If your customer base is a mile wide, sometimes its tough to detect fraud when its hidden more than an inch deep. The Government Employees Hospital Association (GEHA) felt that way. GEHA is a national health plan serving federal employees, retirees, and their dependents. With subscribers spread throughout the country, traditional fraud detection systems werent doing the job for GEHA.

Bob Greene, manager of data analysis for GEHA, believes traditional fraud detection systems need a reasonable number of claims from the same healthcare providers to determine if those providers were being honest in their billing. We dont have a significant share of claims per physician, says Greene. For systems that are heavily dependent on rules, this was a significant problem we struggled with for a number of years.With policyholders so spread out and nearly half of their claims submitted electronically, Greene felt better detection was needed because adjusters were no longer handling the paper they once used and didnt get the feel of the paper claims.

GEHA signed up with Fair Isaac (then known as HNC Software) at the end of 2001. Greene says he was familiar with the neural network technology the software company used, and GEHA was pleased to take part in the pilot program that Fair Isaac ran through the summer of 2002. The system, known as Payment Optimizer, used GEHAs own claims, not external benchmarks or comparisons, to develop its procedures. The system adapts to the local environmenthow GEHA codes, processes, and handles claims. You dont have to tell it by rule that you apply certain types of CPT modifiers and dont use others, says Greene. The neural network trains on your data and learns it for itself.