Fraud detection is basic to insurance operations, and never more so than during a down economy. In addition to ongoing efforts to ferret out organized fraud rings, carriers need to be on the lookout for increased instances of soft fraud, as debt-ridden and desperate consumers pad otherwise legitimate claims.

Since the upswing in fraud slices through all personal and commercial lines business, including P&C, workers' comp and health, carriers are more diligent than ever in their efforts to thwart the bad guys.

Fortunately, they now have a bevy of technological options, including predictive analytics and visual link analysis, to add to their traditional array of fraud detection tools. Read on to see how carriers are melding state-of-the-art with time-tested investigative techniques to separate fraudulent claims from legitimate ones, and improve their bottom lines.

In addition to the usual witches' brew of hard fraud, soft fraud, staged accidents and slip-and-fall fraud rings, insurance carriers must contend with increasing instances of fraud borne of a policyholder's adverse financial circumstances. The homeowner or SUV owner seeking to get out from under a loan with a little help from a can of kerosene is far from apocryphal. "Fraud has historically been committed out of acts of greed," says David Rioux, VP of corporate security and the special investigative unit (SIU) for Erie, Pa.-based Erie Insurance. "Now, given the economic situation we're in, fraud is being committed out of acts of desperation."

Perhaps as troubling as the fraudulent claims carriers manage to detect, are the frauds that escape detection.

"The funny thing about fraud is that nobody knows how much there really is," notes Donald Light, a senior analyst at Boston-based Celent. "Insurers only know how much they have found - they don't know how much they haven't found."

When it comes to fraud detection, experience is often the best teacher. "Insurers are always getting better at it, but, by nature, you almost have to be burned before you find the fire," adds Robin Harbage, C-counsel consultant at San Diego-based EMB in North America.

While the totality of insurance fraud is likely unknowable, new data from the Des Plaines, Ill.-based National Insurance Crime Bureau (NICB) offers one way of quantifying the problem. NICB data encompasses property/casualty, commercial and vehicle data, but the most pronounced surge occurred in auto claims, as suspicious car fires were up 20% from last year, suspicious auto glass claims up 76% and "phantom" accidents were up 46%.

In fact, for the first half of 2009, NICB data shows increases in nearly all referral categories compared with the first half of 2008. Overall, a total of 41,619 questionable claims were referred to NICB for closer review and investigation by member insurance companies in the first half of 2009, compared with 36,743 received during the same period a year earlier.

This upswing in fraud is not limited to the property/casualty sector. Workers' compensation claims can swell as workers anxious about layoffs file claims, aware that workers' compensation benefits are generally larger and longer-lasting than unemployment benefits.

So how do insurers stem this tide? By making better use of their most abundant and critical asset: data. Using predictive analytics to fight claims fraud is not new. Once the province of the well-heeled, the proliferation of predictive analytics technologies means now most any insurer has the financial and technical ability to bring these solutions right to their desktops. While the solutions have improved significantly in the past five to 10 years, sporting better graphic interfaces and greater use of visual link analysis to aid in the identification of suspicious claims, they are beginning to plateau. "From a pure technology perspective, there is not much new in the last few years," acknowledges Light.

The Human Factor

Absent technological leaps forward, how are insurers advancing the use of predictive analytics to fight fraud? One strategy is to use predictive analytics to augment human decision-making, or make better use of human capital. As analytic models shift from batch mode to real-time, they can help with areas such as adjuster assignments. For example, an experienced adjuster could be assigned to an accident scene while remnants of the accident are still present.

"I look at predictive analytics as a data watchdog - a guide for adjusters and investigators to be on the alert to critical data," says Gregory Melanson, manager, special investigation unit, for Warwick, R.I.-based MetLife Auto & Home. "You have finite resources, so you only want to investigate files that really need it."

Melanson says it is important to synthesize information culled from analytics with human interpretation. Insurers can integrate predictive analytics with existing business rules, such as special consideration for any claims that occur within the first two weeks of a new policy. "You are not going to pay three years' worth of premiums and then make a claim," Harbage says. "Many fraudsters will make a claim before the check clears or bounces."

While Melanson says the insurance industry is not trailing fraudsters, he sees areas where insurers can be more proactive. For example, incorporating existing technologies such as geo-coding can help investigators ferret out fraud, he says. "It may prompt an investigator to ask: Why would someone go to a medical clinic 50 miles away?"

Yet, just because a model flags a claim it doesn't necessarily mean the claim is fraudulent. Analytics can highlight exonerating behavior in a seemingly suspicious claim as well, saving the adjuster the time and aggravation of following a false lead. Analytics works best when paired with solid claims management and diligent investigative techniques. Melanson stresses that the technology is not an exact science, and companies need to guard against becoming over-reliant on analytics. "Often there are mitigating factors that are not in the data," he says.

Likewise, Rioux says predictive analytics work best as a conduit to the information needed by adjusters and investigators to make informed decisions. "From the first notice of loss throughout the entire life of the claim, we continuously score the claim, and there is a threshold we set," he says. "When it breaks that threshold, a notice will go to that claim handler."

Rioux says analytic models now account for one-third of referrals to SIU. "That's actually a very good number," he says. "The reality is there are some adjusters that don't benefit as much from predictive analytics; they find these things off gut feeling."

One area where Rioux concedes humans come up short is in consistency. Automation breeds consistency, he says. While adjusters and investigators can suffer from workload or training issues, predictive models do not. "It levels the playing field across an organization," he says. "One thing about technology is that it doesn't have good days or bad days. Our claims are being scored consistently across the organization."

