AI-Driven Fraud Detection in Homeowners and Renters Insurance Claims

Abstract

Homeowners and renters insurance claims are numerous and variable. Insurance carriers have every intention to pay legitimate claims, yet HRI insurance claims present unique challenges because the insured's loss is often due to a perils insured against. It is precisely this unpredictability associated with perils insured against that creates the greatest opportunity for fraud. The urgency to relieve the insured of his or her distress generally short-circuits the process of investigation. Moreover, the norm within the industry is to settle first and investigate later. This environment creates a breeding ground for both opportunistic and organized fraud. A substantial portion of the claims activity in the HRI space represents organized fraud perpetrated by groups who, based on their prior experiences, believe they can navigate around the later-in-the-process investigation hurdles without consequence. If these fraudsters can bypass the case management system and the carrier's investigative staff, then in the months and years ahead, they will refile and settle again and again and again, with impunity.

Internet search activity data provides unique insight into fraud risk. Search activity is a leading indicator of demand for many types of goods and services. Furthermore, the Internet search process is relatively frictionless. Estimates suggest that over 200 billion searches are conducted each year using major search engines; this number is rapidly growing. Consequently, the problem addressed in this paper is to develop a case management and decision support system that leverages existing online activity data to help insurance carriers identify potentially fraudulent HRI claims before settlement. As a response to the market need, the developed system provides a near real-time predictive infrastructure that allows for complementary offline predictive analytics and online predictive queries driven by state-of-the-art prediction models and predictive queries. Additionally, the proposed case management infrastructure allows for the development of both generic and customized backend and frontend case management modules that support claims processing across carriers.