Consider voice stress analysis, a technology that has been around for decades but recently has received more interest in antifraud applications, primarily among overseas carriers. Combined with savvy interviewing techniques, voice stress analysis serves as a phone-based lie-detector test that's predicated on the fact most people are unable to lie about important matters, such as an insurance claim, without the stress of that lie showing up in their vocal patterns.

Granted, voice analysis is an investigative post-claim technology, but UK-based home and auto insurer esure has found it to have preventive benefits, as well. For the past three years, esure has been using DigiLog's Voice Risk Analysis (VRA) technology in its automobile claims processing.

“The system doesn't actually catch fraudsters,” notes Adrian Webb, head of corporate communications at esure. “What it does is cause potentially fraudulent claims to evaporate,” once fraudsters know the insurer is wise to their game.

At esure, all auto claims aren't put through the system, Webb states, rather only a subset with characteristics previously identified as high risk, such as vehicles stolen and not recovered (where the company suspects the insured simply may be hiding the vehicle). “It is not a system that will work for every single claim because of the cost in using it,” Webb says, although esure would not reveal that cost.

On suspicious claims, adjusters notify claimants they will be recorded by the VRA technology. Adjusters begin the interview by asking factual information–name, address, and so on–to establish a claimant's baseline voice stress pattern. As the interview progresses, the system analyzes what DigiLog calls “emotional data parameters,” combining them into identifications of points in the conversation that indicate truth, uncertainty, or deception. The claims adjusters monitor the system in real time, drilling down into areas of possible deception by using cognitive interviewing techniques.

The result of a voice analysis never is used by esure as the basis for claim denial, Webb emphasizes, but rather only as a springboard for additional investigation when a suspect claim doesn't disappear. Also, by being public about its use of voice stress analysis and its overall antifraud program, Webb believes the insurer has established a reputation in the industry that has a preventive benefit. “People interested in making a fraudulent claim are like the type interested in robbing a house. If they see two homes with burglar alarms and one without, they'll go for the one without,” he explains.

While esure asserts the system has had an impact, “it's difficult to value the claims we don't pay because they are withdrawn,” Webb says, adding the system also has an important ancillary benefit: helping esure fast track valid claims that initially raised suspicion.

“If someone goes through the interview and shows no stress, there is virtually no chance [that individual is] lying, and we can cut the check right there,” he says. “There is a great efficiency savings [to the system] as a truth validation.”

Fortis Corporate Insurance, a marine insurer based in the Netherlands, is using a modern incarnation of the traditional salvage auction to deter fraud. When claims occur, Fortis typically finds itself with damaged, insured cargo on its hands. When that cargo is steel, Fortis today uses SteelSalvor, an online auction service, to dispose of it.

Like voice stress analysis, the auction system is used post-claim, so its fraud prevention benefits aren't immediately apparent. However, Fortis has found SteelSalvor useful to prevent fraud that might be perpetrated by the third-party field damage surveyors the insurer uses as well as to reduce overall leakage in the claims process.

“We had no control in the possible contacts between the surveyors and potential buyers,” indicates Guy Van Spilbeeck, senior claims adjuster at Fortis. “Most of the surveyors are trustworthy, but you never can know whether [any surveyors] might have made some deal with a buyer that is to their benefit.”

Whereas previously surveyors would work directly with a small pool of buyers they knew, today Fortis requires surveyors to work with SteelSalvor to run online auction listings of salvage steel the service markets to thousands of interested buyers worldwide. Since bids are open, the salvage process now is transparent to Fortis.

“The important part is the preventive aspect. There is no opportunity for collusion between the surveyor and the buyer,” Van Spilbeeck says. Fortis pays SteelSalvor a five percent fee based on the sale price, which he reports has been running 15 percent to 20 percent higher based on greater buyer competition since the insurer began using the system.

From Prediction to Prevention

Among U.S. insurers, the antifraud technology of most interest in recent years has been predictive analytics, which tends to be the third stage of including automation in the process to detect fraud in individual claims. Stage one typically involves providing claims adjusters with fraud flag checklists in the workflow system that will route electronically suspicious claims to SIU. It also involves using data-matching systems that compare claim data against both internal and external negative file databases of known fraudsters.

