Worst % % 0.00% Q1 Q2 Q3 Month 20

21 Predictive modeling as UW assistance on the account level Heat map concept: - Information is presented on an account-level basis - The decile indicates to the UW the accounts predicted loss ratio relativity - Five variables will be presented using red, yellow, and green to provide a quick view of the contributors to the overall model score Heat Map (Best, Good, Average, Bad, Worst) Decile # Pay Exp Yrs in Buz CCSP Trans Use Lightning 2 Bad Average Best Good Average 21

22 Underwriting Actions viewed just as colors and not using words Severly underpriced Underpriced Risk Adequate Price Well Priced Strongly Priced 22

23 Underwriting Actions viewed just as colors and not using words Severly underpriced Underpriced Risk Adequate Price Well Priced Strongly Priced Significant Rate<80 AP/TP AP/TP AP/TP AP/TP >130 AP/TP 23

24 Start off Simple Displays policy and modeling results for selected renewal policy Primary tool for Underwriters Displays detailed data Drop downs for policy search Basic policy data Premium and loss data D&B data Company, Exp and Schedule mod PM Indication Reason Code Indications Notes field for UW comments Joe s Sandwich Shop M/D/2009 Program Administrator Work Comp Modeling Training 24

25 Company Selection Worksheet All American John Doe Big Gas OK /01/2010 M/D/2009 Program Administrator Work Comp Modeling Training 25

26 Other Avenues of Value in Implementing a WC Predictive Model Automatic and Consistent Company Selection guidance Presentation and Reuse of D&B Data for multiple purposes throughout process steps (reduced double hits to D&B database) Combining External data sources (D&B info, experience mods, sic codes) and proactively enhancing data and scoring potential accounts Scoring of Prospective Books of Business Premium < $50K Loss Ratio Capped Loss Ratio (100K) Capped Loss Ratio (250K) Predicted LRR 26

27

28 Industry Specific Considerations - Commercial insurance Market Cycles - Right price for the risk vs. what the market will bear - Competition - Marketing insight Underwriting Actions - Directional - Continuum of actions - Renewal strategies Frequency vs. Severity - Frequency based = more stable; more credible - Severity based = less stable; less credible Market Forces - Expense reduction - Customer expectations - Loss inflation trend 28

29 Commercial insurance Market Cycles Average Complete Market Cycle lasts 5-10 Years Market Softening Average soft market lasts 2 6 years Market Hardening Average hard market lasts 3 4 years *Trigger N Months Filing Process 6 Months Premium change 15 Months Typical Duration of Rate Response = 21+ Months (17%-35% of the market cycle) * Rate Response Triggers 1) ISO Rate changes and/or 2) Annual pricing projections 29