Taking Underwriting to the Next Level with Predictive Analytics Innovative Approaches in a Tough Market Hansong Choi Underwriting Strategist/ Data Specialist/Modeler Prudential of Korea Nitin Basant Analytic Science FICO © 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation’s express consent. The Power of Predictive Analytics How can predictive analytics help transform underwriting in the insurance industry? © 2014 Fair Isaac Corporation. Confidential. Agenda ►Introduction ►Life Insurance Market in Korea ►An Utilization of Predictive Model in Underwriting ► EUS (Expert Underwriting System) Model ► Preferred Underwriting Model ► Tele Interview Model ►New Ideas © 2014 Fair Isaac Corporation. Confidential. Introduction ► Who Are We? © 2014 Fair Isaac Corporation. Confidential. Introduction ► Prudential of Korea Asset Total asset of $US 11.2 billion Net profit of $US 183.6 million RBC The highest level of RBC Ratio in the industry = 432.2% CSI © 2014 Fair Isaac Corporation. Confidential. MDRT 92.3%, CIC 35% 13th Persistency rate 87.9% Life Insurance Market in Korea © 2014 Fair Isaac Corporation. Confidential. Life Insurance Market in Korea ► Enhanced Financial Supervisory Services (FSS) regulatory oversight The size of Korea life insurance market 91.2 billion USD, M/S 3.5% Global rank 8th ► Blind spot of National Health Insurance Medical cost Personal Burden ► Intensified © 2014 Fair Isaac Corporation. Confidential. National health insurance moral hazard Feature of Underwriting Introduction of predictive modeling No discrimination of underwriting Limited collection of personal information The claims paid by insurance fraud is estimated $US 3.4 billion © 2014 Fair Isaac Corporation. Confidential. Utilization of Predictive Models in Underwriting ► EUS (Expert Underwriting System) Model © 2014 Fair Isaac Corporation. Confidential. EUS Model EUS (Expert Underwriting System) ► Development Objectives Target Method Strengthen loss ratio management Build risk DB, predictive model Enhance underwriter’s professionalism Improve underwriting efficiency + rule set compliment Objective 2011 Introduction Risk DB and predictive model N Y Automated Issue Ratio 8% 45% © 2014 Fair Isaac Corporation. Confidential. EUS Model EUS (Expert Underwriting System) ► Development review BRMS(Business Rule Management System) How to handle the policies? (Operation’s aspect) FICO® Blaze Advisor Predictive Model There are subsidiary Information FICO® Model Builder © 2014 Fair Isaac Corporation. Confidential. EUS Model EUS Process and Utilization of Model Operation Work with tiny risk Agency Score Validator Medical Examination Tele Interview © 2014 Fair Isaac Corporation. Confidential. Underwriter Death Score Surgery Score Hospitalization Etc. Risk Mart Predictive Model DB EUS Model Predictive Model Development DB upgrade Risk Mart risk factor1 risk factor2 risk factor3 risk factorN Defining a target incidence within 3 years © 2014 Fair Isaac Corporation. Confidential. GAM (Generalized Addictive Model) Modeling algorithm Predictive Model (scoring) Death Surgery Hospitalization : Apply to new business underwriting Focus underwriting on high risk (score) cases EUS Model Definition of Hospitalization Coverage Coverage: all of hospitalization in terms of disease and accident ► Guaranteed ►3 amount: KRW 10 thousand won(10$) per a day days contestable period, maximum 180 days ► *5,000 face amount = 5 x 10 thousand won = 50 thousand won (10$) per a day © 2014 Fair Isaac Corporation. Confidential. EUS Model Utilization of EUS Model Hospitalization Score grade Low Risk ► Give Automatic underwriting ► Enlarge maximum face amount ► Mitigate financial/medical underwriting 2 3 25% 21.4% 4 ► General Underwriting Selective tele Interview Expected morbidity ratio within 3 years 15.0% 14.0% 10% 7.5% 6.2% 5.6% © 2014 Fair Isaac Corporation. Confidential. for all policies 7.4% 5.4% 