Despite years of ongoing efforts to identify and curb insurancefraud, it remains a significant problem. Conservative estimatesfrom the Insurance Information Institute (I.I.I.) place the figurefor annual P&C payouts on fraudulent or padded claims at morethan $30 billion. A further disturbing statistic suggests that 10percent of losses and loss adjustment expenses (LAE) are associatedwith fraud and abuse. Thus, a carrier with $100 million in directwritten premium (DWP) and running at a 70-percent combined ratio islikely leaking more than $7 million annually because of fraudulentclaim activities.

As the statistics indicate, fraud continues to be a profitableenterprise, with fraudsters often operating across vertical lineslike P&C insurance, healthcare, and mortgages. Economic factors play an important role aswell, with the down economy driving new entrants to practicefraud—both opportunistic (such as padding, waste, and abuse) aswell as organized fraud schemes. Perhaps most noteworthy are theexperienced fraudsters who continually revamp and innovate,developing entirely new schemes or improving old ones to avoiddetection.

Developing Corporate Culture

Fighting fraud effectively requires industry-wide collaboration.With the fine-tuned tactics that hardcore fraudsters employ totarget their victims, no carrier can afford to lack a strong,unified strategy against fraud. Some collaboration already exists,as evidenced by an industry-wide claims database (ISOClaimSearch®), numerous state fraud bureaus, and the NationalInsurance Crime Bureau (NICB). However, individual insurancecarriers span the spectrum in terms of their capabilities to detectsuspicious claims, their investment in investigative resources, and their corporate appetite to deny bogus claimsand prosecute discovered fraud.

To enable more effective industry-wide collaboration, individualcarriers must develop the right kind of corporate culture, createpractical processes for handling suspicious claims, and implementsystematic fraud detection solutions. Corporate culture is arguablythe most important. Fighting fraud must be an enterprise-wideendeavor, with commitment beginning right at the top. Investigatingsuspicious claims cannot be a unilateral decision of the claimsdepartment because it also inevitably has implications for thelegal department (in cases of disputes and suits); business areas(lost policyholders from incorrectly referred claims); and publicrelations (overcoming negative publicity such as blogging fromdissatisfied claimants).

Organizational measurements must also be aligned to create theoptimal environment for fighting fraud. Effective anti-fraudtraining is critical for the success of any fraud detectionprogram, and much depends on the adjuster workflow for detectingand referring suspicious claims. The appropriate integration of process andtechnology is a must.

The third component of effective fraud-fighting is thedevelopment and use of solutions that can identify suspiciousclaims systematically. In the remainder of this article, let'sfocus on current and emerging analytical innovations that allow usto build more effective tools for systematic and semi-automateddetection of suspicious claims.

Systematic Detection

Building effective fraud-detection solutions requires a strongpartnership between business experts, such as investigators and adjusters, aswell as analytical staff, including predictive modelers and dataanalysts. Given the ever-changing nature of fraud, this process isnecessarily a virtuous cycle of continual development andimprovement.

Until recently, a simple special investigation unit (SIU)scorecard has been the tool for systematically detecting suspiciousclaims. Typically composed of expert-determined red flags such asindication that the accident was a set up, the claim was reported more than 20 days after loss, or there were unrelatedclaimants with same doctor, the scorecard has points, or weights,associated with each red flag. The adjusters determine which onesapply to the claim, total the associated points, and refer theclaim to the SIU if the total exceeds a preset threshold.

While such scorecards serve a valuable business purpose, theycan be further augmented with data-driven predictive modelingtechniques. When presented with a historical body of claims,supervised techniques, such as regression models, decision trees,Naïve Bayes, and neural networks, and in some instancesunsupervised techniques, such as clustering and outlier detectionmechanisms, can produce effective models to determine suspiciousclaims. The features, or predictors, used in such models are oftenthe expert-determined red flags, with their coefficients or weightsbeing determined by the historical data and its correlation to theSIU action.

Beyond Structured Data

While conventional predictive modeling techniques can yieldeffective fraud-detection models, they only work with structureddata such as code and value fields found in claims and policydatabases. Date and time of loss, policy limits, insured driverage, claimant vehicle details, nature of injuries, and treatmentduration are all examples of structured data. Often, valuableinformation that can identify suspicious claims may be buried inother data sources such as the adjuster's notes and claim networks.

Text mining is a recent application in extracting insights fromclaim notes. While claim notes pose special challenges such astypos, nonstandard abbreviations, grammatical issues, sentencefragments, and changing concepts over time, even simple and crudeconcept extraction can yield actionable insights. Text profiles canbe created to look for adjuster descriptions of interestingconcepts such as a low-impact accident¸ an accident near ahighway exit, and when the claimant waives EMR/ambulance. Anadjuster can use a structured field to denote the presence orabsence of this concept in the notes section of each claim. Suchstructured fields can then be fruitfully leveraged by predictivemodeling techniques.

Future Capabilities

Networks that relate claimants, attorneys, doctors, serviceproviders, and other entities on a claim contain valuableinformation not easily accessible in a structured form. Given aparticular person of interest, one can identify all claimsassociated with that person as well as all other entitiesassociated with those claims. This querying can be appliedrecursively to multiple degrees by then identifying claimsand other entities associated with the entities discovered in theprevious query. SIU personnel currently use interactive and deepening queriesof this kind as a key capability in the identification andinvestigation of fraud rings.

However, new capabilities are emerging to exploit claim networksin innovative ways.

Networks—and the participants therein—can be measured bycharacteristics such as density (number of entities or transactionsto which a particular entity is connected) and betweenness (theextent to which an entity is directly connected only to thoseentities that are not directly connected otherwise), among others.A high density might be expected of a medical provider but not fora claimant in auto accidents. Calculating norms for these characteristics bytype of entity can help flag abnormally deviant—and thereforeinteresting or suspicious—entities and behaviors. Tagging claimsand entities with their network characteristics results inadditional structured data that can be leveraged by the predictivemodeling techniques mentioned earlier.

Emerging technical capabilities, including speech recognition,image and video analysis, telematics, and novel data sources, suchas license plate readers, Facebook, and Twitter, will play a keyrole in creating structured data attributes that can significantlyenhance fraud solutions in the future.

Facing the Challenge

Claim fraud continues to be a big business—and a big challenge.In addition to corporate culture, training, and processimprovements, fighting fraud requires creating systematic and semi-automatedtools to identify suspicious claims. Inventive applications ofpredictive modeling techniques in fraud detection have been growingwith success. Text mining capabilities are expanding the scope ofthese solutions by discovering and integrating powerful insightsfrom claim text. Growing advances in social network mining areunearthing and characterizing claim networks and participants.Because of the ever-changing nature of fraud, new technologies, capabilities, and data will continually needto be harvested, and innovations pursued, to produce betterammunition in the fight against fraud.

Karthik Balakrishnan, Ph.D., is vice president of analyticsat ISO Innovative Analytics (IIA). IIA delivers advanced predictiveanalytics tools to the property/casualty insuranceindustry.

Thomas Mulvey is national director of claims and SIUsolutions at ISO, an operating unit of Verisk Analytics.