Establishing an effective anti-fraud program requires a vigorous, hybrid approach to detecting fraudulent claims—one that encompasses business rules, anomaly detection, text mining, predictive modeling techniques, and social networking analysis—all accompanied by process and operating model designs that drive claim handling based on analytics-based outcomes.

The foundation for such a program is good data. Insurers clearly need clean, robust, and “trustable” data about the claim attributes themselves, but should also look to round out the attributes of fraudulent claims with as much ancillary data as possible. Claim-handling data will help expose processes within certain claim segments that lend themselves to fraudulent activity, while also exposing data about “insiders” who perpetrate fraud from within the claim-handling organization. Background information on the customer, such as billing and policy history, contact history, and historical claim frequency is essential, as well. External data, from sources such as the ISO, CarFax, credit bureaus, customer medical records, and provider information will likely prove valuable when added to the mix. Finally, data indicating founded fraud, e.g. fraud that has been identified and quantified in prior claim-handling activities, is critical when looking to develop predictive models to better identify fraud going forward.

Opening the Analytics Toolbox

Business rules and anomaly detection are typically the first lines of defense in fraud screening, testing each claim against algorithms that are designed to detect known types of fraud by identifying specific activity patterns. Given the large number of variables to examine, however, business rules are inherently inefficient if used in isolation, e.g. to generate a report that is passed on manually to a special investigation unit for additional review. Business rules often lack the multi-variate analytic power that comes from more advanced techniques. In addition, organized fraud rings soon learn the carrier's rules and red flags and take actions to avoid detection at this stage.

Effective fraud prevention programs employ several additional techniques to improve the overall efficiency of the process and reduce the incidence of false positives. Text mining, for instance, takes advantage of the large quantities of text-based information generated in the claim process. These documents might include e-mails, customer service calls, interviews with individuals pressing a claim, adjuster notes, and other related data. This unstructured information can be a valuable source of clues leading to indications of possible fraud.

Social networking or link analysis is another effective tool in detecting fraud-ring activity. Using social networking analysis, insurers can identify relationships between and among claims—including telephone numbers, vehicle identification numbers, and medical providers—and establish links indicating suspicious activity. Social networking analysis provides insurers with the ability to identify patterns and possible links to suspicious activity across large numbers of claims. The right software can map such activity across the entire base of customer claims, providing graphic representations of how such activities are connected.

While social networking analysis helps insurers avoid paying potentially fraudulent claims, it can also be used to screen new policies to identify historically problematic elements. As is the case with so many aspects of claim analysis, the best and most efficient way to avoid a loss is not to write the policy in the first place.

Proper segmentation of claims—using cluster analysis of internal and external benchmarking data to establish claim profiles, which are then input to a pattern analysis engine to identify appropriate assignments and behaviors throughout the life cycle of the claim—can make the claim-handling process more efficient and more effective. Analytics in claim handling cannot be a back-office function that generates results to a report that organizations then need to figure out how to implement. Carriers that imbed analytic outcomes within claim-handling practices to drive processes based on those outcomes will start to put distance between them and their non-analytic competitors.

Another significant benefit of segmentation is that intelligence from claim handling can be fed back to the underwriting operation, creating a better understanding of risk and leading to more accurate policy writing and pricing.

Measurements of Success

How do insurers know when the incorporation of analytics into anti-fraud efforts has been successful? The referral rate may remain the same, since better application of business rules and anomaly detection may generate a greater number of exceptions related to fraud while reducing the number of exceptions that are not. There should, therefore, be a reduction in the rate of false positives. This will vary greatly carrier by carrier, driven by the sophistication of a company's current fraud-detection techniques.

The unit cost per referral to the special investigations unit should go down, as well. Data made available as an output from analytic techniques in play should give fraud investigation personnel much more data upon which to confidently act, thus driving down the time to piece together a case to prevent payment or pursue recovery.

The company, overall, should experience a reduction in the average fraud analysis effort per claim handled and a reduction in total indemnity payments made, along with a reduction in unallocated loss adjustment expenses incurred to investigate fraud.

Assembling the proper analytics toolbox to enhance fraud prevention and detection activities is not a simple task. The insurer must assess and, if necessary, improve the quality and structuring of data; establish proper thresholds to trigger notifications; and properly test and tune the analytics package. Analytical models must be integrated into the underwriting and claim-processing cycle. Lastly, the requisite analytic skill-set is required within the carrier's workforce to maintain and refine these analytic toolboxes over time (oftentimes a skill-set that is net-new to the claim organization).

The return on the investment in a properly implemented anti-fraud analytics initiative is quite high. A carrier with a billion dollars in claims paid can expect significant real-dollar impacts, for instance, from a fraction of a percent of reduced indemnity payouts due to fraudulent claims. More difficult to measure—but equally important—is the improvement in customer satisfaction and retention as the insurer concentrates its skills on identifying and investigating claims that are truly fraudulent, while using the cost savings from lower payments to improve the overall customer experience.

Savvy carriers will likely determine a way to market their investments in fraud detection as a way to ultimately save their non-fraudulent insureds money through lower premiums thanks to their dedicated pursuit of fraud.