Artificial intelligence that mines injured workers’ medical and claims records is helping employers and insurers cut costs by spotting difficult or suspicious claims, but the technology remains a tool for claims handlers rather than a replacement. 

By analyzing injury and claims handler notes, background reports, prognosis details, and emails and transcripts of phone calls, AI can help claims staff predict the trajectory of a claim and intervene. 

Some experts say the technology, which has saved employers and insurers millions in workers comp costs, holds promise in uncovering workers comp fraud — both among claimants who fake or exaggerate injuries and medical providers who prescribe bogus or unrelated treatments, or charge for unfulfilled treatments. 

The Coalition Against Insurance Fraud  says that typically 1% to 2% of workers comp claims are fraudulent, and the organization published a report in 2023 that found that up to 16% of comp claims have some elements of fraud, accounting for $9 billion in losses for insurers and employers. 

There is little, if any, available data on fraud uncovered with AI, but experts say the industry is poised to deploy the technology. Several companies have launched products in recent months or are in the process of developing the capabilities to find fraud patterns in claims data. 

“We can use techniques like pattern recognition and predictive models, which can analyze historical claims data to identify any kind of patterns indicative of fraud,” said Sam Krishnamurthy, Alpharetta, Georgia-based chief technology officer in digital solutions with Crawford & Co. 

AI has been found to be “incredibly powerful in the extraction of information,” said Jeff Gurtcheff, chief claims officer at CorVel Corp.

On the claims side, workers comp fraud often shows similar characteristics, and technology can aggregate details found in mounds of claims data. The reports generated can help uncover patterns such as exaggerated injuries, lack of proof that an injury occurred at work, and malingering, experts say.

For provider fraud, spotting similarities or, in some cases, anomalies in prescribing and treatment patterns among doctors could help uncover fraud rings, they say, adding there is also some promise in finding litigation patterns among plaintiffs attorneys. 

“The more unstructured and structured data we have we can create a target data discovery model, which is a very powerful approach for detecting and mitigating fraudulent behaviors,” Mr. Krishnamurthy said. 

“We then use these advanced data analytics and machine learning techniques, not only to uncover the hidden patterns, but mainly to also detect these anomalies and prioritize them for investigation. And that will eventually lead to more efficient and effective fraud prevention strategies.” 

In June, Sunnyvale, California-based Clara Analytics Inc., which provides insurance technology services to insurers and employers, launched a fraud detection product that it says can leverage the company’s AI platform and large workers compensation datasets “to increase visibility into suspicious claims.” 

Mubbin Rabbani, Clara’s Medford, Massachusetts-based chief products officer, said the new capabilities are a natural progression from what Clara and many similar companies already do: provide data to help organizations manage claims. 

“Throughout our engagement with our customers we have built a very large data set of performance data around providers, doctors, hospitals, etc., as well as attorneys, both defense counsel and plaintiff attorneys, on these claims,” he said. “Using that allows us to now start to build on additional capabilities to what we currently offer, all in the spirit of deriving optimal claim outcomes, specifically now in fraud.” 

Tyler Kennedy, Boston-based lead developer for Gain Life Inc., which also provides AI services in the comp industry, said the capabilities show promise. “You’ve got this data and you’re trying to search for anomalies, or you’re trying to search for patterns, which is kind of what fraud is,” he said. 

Both Mr. Rabbani and Mr. Kennedy say the endeavor requires caution (see related story below) and that a good AI fraud program won’t investigate fraud — it just points out the possibility of fraud, given certain claim characteristics. 

“It’s not in our role to say (whether) it’s fraud or not,” Mr. Rabbani said. “What we are trying to do is refer (the information to a client), which will do that deeper investigation and take the corrective actions that may be needed if it is indeed fraud.” 

False positives, burden of proof slow AI adoption among insurers

Companies looking at artificial intelligence as an avenue for uncovering workers compensation fraud should tread carefully, experts say. 

AI is imperfect and can misrepresent findings and conclusions, finding coincidences instead of fraud and leading investigators astray, they say.

“This technology is not without its challenges,” a spokesman for the Coalition Against Insurance Fraud wrote in an email. “It can produce false positives, leading to unnecessary investigations and customer dissatisfaction.” 

Jeff Gurtcheff, chief claims officer at CorVel Corp., said the challenge is not “the ability to identify lots of red flags through the use of AI” but whether insurers can “truly prove fraud or intent (to commit fraud) or persuade an administrative body.” 

Mr. Gurtcheff gave the example of a malingering workers comp claimant video-recorded on a golf course while attesting he still had a bad back. “You take that video in front of an administrative body, which then says maybe he was just having a good day. Therein lie some of the challenges in this space.” 

Brad Young, Chesterfield, Missouri-based member at Harris Dowell Fisher & Young LC, which represents employers, said workers comp fraud is “very difficult to prove” to begin with. 

“The insurance carrier is going to have the burden of proof to demonstrate fraud,” he said. Investigations often raise privacy concerns, and the process for prosecuting takes on “a general bias in favor of claimants.” 

“I’ve had numerous cases over the years where I’ve had clear and convincing evidence of claimant fraud,” he said. “I had one where the claimant looked to his left, looked to his right, made sure no one was watching, laid down on the ground, and then started screaming. And I submitted that to the (state) fraud unit, and the fraud unit said that was insufficient evidence to demonstrate fraud.” 

Jeff Adelson, a partner with Irvine, California-based Bober, Peterson & Koby LLP, is also skeptical of what AI can do when it comes to fraud convictions. 

In California, where regulators and researchers contend the Los Angeles area is a hotbed for suspicious workers comp claimant activity, it wouldn’t be “fair” to investigate every claim as possibly fraudulent if it shares the same characteristics of another claim found to be fraudulent, Mr. Adelson said. His opinion aside, he said, the state limits the scope of some investigations. 

California law requires that surveillance be supported by an “articulable suspicion of a fraudulent insurance claim” before an investigation is triggered, he said, adding that forgoing that burden could lead to civil liability.