Stage two has been to automate the checklist process by using a rules engine to match claim data against a set of known fraud parameters and to assign scores that help SIU focus on the highest-probability fraud cases. In addition, claims are reprocessed through the engine any time changes occur to the file. In stage three, an insurer augments adjusters' intuition by deploying a predictive modeling engine that continually refines its understanding of what fraudulent claim patterns look like based on internal claim data as well as third-party fraud databases.

Nationwide does use predictive modeling as do other early adopters such as Erie Insurance Group and MetLife Auto & Home, both of which have been featured in past issues of Tech Decisions. “Typically, [only] the larger insurance companies are beginning to make use of predictive analytics to help shrink the time line for identifying and investigating suspected scams,” says James Quiggle, spokesperson for the Coalition Against Insurance Fraud. “Predictive modeling really still is a cutting-edge tool for insurance companies.”

While most insurers grapple with whether–and how–to use predictive modeling, those already applying it are looking for more ways to leverage the technology. “With our current application of predictive modeling and using business rules against our data mining of claims, we've gone from anecdotal-type fraud detection based on adjuster intuition to a technology-enabled approach to predict where fraud has occurred,” says Kearns, who declined to name the modeling technology in place at Nationwide. “But we need to apply predictive modeling even earlier in the process to predict where fraud could occur. That's the next level.”

Getting to that level won't be easy. It will require access to data that's not in the claims system because no claim has yet occurred. “We need to assess what changes need to be made on the front end to [do that analysis],” Kearns points out. “What are the things that need to be captured and measured, and do we already capture [that information] in other systems?”

This might mean crossing technical boundaries into underwriting and policy administration systems to evaluate policy changes that indicate fraud might occur. And there are organizational obstacles. For instance, will underwriting be willing or able to take action, based on a suspicion raised by SIU, on a policy on which no claim has occurred?

“You can't just push a button and delete a policy,” Kearns says, “but you might find actionable things in the detection and investigation process where you can minimize your risk or exposure.”

Beyond predictive analytics, data visualization tools have been on the radar of insurers' SIUs. Rather than detecting the one-time fraud perpetrator based on individual claim data matching, visualization tools help insurers connect the dots among claims to find fraud rings, thereby stopping future claims from those rings from occurring.

“Taking data out of its normal format of columns and reports and putting it into a graphical representation adds a lot of power to the investigative process,” says David J. Rioux, vice president and manager of Erie Insurance Group's Corporate Security and Investigative Services.

Erie simultaneously deployed Fraud-Focus predictive modeling from Magnify (now part of ChoicePoint), ISO's NetMap data-mining software, and ISO's ViewLink manager data visualization tool in 2003. “Another advantage those data-mining and visualization tools have brought us is they can take data from disparate sources. Sometimes we first need to extract it into a common repository but most often not,” Rioux says.

Having created the ability to data mine against the structured data of its claims databases, Erie now is looking to text mine the unstructured data of claim file documentation. “The richest information is kept in the claim log, where it is not captured in data fields. We already do text mining around the loss notice, but we'd like to text mine against the log notes, as well,” Rioux adds.

Still Slow Movement on The Tech Front

Every insurer fights claims fraud, and individual insurers as well as industry associations have made a concerted effort in recent years to educate the public on the true, shared costs of fraud as part of a larger preventive effort. “Insurers have no choice but to take a robust public stance against both hard and soft fraud,” says Frank Zizzamia, director at Deloitte consulting. “If there's any perception in the external world they're not, they will become an instant target.”

Yet insurers that use automation in the fight remain the minority. Annemarie Earley, managing vice president for Gartner's insurance advisory service, says in an unpublished survey the research firm did in 2005 only 13 percent of insurers reported they were assessing or investing in fraud detection and analysis tools. “The adoption isn't what I would expect it to be,” she says.

“[Insurers] like to talk about [antifraud technology]; they get interested and excited about it, but when it comes down to doing something, it stalls,” observes Lewis Rogers, director within the claims and legal product development group at CSC, which markets fraud-detection solutions.

The main reason for slow adoption, Earley asserts, is the timing simply hasn't been right for insurers. With their claims IT spending over the past several years tied up in core claims administration system projects, fraud technology has been a low priority.