3.2% 2.6% 2.2% 6 5 4 6 ► Inspection 11.3% 9.2% 5% 7 (Cumulative Morbidity ratio) After 2 years 11.1% 5 High Risk After 1 year 20.5% 20% 15% ► ratio of accepted policies preferential treatment ► 1 ► Morbidity 5.3% 3.8% 2.9% 1.7% 1.4% 0.0% 2.3% 0% 7 3 2 1 Total EUS Model Effect of Scoring Model Introduction ► The underwriters are starting to pay attention to not only medical information, but also non-medical information Medical Examination High Risk Financial inquiry Tele Interview Review All paid out claim © 2014 Fair Isaac Corporation. Confidential. Low Risk Automated Underwriting EUS Model Benefit of Score Model ► A&H Loss ratio <A&H Loss ratio by coverage(≤ 2 years)> 120% CY'10 CY'11 CY'12 CY'13 100% 80% 60% 40% 20% 0% Surgery © 2014 Fair Isaac Corporation. Confidential. Hospi. Cancer CI TOTAL EUS Model Utilization of EUS Model Projection Score Underwriting Accept Loss Ratio Morbidity Ratio Score Risk Factors 1. Agent’s hospitalization loss ratio for 1 year 2. Region 3. Occupation section 4. Insured age 5. Gender © 2014 Fair Isaac Corporation. Confidential. Underwriting Effects Reject How did you reject low score? Reject policy’s Morbidity & Loss ratio projection Utilization of Predictive Models in Underwriting ► Preferred Underwriting © 2014 Fair Isaac Corporation. Confidential. Preferred Underwriting Preferred in Korea Market ► Preferred Condition in Korea As a person can apply base plan… Smoking Habit ►Non-smoker for the last 1 year Blood Pressure ►Systolic Pressure 110~139 All conditions should be met © 2014 Fair Isaac Corporation. Confidential. BMI (Body Mass Index) ►Weight(Kg) Height(m²) ►20–27.9 / Preferred Underwriting Why does Preferred Insured Show Higher Loss Ratio in Every Years? ► Loss ratio 60% 50% 40% 34% 26% 30% 20% 10% 0% Preferred Insured © 2014 Fair Isaac Corporation. Confidential. Standard Insured Preferred Underwriting Factor Selection ► What kind of factors affect death coverage? Factors (= Information) © 2014 Fair Isaac Corporation. Confidential. Preferred Preferred Underwriting Preferred Differentiation Method ► How to choose preferred insured ► Should ► meet all conditions Knockout Point Current Korea © 2014 Fair Isaac Corporation. Confidential. Meet at least several items out of all conditions ► Reaching a certain score that gives a specific item’s weights and score Debit/Credit Preferred Underwriting We Knew Lifestyle Shows a Greater Impact than Physical Condition © 2014 Fair Isaac Corporation. Confidential. Preferred Underwriting How Much Weight Was Given? © 2014 Fair Isaac Corporation. Confidential. A Utilization of Predictive Models in Underwriting ► Tele Interview Model © 2014 Fair Isaac Corporation. Confidential. Selective TI (Tele Interview) Model Background ► In Korea, Only specified information can be collected. Cannot be rejected as a direct cause of MIB and RX profile, MVR report. So, some information from customers should be collected in order to underwrite through Tele Interview © 2014 Fair Isaac Corporation. Confidential. Selective TI (Tele Interview) Model Who can be target to call? Targeting Predictive Modeling Information 28 © 2014 Fair Isaac Corporation. Confidential. Selective TI (Tele Interview) Model Definition of Target Hit ratio Medical rejection + extra charge+ exclusion rider + reduced Face amount Tele Interview Target © 2014 Fair Isaac Corporation. Confidential. = 15.3% Effective Tele Interview ► Selective ► TI Target Ratio Insured Age General Death Benefit Amount ≤40 ≤50 ≤60 >60 ≤10 mil. >10 mil. >30 mil. Non-medical Exam Selective TI >50 mil. Tele Interview >100 mil. >200 mil. >15 bil. © 2014 Fair Isaac Corporation. Confidential. 