Business Objective
Our Client, one of the largest life insurance companies in the US, wanted to develop a straight through processing (STP) framework for the accelerated underwriting (UW) of life insurance policies. The framework was to use a predictive model-based approach leveraging only fluid-less, examination-less, and non-APS (Attending Physician Statement) evidence data
Challenges
- Interpreting diverse evidence sources like application data, Motor Vehicle Records (MVR), Medical Information Bureau (MIB), prescription data, diagnosis data, etc. and understanding how and where they can be used
- Balancing data-driven insights with domain experience-driven decisions from underwriters
- Ensuring the models are interpretable with clear reasoning for the final decision, given UW risk class assignment is a regulated decision for adverse cases
Solution Methodology
- Developed a 2-step predictive model framework to process incoming applications using fluid-less, non-APS evidence data:
- Decline triage model: Classifies all applications into 3 groups – high mortality risk (automated decline), low mortality risk (automated approval and risk class assignment), and medium mortality risk (sent for manual UW assessment using additional evidence data)
- Risk class model: Classifies all the low mortality risk applications into the right risk classes
- Leveraged advanced boosting techniques for both decline triage model and risk class model
- Overlaid the models with a knock-out rules framework to capture rare mortality events
- Created an Adverse Reasoning Framework to provide top contributing reasons for decline and adverse risk class decisions
Business Impact
- The new AI-driven accelerated underwriting workflow estimated to save ~39% of evidence costs with ~40% STP and less than 4% decline leakage into the STP approved cases
- Decline triage model showed 86% accuracy for all non-APS ordered policies
- Risk class models showed 30% reduced mortality load compared to previous model