Additionally, there are several difficulties in assessing the cost-benefit of claims fraud technology, the first of which is how much fraud insurers actually incur. A frequently cited statistic is 10 percent of claim dollars insurers pay are fraudulent. However, there's no way to prove that empirically. “[The 10 percent figure] is more 'common wisdom' than statistical calculation,” says Frank Scafidi, director of public affairs for the National Insurance Crime Bureau.

“We don't know for sure how much fraud is being caught and how much isn't because what we need is a reliable way of precisely measuring the volume of scams and attempted scams,” Quiggle explains. “How do you measure it? Is it measurable with a court conviction? Is it when an insurance company refers a claim? Is it when a company knows there's a scam, but [the fraud] isn't large enough to pursue?”

The second difficulty is assigning a monetary value to the impact of antifraud technology. “Many companies I talk to struggle with developing a return on investment because they don't have the historical data needed to benchmark their post-implementation results,” says Rioux. In contrast, he reports Erie historically had captured data regarding the result of SIU-referred claims. Consequently, Rioux's business case for the technologies Erie now has deployed had projected an additional loss impact (dollars not paid on fraudulent claims) of between $2 million to $4 million. These savings came from claims referred by the new systems that would not have been flagged manually by adjusters as well as from scoring higher-probability fraud cases that could be given higher priority by SIU.

Erie actually beat Rioux's projection, recording additional loss impacts of $5 million in 2004 and $4.4 million in 2005. Although Erie won't divulge costs related to the system or to the concurrent creation of a new Case Intelligence Unit that works with the system in support of field investigators, Rioux says the project has had positive net returns each year.

MetLife Auto & Home is more circumspect about any monetary benefits. The insurer uses CSC's Fraud Evaluator, which it co-developed with the vendor and deployed in 2003, incorporating predictive modeling, data mining, and business rules components. “The product has increased productivity and the referral rate [to SIU]. If you continue down that path of what that translates into, it does create a monetary value,” affirms John Sargent, SIU director at MetLife Auto & Home. “We have quantified [the cost-benefit], but we feel uncomfortable quoting it.”

Yet another issue is companies struggle with how to measure ultimate success–the prevention of fraudulent claims. “How do you value deterrence, and do you wrap that into the return?” Kearns muses.

As a result of these difficulties, advocates of antifraud technology often express its impact in nonmonetary terms: the increase in total referrals to SIU, the increased percentage of referrals coming from automated vs. adjuster-initiated referrals, and/or the “quality” of referrals based on successful detection of fraud. “ROI comes in more forms than just dollars,” Kearns says.

The difficulty in calculating a cost-benefit has led some carriers to wait on deploying automated detection systems. Consider EMC Insurance Companies. In its project to update its claims system, which began in 2002 and ended in 2005, the insurer added an electronic workflow between claims and SIU but chose not to put in an automated referral system based on any type of predictive modeling.

“There's no easy way to look at our existing data and say, 'We definitely could have saved this amount of money,'” with an automated referral system, says Rich Schulz, senior vice president of claims at EMC. “So, to spend a massive amount of money, I just don't have the data to show the expenditure would pay off. There are enough simple things–training adjusters, making sure SIU is handling [cases] accurately, accessing different databases for matching–that we can do first.”

Finally, insurers still struggle with data problems that inhibit the deployment of antifraud technology. “Data is our biggest challenge,” Kearns contends. “It's not having data; it's the condition and integrity of the data. It could be anything from the design of the system to data-entry errors. You have to take the time to know the data sources, determine what is both accurate and significant, and build from there.”

Given the core claims system modernization initiatives many insurers have undertaken, Earley points out the industry has been addressing data issues, and adoption of claims fraud technology may be poised for growth. “Not everybody is jumping on [fraud] as the next area to conquer, but it's slowly migrating to the forefront,” she says. “The reason is timing. The adoption of new claims management technology had to come first.”

Ultimately, however, technology never will replace the human element of both detection and prevention of claims fraud. “At the end of the day, it's back to the street smarts and back to basics: interviewing people, collecting documentation, and gathering evidence,” concludes Rioux. “No technology will prove fraud; only a good investigation proves fraud.”