30%↑ (Step 1) Special Exam C >300 mil. >13 bil. Special Exam A 4.3% Special Exam B 0.6% 4.0% >7 mil. 50%↑ (Step 2) 91.8% 0.1% Special Exam D 0.1% 15.3% (Now) Special Exam E 0.0% Selective TI (Tele Interview) Model Modeling Variable Apply신청여부 preferred 우량체 Apply preferred y/n 0.4017 Number of riders 0.3193 Sum of general death benefit amount Hit history 0.1798 0.0539 Elapsed period Product Category Replacement contract Y/N Insured Age 0.043 0.0317 0.0157 © 2014 Fair Isaac Corporation. Confidential. Unscaled 39 0.319 [-, 1) -21 -0.242 -1.411 [1, 2) -34 -0.341 [2, 3) -2 -0.031 [-, 2) -110 -0.889 [3, 7) -3 -0.01 [2, 3) -9 -0.073 [7, 10) 16 0.184 [3, 5) 53 0.427 [10, +) 25 0.229 [5, +) 74 0.596 Product 상품종류 Category -86 -0.707 63 0.508 TERM -8 -0.096 5 0.038 WHOLELIFE 4 0.028 [150000001, 250000000) -35 -0.281 36 0.369 [250000000, +) -78 -0.629 -10 -0.078 No 84 0.675 Yes (-, 20000) -79 -0.638 [-, 31) [20000, 35000) -31 -0.247 [31, 38) 2 0.006 [35000, 55000) -5 -0.044 [38, 42) 1 -0.004 [55000, 115000) 18 0.145 [42, 47) -7 -0.07 [115000, 290000) 18 0.146 [47, 52) 0 0.018 [290000, +) -1 -0.008 [52, +) 27 0.294 Sum of general Death Benefit amount 기계약합산 일반사망 가입금액 RI CHILD DUHC FAMILY_INCOME MULTI_PLUS PEN_SAV Replacement contract Y/N 대체계약여부 Hit History 적출이력 N 없음 Y 있음 0.0378 Scaled Elapsed period POK 최초 청약후 기간(년) -175 [50000001, 150000001) 0.0498 Variable Unscaled N 없음 Y(Preferred) 우량체 Number 특약건수 of riders (-, 50000001) Real premium Scaled y/n 3 0.028 -125 -1.249 -25 -0.232 Insured 가입연령 Age Real Premium 실납입보험료 Selective TI (Tele Interview) Model ► Roc curve © 2014 Fair Isaac Corporation. Confidential. ► Model Result setID Score Variable train Model_score test Model_score Divergence 1.127 1.058 AUC KS KS percentile 0.769 41.529 62.70% 0.762 40.094 61.40% Selective TI (Tele Interview) Model ► Total A&H loss ratio (≤2 years) 100% ► Target ratio 35% 90% 30% 80% 70% 25% 60% 20% 50% 15% 40% 10% 30% 20% 5% 10% 0% Before 2014 0% CY'10 CY'11 © 2014 Fair Isaac Corporation. Confidential. CY'12 CY'13 2014.1Q 2014.2Q 2014.3Q New Ideas © 2014 Fair Isaac Corporation. Confidential. New Idea Underwriting Model Simplified issue Fraud detection Preferred Medical Exam Extra Charge Loss ratio © 2014 Fair Isaac Corporation. Confidential. Inforce Marketing Thank You! Hansong Choi hansongchoi@prudential.co.kr © 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation’s express consent. Learn More at FICO World Related Sessions ►Harnessing Network Analytics to Better Combat Fraud ►Product Showcase: Multichannel Communication Solutions for Insurance ►FICO Roadmap for Insurance Fraud Products in Solution Center ►FICO® Model Builder ►FICO® Blaze Advisor® business rules management system Experts at FICO World ►Nitin Basant ►Scott Horwitz White Papers Online ►Connecting Insurance Customers – and Decisions – with Technology Blogs ►www.fico.com/blog © 2014 Fair Isaac Corporation. Confidential. Please rate this session online! Hansong Choi hansongchoi@prudential.co.kr © 2014 Fair Isaac Corporation. Confidential. Nitin Basant nitinbasant@fico.com # Appendix: Selected Factors by Coverage EUS (Expert Underwriting System) Model Death 1. Agent’s total loss ratio for 1 year 2. Contractor=Beneficiary Y/N 3. Insured age 4. Occupation section 5. Disease diagnosis below 5 years 6. Region Disability 1. 2. 3. 4. 5. Gender Agent’s loss ratio for 1year Change job within 2year Job LP loss ratio fro 1year in terms of accidental death 6. Hospitalization, surgery rider y/n Surgery 1. 2. 3. 4. LP loss ratio for 1year Age Gender Cumulative surgery Face amount 5. Number of no-warrant contract © 2014 Fair Isaac Corporation. Confidential. CI 1. 2. 3. 4. 5. 6. Gender Diagnosis History ≤ 5 years Age Voluntary contract LP Job grade BMI Cancer 1. 2. 3. 4. 5. 6. 7. 8. 9. Age LP loss ratio for 1year Gender Contract date Cancer rider y/n Relationship of policy owner and insured Job Notice Examination within 5 years LP loss ratio for 1year Hospitalization 1. LP Hospitalization loss ratio for 1 years 2. Region 3. Occupation section 4. Insured age 5. gender Etc. 1. 2. 3. 4. Region Notable LP Gender Relationship of policy owner and insured 5. LP Job grade