The Notes

Oddly, one bountiful source of predictive information - the copious amount of notes created by adjusters - has only recently been incorporated in predictive models. Traditionally, carriers would use structured data such as historical records of known claims fraud, plus policy and public data to build their predictive models. Presently, steady advances in text mining capabilities have made this once-underutilized asset another primary feedstock for analytics. Rioux says Erie now incorporates adjuster notes into its predictive models, which were originally built using hundreds of predefined data fields such as names and addresses.

"One of the more recent additions to our solution was the introduction of adjuster loss notes because text mining is huge," says Rioux, who is also president of International Association of Special Investigators, Baltimore. "The notes tell the story. Data fields just hold pieces of information. Being able to read notes, and organize them in a fashion useful for the model, has really given predictive analytics a substantial lift."

Keith Ellis, sales engineer at Chicago-based SPSS Inc. concurs, noting an estimated 80% of claims data is in unstructured notes. "Being able to extract meaningful data from that text and apply it predictive models is a significant capability," he says.

The Growing Mix

Insurers are always collecting various types of information that might be predictive. Increasingly, they are adding ancillary financial data and automated public records, such as a bankruptcy data, to the mix as sophisticated anomaly detection models enable adjusters and investigators to easily identify suspicious data. What's more, new sources of data for predictive models are constantly becoming available. "We refresh our models each year because the data changes, and the types of things we investigate change as well," Rioux says.

Digitizing medical records presents a vast new feedstock of information for predictive modelers. In addition to the obvious applications for health insurers, such records could be helpful to auto investigators looking at a low-speed car accidents where significant injuries are claimed. Likewise, the digitizing of legal records can help carriers realize links between seemingly unrelated claims with common attorney representation.

"You have to be creative because it's not always something that is going to be on the claims file," adds EMB's Harbage. "You look for outliers - things that just don't fit. If somebody is insuring a home and its clear that the insurance to value is way above average for that geographic area, that would be a huge red flag, even before you do the inspection."

The geographic aspect is important for carriers that operate in many states. Harbage notes that fraudsters tend to gravitate to states with higher minimum limits for coverage, such as New York or Florida. "It just makes them a rich target area," he says.

The ability to integrate disparate sets of data inevitably leads to more accurate predictive models. A model that produces a high degree of false positives will quickly lose adherents, says Ernie Feirer, VP and GM of LexisNexis Claims Solutions. "The more data you have, the better the performance of technology, the better acceptance of the technology within an organization," he says.

Another question with which all carriers must grapple is how sensitive to make their models - how many claims do they want to flag? If they set their threshold too low, there will be too many claims at which to look, to the chagrin of adjusters of investigators. Conversely, set the threshold too high, and fraudulent claims will undoubtedly get through. What's more, if adjusters and investigators start questioning the results they get from the technology, they will walk away.

"You are always adjusting your threshold," Rioux says. "You want to lower the number of claims your adjusters look at, but balance it against missed opportunities."

The Balance

One other balance carriers need to consider is how to employ analytics. One choice is use modeling to get an evermore granular view of an individual claim.

"You have the ability to create predictive analytic models based on a customer's unique data," Ellis says. "You can apply algorithms to a customer's unique claims data and find patterns in the data."

One consideration to this approach is that a carrier must be unerringly precise with the data used to populate the model. An honest claimant with a name similar to that of a known fraudster can produce a flawed model and an angry customer. "You have to be very good at matching the data to the right individual," Feirer warns.

Another option for insurers is to leverage analytics to get a holistic view that individual adjusters, no matter how intuitive or experienced, could not surmise on their own.

"In order to be effective, analytic models need to reach across claims," says Michael Costonis, executive director, insurance industry, North America, for Bermuda-based Accenture. "The old paradigm was using binary scoring to assess a claim to see if it was fraudulent. In the new world of analytics, the aim is to look across claims patterns."

In addition to these trends, experts see a larger acceptance of predictive analytics as a vital tool in combating fraud. "Large insurers have been doing this for a while," Harbage says. "Now, we're seeing a wider recognition among smaller insurers that they need to do something."

Celent's Light predicts that there will be a broader adoption of predicative models in the claims process over the next couple of years as some of the cultural issues around the use of analytics abate. "More people exposed to quantitative analysis are coming through the pipeline," he says.

Erie's Rioux says despite the advances in the use of predictive modeling, many things in the claims process will remain the same.

"We're not cutting the adjuster out of the process," he says, noting that since models are based on past experience, they are a good way to lock in knowledge from experienced adjusters. "You need somebody to look through it. Human intelligence is far superior."

New Database for Cargo Theft

Jersey City, N.J.-based ISO and the Des Plaines, Il.-based National Insurance Crime Bureau (NICB) are partnering to create a national information sharing system aimed at thwarting cargo crime.

Launching in early 2010, the core of the network is a new database called CargoNet. Intended for use by theft victims, their insurers and law enforcement, the endeavor will network existing databases and provide secure reporting and analytic functions.

"NICB has been making progress against cargo theft on many fronts," Joe Wehrle, president and CEO of NICB, said in a statement. "We have recovered stolen cargo, developed intelligence and dissolved organized groups behind the thefts. If CargoNet were in place today, I'm sure we'd be seeing a lot more recoveries, and we'd be making thieves think twice about stealing these loads."

Proponents of the system note that, currently, cargo thieves are able to exploit existing gaps in the nation's information-sharing framework, costing insurers and, ultimately, consumers.

"We are greatly encouraged by the strong support we are receiving from leading cargo insurers," said Vincent Cialdella, ISO senior VP. "This initiative would not be possible without it. We are also encouraged by discussions we have had with transportation companies, manufacturers and retailers, given the crucial role they play in this initiative."

This article can also be found at

Register or login for access to this item and much more

All Information Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access