Having saved $7.5 million through analytics-driven fraud detection over the past two and a half years, CNA Insurance has signed a two-year extension with SAS for an undisclosed sum.
At the ACORD LOMA conference, Tim Wolfe, director of CNA's special investigative unit, said under the extension, CNA will refresh the five models the company has built with SAS’s Fraud Framework for Insurance (FFI), including workers’ compensation, general liability, commercial property, commercial auto and i’s social network link analysis model, which initially was launched in January 2012.
Also see:
Wolfe explains that CNA currently scores claims individually for a likelihood of fraud; then they are scanned again to search for patterns, including IDs of medical and legal providers. Those claims then are triaged and considered for father investigation. Three CNA analysts then consider the results, referred to as network alerts, and drill down to determine why those claims have been highlighted.
Wolfe said CNA began using analytics for fraud detection in 2008 with a homegrown application and soon switched to a vendor solution that offered marginally better hit rates before selecting SAS FFI. In 2011, SAS built four models for the company, including workers’ comp, general liability, commercial property and commercial auto. Social networking analysis, which uses link analysis visualization, followed later. “That went live in January of 2012,” Wolfe said. Open claims are run through those models on a weekly basis, he explains.
“On Monday morning, the claims are scored individually for fraud potential; these are not claims that have been referred by adjusters. Then the other inspection is for social networks, which typically identify legal and medical providers. Those networks contain dozens or even hundreds of claims.”
Wolfe has two teams, the first does the individual claim review after it has been tagged by the model. The claim then is triaged, and if accepted, the analyst contacts an adjustor to discuss a plan of action and follow through with an investigation. The social networking analysis is more complex and involves medical billing details. Three analysts review those network alerts and drill down into the claim to determine why the model has identified the network for fraud potential.
Claims are then investigated, declined or monitored, with decision information fed back to the SAS Reporting Studio, Wolfe said. “That was one thing I was keen on doing right from the onset: to track the results of the claims we investigate from the model vs. the ones that come from the adjustors. The results have been really interesting to see because when I put my business case together, I projected an average savings per claim. And the hit rates have varied by lines of business. Workers’ comp being the highest-performing model. The hit rate is around 27 percent, which is really good,” Wolfe said, explaining that for every 100 claim alerts, 27 are accepted for investigation.
“Those are typically the highest scoring ones. But that was something else I wanted to see. Of the claims we accept, are they all in the top scoring categories? And the interesting thing is not necessarily. It depends on the line of business. With Auto and general liability, they are in the top-scoring sector. Workers’ comp is more middle-of-the-road. You never would have expected that. The other surprise is that they average savings is lower. The savings are 50 percent of what we originally projected. The reason is that the longer a claim stays open, the more it costs. We are getting in earlier on these claims, investigating them and doing it more cost effectively.”
Workers comp accounts for 40 to 50 percent of the investigated claims, and Wolfe said adjustors are encouraged to escalate suspect claims immediately. “Where we are getting the best bang for the buck is getting suspicious claims early, where we are doing more investigation vs. surveillance; we are saving money there. We are shutting the claim down early rather than relying on getting lucky with surveillance.”
The recent fraud case against California hospital owner Michael Drobot, who according to the Federal Bureau of Investigation, allegedly ran a health care fraud scheme that involved tens of millions of dollars in illegal kickbacks in exchange for referrals of thousands of patients who received spinal surgeries, is providing CNA with new opportunities to analyze for fraud, Wolfe said. The hospital referrals led to more than $500 million in bills being fraudulently submitted during last five years of the scheme, much of which was paid by the California worker’s compensation system. Wolfe said CNA was affected by only $5 million of that, and that the data is analyzed to figure out the strengths and weaknesses of the models so they can be improved.
“If it didn’t show up, I want to know why and reverse engineer things,” Wolfe said.
Also see:
