Insurers appear to be waging a losing battle in their war againstfraud. Consider this: Property/casualty insurance fraud cost carriers $21billion in 1998, according to the Insurance Information Institute. Andexperts peg the annual cost of life and disability fraud at $5 billion, andhealth care fraud at $95 billion.Why is fraud so rampant? For starters, the sheer volume of claims-anestimated 116 million property & casualty claims were submitted in 1998-isoverwhelming carriers' claims adjusters and trained special investigativeunit staffers. Also, the complexity of fraud is increasing as professionalfraud "rings" continually devise new methods of bilking carriers.
Technology certainly has helped insurers improve the speed and accuracyof identifying fraud. The development of massive, Web-accessible databasesthat enable claims investigators to search for clues, coupled with softwarethat helps investigators piece together small bits of information, havevastly improved the detection of fraudulent claims. But in most cases,insurers are identifying fraud long after the check has been cut to pay theclaim.
"Fraud detection is a complex issue for carriers," says Judy Johnson,vice president, insurance information strategies, for Meta Group Inc., aStamford, Conn.-based research and consulting firm. "Fraud detection in theindustry is an application that's used by a carrier's special investigativeunit that sits on desktop. To be successful, it needs to become the natureof how a company operates and part of the company's IT infrastructure."
Although more carriers are using visualization software and data miningtools that use artificial intelligence to piece together bits ofinformation that to the trained eye may seem unrelated, experts say thatcarriers aren't doing enough to fight fraud. The problem, they say, is alack of proper tools, and, in some cases, the organizational commitment towage an all-out war against fraud.
"For many insurers, fraud is just considered a cost of doing business,"Johnson says. "It's easier for them to just pay claims withoutinvestigating them for fraud, because insurers don't really know how muchfraud is being committed."
Manual process
Fraud detection traditionally has been a labor-intensive process ofmanually piecing together bits of information from claims and data sources.In 1998, an estimated 150,000 claims adjusters handled more than 116million new claims, an average of 750 claims per individual per year,according to a recent report by Deutsche Bank Alex. Brown, San Francisco.
Of that total, only 3,218 cases were presented to prosecutors by statefraud bureaus, according to the Washington, D.C.-based Coalition AgainstInsurance Fraud, a not-for-profit organization made up of consumers,insurers and government agencies.
Claims professionals don't have the proper tools to quickly identifyfraud and are too overwhelmed by their caseloads, says Gary Craft, managingdirector of e-finance research for Deutsche Bank Alex. Brown.
"Claims professionals are dealing with huge volumes of claims, andinsurers need to invest in workflow systems," Craft says. "Carriers areunder pressure to pay their claims promptly. In many cases, carriersblindly pay their claims, and then they retrospectively look for fraudafter the claim has been paid."
Searching for clues
Insurance companies understand that industry collaboration is essentialto effectively fight fraud. The creation of ISO's ClaimSearch database,some industry observers say, is a collaborative effort which demonstrateshow technology is being deployed to combat fraud.
The comprehensive database has 147 million records containing bodilyinjury, property loss and auto claims information. More than 4,000insurers, self-insured entities and third-party administrators report morethan 150,000 claims per day to the database.
When insurance companies and other participants submit a claim to thedatabase, the system automatically enters the information and searchesrecords stored in the database to match a name, address, Social Securitynumber, vehicle identification number or other claims information.
"Forty percent of the time we are able to come up with a match, andthat information is returned to the insurance company that sent us theoriginal claim," says Richard Boehning, ISO's senior vice president. "Justbecause there's a match doesn't automatically indicate there's fraudinvolved. There could be some crossover if two companies file the samereport."
One benefit of the ClaimSearch system is that it can match informationthat's reported for different lines of business. For example, if apolicyholder files a bodily injury claim with one carrier following an autoaccident, and later files a worker' compensation claim with a differentinsurance company to collect on disability coverage, the ClaimSearch systemcan match that information.
"We don't tell the carriers that we believe there's fraud involved, welet them make the decision," Boehning says.
If a carrier needs more information once a match has been made, theClaimSearch system enables claims adjusters and investigationsprofessionals to submit queries to search specific fields of information.
Such searches may uncover previously hidden connections between names,addresses and other identifying information.
"An SIU professional doesn't want to be constrained by the system, sowe enable them to configure how they would like the search to be done,"Boehning says. "For instance, someone might want to do a search for all'John Smiths' in every state. A lot of information is gathered, and some ofit is meaningless, but the system attaches a score to the information basedon its relevancy."
Last November, ISO introduced a new service, called ISOnet, thatenables ClaimSearch users to submit insurance claims via the Internet. InJuly, ISO finalized work on a universal format for standardizing howinsurers report information to the ClaimSearch system.
The new format enables the system to issue a single matching report forall types of claims.
"ClaimSearch is an invaluable tool," says Kent Davis, corporate claimsmanager in the fraud management unit of St. Paul Cos., St. Paul, Minn. "Youneed that ability to search your internal systems, as well as othercarriers' claims, to help identify fraud. It's a great tool, but it's onlyas good as the data that insurance companies contribute."
