Client Details
A top-10 U.S. auto insurance carrier insuring over 20 million vehicles. The company is known for its digital-first approach and aggressive stance on fraud prevention.
Challenge
The traditional fraud detection system relied on post-claims review, which allowed fraudulent claims to slip through and required expensive recovery processes. The absence of real-time alerting mechanisms exposed the company to increasing fraud losses.
Solution
The team developed and deployed a real-time fraud detection solution using Databricks, MLflow, and Azure Stream Analytics. A combination of logistic regression and isolation forest models were trained on labeled fraud cases and enriched telematics data.
Streaming data from claims intake systems, vehicle sensors, and third-party providers was ingested and scored in real time. Suspicious claims were automatically flagged for manual review or routed to a specialized handling path.
The Impact
✅ Fraud Reduction: Detected 30% more fraudulent cases within seconds of submission.
✅ Cost Savings: Estimated $8M in annual savings from fraud prevention and early intervention.
✅ Real-Time Alerts: Enabled proactive investigation and reduced false positives by 45%.
✅ Scalable Framework: The architecture is now used to build other real-time ML use cases like accident severity prediction.
