The Fraud Squad

Whether it's done by customers, employees or organized criminals, fraud takes a bite out of business's bottom line. Here's what CSOs can do about it.

Companies can either buy customizable software or write their own rules-based programs that analyze network activity for specific indicators of fraud. For example, if corporate policy decrees that all purchases above $20,000 require approval, then a program that flags purchase orders for amounts between $19,000 and $20,000 could be useful in fraud monitoring. Similarly, a program could compare vendor addresses with employee addresses to detect "ghost" vendors.

The insurance industry is a frequent target of fraudsters. According to the Insurance Information Institute, property and casualty insurers alone pay about $30 billion annually in fraudulent claims (which includes the administrative and investigative costs of fraud). This leads, as we're often reminded, to higher premiums for consumers.

To drive down the cost of fraud in its auto and home division, MetLife has teamed with Computer Sciences to develop an early fraud-detection system. The program, called @First, combines rules-based technology with predictive modeling to identify possible fraudulent activity. Previously, MetLife Auto and Home relied exclusively on the company's claims representatives to spot possible fraud. But picking up on many of the common red flags (for example, an individual who files a claim within the first 30 days after obtaining a policy) required that claims reps note every policy's inception datewhich didn't always happen. A claim that came through on a Friday before a holiday weekend, or at some other time when reps were unusually distracted, could slip through unnoticed.

John Sargent, manager of the corporate SIU for MetLife Auto and Home, wanted to provide a safety net. The @First system scours claims for signs of possible fraud: vehicle ID numbers and addresses similar to those of other claimants, drop boxes that could indicate a fictitious address, or the names of doctors and auto body shops that have been previously sanctioned. Using predictive modeling, the program looks at historical patterns of fraud and scores each claim for characteristics that in the past have indicated fraud. MetLife is currently using a test version of the technology and expects to have the software fully rolled out by the end of this month. To date, Sargent estimates as much as a 10 percent increase in the flagging of suspicious claims. But he cautions that even the best technology won't replace the skills of a seasoned claims rep. "No system captures a reluctant voice on the phone or somebody who can never be contacted by phone but is able to call the claim rep," he says. "We rely on their gut instincts."