Large Health Insurance Plan:  Claims Fraud

Old Way

Six million claims a year were processed by using a simple rules-based, largely manual method. When fraud was detected, typically it was after payments had already been made, and it was too late to recover the funds. The pace of claims was increasing, which meant more staff were needed. In addition, some fraud was hard to uncover manually, because the small volume of claims per provider made it hard for staff to spot patterns.

EDM Way

Predictive analytic models -- in this case, neural nets designed to spot new and emerging patterns -- are integrated with business rules in a fraud detection decision service. The neural net "learns" new fraud patterns as they occur. Combined with dynamic profiling, it identifies a claim as fraudulent much earlier in the process. The decision service flags suspicious claims so that they can be investigated manually, which allows automatic payment of remaining claims without intervention. This method improves straight-through processing rates and meets regulatory requirements for prompt payment. Investigations of fraudulent claims are more effective, because the decision service clearly states reasons for the referral, allowing staff to focus immediately on suspicious elements. The service also detects errors in billing, because it detects any kind of aberrant behavior, from fraud to billing errors to clinical errors.