Analytical software
Most large carriers have their own Special Investigative Units, oftenstaffed by former law enforcement professionals, that are responsible forclosely examining suspicious claims.
The process of identifying fraud is often tedious work, compounded bythe complexity of professional fraud rings and, until recently, the lack ofsoftware tools that can help uncover clues that aren't always apparent tothe trained eye.
As recent as 1999, St. Paul had 72 investigation professionals workingin its SIU. "We were very involved in investigation and link analysis thatwas done manually," says Paula Schlotfeldt, vice president of St. Paul'sClaim Shared Services. "We elected in late 1999 to find a less-expensiveoption."
This year, St. Paul began testing software developed by Columbus, Ohio-based Alta Analytics Inc. that enables investigators to link seeminglyunrelated claims on the basis of such information as names, addresses,phone numbers, social security numbers and other data. St. Paul took therecords it had gathered for a fraud ring investigation that the insurer wasprosecuting and fed the information to Alta Analytics' NetMap for Claimssoftware.
"It took us more than two months to manually pull together 23 relatedfiles that we used to prosecute this fraud ring," Schlotfeldt explains."With NetMap for Claims, that process took just 15 minutes to identifythose linkages, and just over two hours to identify 51 files that werelinked to the case."
NetMap for claims combines link analysis, visualization software anddata marts to help carriers identify fraud and plot the relationshipsbetween suspects.
The system uses various colors to illustrate how seemingly unrelatedinformation and individuals are linked, indicating patterns of fraud thatare difficult to uncover with manual research.
St. Paul began using the product in July to investigate potential fraudfor its commercial lines. Schlotfeldt declined to say how much the companyinvested in the fraud detection system, but she says St. Paul expects a $3return on every dollar of its investment.
"That's a very conservative estimate," she adds.
ASP option
Currently, 10 large insurance companies are using the NetMap for Claimsproduct; the annual license fee, which ranges from $200,000 to $400,000, istoo costly for most small and midsize carriers, company executives admit.
That's why Alta Analytics recently introduced an application serviceprovider model, which starts at $10,000 a year, based on a carrier's annualdirect-written premium and the number of people that will be using theproduct.
Fraud detection software is just now gaining acceptance in theinsurance industry, says Herb Jones, president of Alta Analytics.
"Three years ago, insurance company executives did not understand howtechnology could help identify patterns of fraud," he explains. "With thenew technology on the market, there's a growing realization among carriersthat fraud is a significant issue that can be addressed with the morecomplete systems on the market."
Quantum leap
The development of fraud detection software which combines suchtechnology as neural networks, statistical scoring and link analysis,insurance executives say, has greatly accelerated the process ofidentifying suspicious claims.
"Going from a manual process to a computerized process is a quantumleap," says Thomas Mulvey, national director of Prudential Property &Casualty Insurance Co.'s special investigation unit.
State regulations restrict the amount of time that insurance companieshave to process a claim and send the check to a policyholder. "You have tosort through a lot of information in a short amount of time, and if you dothat manually, you can't get a lot done," he says.
Prudential's staff of fraud investigators use ISO's all-claimsdatabase, in conjunction with the NetMap for Claims software, to uncoversuspicious activity.
"In the past, the volume of data would hide the associations of fraudring activity," Mulvey explains. "In addition, fraud has become moresophisticated. The power of ISO's ClaimSearch and NetMap for Claims is thatcombined, they are able to find bits of information and tie them togetherthat may have gone unnoticed when done manually."
Prudential also uses the technology to investigate individual fraudcases. For example, if a policyholder files a claim for a stolen ring,Mulvey explains, the system can determine whether a previous claim had beensubmitted by a different policyholder.
Technology not only has improved how quickly fraud investigators canpiece together disparate information to build a case for prosecutors, ithas enabled insurers to investigate suspicious claims before they've beenpaid.
"Ten years ago, large-scale investigations were often conducted afterall of the claims had been paid," Mulvey says. "Typically some informationwould turn up, and then all of the related claims files were pulled up, andthen we would gather testimony and other evidence. There has been atransformation in the speed to process all that information forinvestigations-we can quickly identify all of the cases involved, and it'snot all done retrospectively."
Eliminating guesswork
The incorporation of neural network technology has taken frauddetection software to a new level, experts say. Neural networks are used tocompare new data against stored information to identify patterns of fraudthat are indiscernable to the trained eye.
The real value that neural networks bring to fraud detection, industryobservers say, is that the more information that is fed into a neuralnetwork, the "smarter" it becomes (see "Smart Technology That ReplicatesHuman Reasoning," page 55).
"Neural network technology was first developed for the CentralIntelligence Agency during the 1970s, and now it's being incorporated inproducts to fight financial fraud," says Meta Group's Johnson. "Thetechnology has been successful in combating credit card fraud, andcompanies are now combining it with the best elements of fraud detectiontechnology."
