A rather typical business story became a head-scratcher of aWorkers' Comp claim — and an illustration of the power ofpredictive modeling — when a seemingly “normal” claim was flaggedfor potential fraud.

A person traveling for business stayed in a hotel, as manypeople do when they travel, but with a twist. He filed a Workers'Compensation claim for supposedly suffering an injury after fallingoff a chair in his room. According to the claimant, he could nolonger work, but he didn't expect that a predictive modeling toolwould flag his case based on his occupation and other factors,bringing it to the attention of the insurer's SpecialInvestigations Unit (SIU).

After a little digging, an SIU investigator found that theclaimant had not reported the incident to the hotel and did nothave any doctor bills. And when the investigator interviewed theclaimant at his home, he spotted a broken chair on the outsidepatio. The claimant noticed that the investigator saw the chair,and immediately told the investigator that he wanted to withdrawthe claim. Although there was no apparent reason for the claimant'sreaction, the investigator suspected that the claimant may haveinjured himself falling off this chair at home, not a chair at thehotel.

This real example demonstrates how predictive modeling empowersa claims staff to be better, faster and smarter when it comes tofraud detection, particularly for Workers' Compensation and autobodily injury claims. Although it can be a balancing act toidentify potentially fraudulent claims and still provide superiorcustomer service for justified claims, a handful of factors canensure success. They include the increasing speed with whichpredictive modeling allows claims teams to act, the ability toidentify potentially fraudulent cases throughout the claimsprocess, maintaining and enabling the human element in claimsadjusting, and strengthening an insurer's anti-fraudreputation.

Speeding it up from start to finish

Before predictive modeling at Chubb, it could take up to 180days to spot potentially fraudulent Workers' Compensation claimsand assign them to the SIU. Now that number is down to six days.For auto bodily injury claims, four months was the average timerequired for an SIU referral; today, it is only four days.

The greater efficiency has resulted in significant savings.Predictive modeling has led to a significant increase in acceptedreferrals to the SIU for both Workers' Compensation and auto bodilyinjury. As a result, the number of investigation days hasdecreased, and the company has achieved significant costsavings.

How is this possible? Quite simply, speed is crucial forinvestigating a claim. Evidence is still fresh. Claimants' andwitnesses' memories are still vivid. Additionally, reacting to aclaim quickly can serve as a deterrent to fraudsters who are “onthe fence,” as it did with the claimant with the broken chair. Andcompanies know that the longer it takes to settle a liabilityclaim, the more likely the claim will be litigated.

Related: 3 keys to a successful Workers' Compensation fraudinvestigation

Speed is also critical to delivering excellent customer service.By quickly identifying potentially fraudulent claims, the companyis able to be more efficient and spend more time on customers whohave valid claims.

(Photo: Thinkstock)

Continuous vigilance

Although predictive modeling can identify a potentiallyfraudulent claim faster than many cases in the past, successfulclaims adjusting operations continue to monitor claims as theyprogress and flag them later if certain warning signs appear.

Predictive modeling begins with the first notice of loss andthen continues to monitor for certain trigger points and specificactions during a claim's lifecycle, such as the number of priorinjury claims submitted by a claimant and the amount of time thatan allegedly injured claimant is out of work. The model flagsclaims based on patterns that have historically proven fraudulent —patterns that a human adjuster often may not detect.

In addition to detecting fraud, predictive modeling can indicatewhether a legitimate claim has the propensity to develop adversely.It can be used to evaluate the likelihood that a claim will resultin litigation. It may also provide the ability to identify Workers'Compensation claims with a greater likelihood of surgery. Suchtools allow adjusters to develop case strategies at first noticeand gain control over the claim as it progresses.

Smarter allocation of resources

Despite the wealth of benefits from predictive modeling, themachines are not taking over. Adjusters are still the criticalelement in claims management and fraud detection.

The “tool” delivers information that gets adjusters thinkingabout fraud and looking for clues. Modeling does not prove that aclaim is fraudulent; it tells adjusters to look closer. After that,they must still conduct their claim investigation. For instance,when predictive modeling registers a claim as highly suspicious, anotice is forwarded to the SIU, where investigators triage theclaim and determine whether it should be accepted. Once the SIUaccepts the claim, the SIU and file handler plan the investigationtogether, with the claims handler maintaining ownership over theadjustment process.

When done well, predictive analytics does more than just inform;it smooths out the efforts of hundreds of adjusters across multipleclaims offices around the country. In any given group of adjusters,some will be more focused on detecting fraud than others.Predictive modeling builds a ground-floor foundation of awarenessso that these differences are minimized.

Human beings also have a say in how the modeling tools aredeveloped, integrated and used. The Chubb Claim Analytics teamseeks input from claims professionals and gathers their feedback toanalyze business benefits and to discover ways to improve existingsolutions and create new ones.

Better reputation management

Most important, predictive modeling can help create a deterrentto fraud. Police and other authorities know that fraud perpetratorstalk to each other. They share information on insurance companies,down to particular claims offices and individual adjusters who areeasy “marks.” Predictive modeling can help insurers send themessage that their company's claims organization is not an easytarget for fraud.

At present, predictive modeling is showing clear signs ofsuccess — increased speed, smarter human involvement, financialsavings, better reputation management and continued vigilance. Asthe technology is further developed, it will become betterintegrated into the claims process. While many claims today arestill handled manually and in the order they are received,predictive modeling will help insurers move even more towardsegmentation. The tools will enable SIUs to identify those claimsthat are best suited for experienced staff at a given office, forinstance, or those severe enough to require immediate review,contact and intervention.

Moving forward, greater success will also involve the ability toinput additional unstructured data into the modeling systems. Postson social media, such as those linking a claimant to knownfraudsters, are examples of information the company hopes to streaminto the systems. Though the use of such data can be sensitivebecause it may contain personally identifiable information, it maybe possible for firms to navigate safely around the potentialprivacy issues. In the near future, systems will be able to trackand match social media with claims data.

It will be a new age of enhanced claims management and customerservice. But it still will be a world in which it will be importantfor an adjuster to spot that broken chair.

Related: Meet the 2015 Insurance Fraud Hall of Shameinductees

We're on Facebook, are you?