Underwriting Knowledge Graph Construction, Maintenance and Its Application on Explainable Exclusion

The current underwriting rule engine (URE) that our industry partner and many other companies have implemented is a rule-based system that streamlined the underwriting manual into a customisable question tree flowchart with question branches leading to exclusion codes as the final result. As the hardcoded process proceeds, the task becomes sub-optimal and loses the customisation aspect for the customers. To address these challenges, we propose constructing an Underwriting Knowledge Graph (UKG) that incorporates an overview of customer risk profiles using explainable exclusion. The UKG leverages interconnected information related to customer attributes, historical data, and exclusion codes to provide valuable insights into risk profiles and potential associations with specific exclusion codes. Our investigation into domain-specific requirements leads to the first-ever UKG, based on real-life underwriting data from an insurance company in Australia. We also present a novel method for UKG maintenance that facilitates multi-label classification. The UKG offers a comprehensive view of the insurance ecosystem, aids risk profile visualization, and enables explainable exclusion classifications. With potential applicability across industries, the UKG promises to improve risk assessment and decision-making for other highly personalized services beyond life insurance.