For example, HNC Insurance Solutions Inc. has developed a fraud andabuse detection system that profiles ongoing activities with workers'compensation claims. The system, called VeriComp Claimant, uses neuralnetwork technology that reviews each workers' compensation claim when it'ssubmitted, then assigns a score ranging from 0 to 1,000 with the higherscore indicating greater probability that some type of fraud is takingplace.
The system also provides an explanation about what aspects of the claimare most suspicious.
"The product takes a set of claims with known outcomes and identifieslegitimate claims and fraudulent claims," explains Sean Downs, president ofIrvine, Calif.-based HNC Insurance Solutions. "We then feed thatinformation into a neural network which takes the information and theprocess begins to recognize subtle patterns in the data and identifiescommon elements for good claims versus fraudulent claims."
This predictive modeling technique is then used to assign a score tonew claims. Once a new claim is entered into the system, the data iscompared against the database of legitimate and fraudulent claims. And,when carriers receive additional information for an existing claim, thesystem automatically updates the fraud-risk score. That way, if bogusinformation is added to a legitimate claim, the system can alert claimsprofessionals to examine the potential for fraud.
In 1997, Workers' Compensation Fund of Utah gave HNC permission to useits nearly 1,000 claims files to help HNC identify elements that canpredict fraud. The Salt Lake City-based mutual insurance company, which has28,000 policyholders and $100 million in annual premiums, was interested indeveloping technology to assist its staff of claims adjusters.
"We noticed that most of our fraud cases came from just a handful ofadjusters, so what we wanted was to be more consistent and not to rely onthe expertise of the individual to uncover fraud," says Bob Short, seniorvice president of the Workers' Compensation Fund of Utah. "We needed a goodmanagement tool that could teach adjusters how to identify fraud."
Improved recovery rate
The organization reviews between 400 and 450 claims cases every year.Prior to using Vericomp Claimant, Short says the company's fraud-recoveryrate averaged between $3 million to $4 million per year. In the past twoyears, the company's fraud-recovery rate jumped to between $6 million and$7 million.
"We run all of our claims at the end of every week, and those claimsthat score above 500 are kicked backed to the adjusters," Short says. "Thesystem red flags the claims and tells the adjusters why a claim needs to befurther investigated."
The process not only has helped the organization better identifysuspicious claims, it has improved workflow.
"The technology makes all of us more accountable for our work," Shortsays. "It makes the adjusters accountable because a report goes to theirmanagers and if they haven't acted on a claim, the managers know about it.Now, the bad adjusters are getting better because they're learning toidentify elements of fraud."
Is enough being done?
Fraud detection technology has the potential to turn claimsinvestigators into super sleuths. Rather than relying on the instincts of abloodhound, these professionals now have the power of massive databases andartificial intelligence at their fingertips.
But some industry observers believe that insurance companies have tocommit more resources in their battle to overcome fraud.
"You can't apply tools such as artificial intelligence until you havegood, clean data and good flows of data, and the insurance industry as awhole doesn't have that yet," says Deutsche Banc Alex. Brown's Craft."Insurance companies need to invest in workflow systems, but the fact isWall Street is not going to give the large insurance companies a lot ofcredit if they invest in their claims systems. Insurance companies arerewarded on the basis of increasing their market share."
Some insurers have been making sizable investments in updating theirclaims processing systems, Meta Group's Johnson says, but their emphasismostly has been on expediting the claims process, not on fighting fraud.
"Most insurers talk about how they're re-engineering their claimsprocess, and in a commodity market, customer service is the way todifferentiate yourself from competitors," she says. "They're looking tohandle claims more quickly, so when they pass a claim off to their specialinvestigation units, it slows down the process." Instead of retroactivelysearching for fraud, Johnson suggests that carriers should begin frauddetection when a policyholder files a first notice of loss.
"The claims process needs to have an automatic trigger so that when aclaim is entered manually or electronically, the first step is for thatclaim to be sent through fraud detection software and match thatinformation against a claims database. Fraud detection should not beconsidered a separate application that sits in the special investigate unitat the end of the process. At that point, mostly likely the claim hasalready been paid."
When St. Paul Cos. went through the process of identifying frauddetection technology to serve its needs, the company wasn't satisfied withany packaged solution.
"The problem with most scoring systems is that they're attempting to bea one-size-fits-all solution, and that just doesn't work well," says St.Paul's Davis. "Fraud indicators are very different from the type of lossand for the policy type, so you need a solution that is very specific.
"For example, the fraud indicators for automobiles and home theft vary,but all of the systems out there take data from fixed fields, and mostclaim systems don't include information that's needed for all types offraud indicators."
What Davis says would be beneficial for insurers is a fraud detectionsystem that enables adjusters to query a claims database for specificinformation, and for the system to give adjusters advice on how to proceedwith a claim.
St. Paul is discussing its plans with a number of technology suppliers,including Alta Analytics, to get a patent on such a system. Davis claimsthe company's suggestions have been well received.
"More companies realize that what's needed is more customization,rather than a broad-band approach," he says. "We need a proactive solutionthat can identify what type of claim it is and to then apply the fraudindicators that are specific to that type of loss. Without that, claimsadjusters are dealing with too many false positives, and that becomesfrustrating and then they don't want to use the system."
