Reactive to Preventive: Managing Fraud with Databricks OpenCV and GenAI

Fraud, costing billions annually, is a significant challenge in the insurance industry. This presentation highlights Manulife's shift from a reactive fraud detection system to a real-time, AI-driven management system. One of the biggest hurdles in preventing fraudulent transactions in real time is due to poor data quality in scanned images. We'll present the solution and architecture of how we improved data extraction and anomaly detection using open source vision and LLMs, Azure Cognitive search, Document AI, link analysis and scaled traditional ML models within the Azure-Databricks platform. Our new system, integrating data from multiple sources, provides risk scores for providers, members, and fraud rings and flags potential fraudulent claims. This transformation helped avoid fraudulent claims worth millions, boost accuracy by 60+%, 12X speedup, and fostered positive behavioral change. We will cover our journey, lessons learned, technical details of our new system, and plans.