Key takeaways
- How a French home insurer's churn model flagged about a third of policyholders likely to leave, against an estimated 900,000 euro annual premium loss
- How the same modeling approach applied to prescription fraud detection for a US health insurer, targeting an estimated 4.5 million dollars a year found among just 1,000 fraud cases in 500,000 claims
- Why explainability matters in regulated insurance, and how Shapley values, partial dependence plots, and surrogate decision trees show why a model made each prediction
- How churn-prone customers were proactively targeted with renewal offers and more frequent outreach ahead of their 21-day pre-renewal window
- Where image analysis is headed for auto claims: detecting whether a car is even present, matching color and brand against policy records, and estimating damage location and severity
- Results cited from other financial services clients, including an 80 percent reduction in fraud cases over three years and a 69 percent cut in false-positive alerts during customer screening
Overview
In this session, H2O.ai's insurance team maps where AI creates value across the European insurance market, from an estimated 16 billion euros in annual undetected fraud to a home insurer's one million euro churn problem. The recording walks through two live customer builds in Driverless AI: a churn model for a French home insurer and a prescription fraud model for a US health insurer. It also previews image-based use cases for auto claims, with Shapley values and surrogate decision trees keeping each prediction explainable.