Fraud losses represent a systemic and increasingly convoluted risk affecting the profitability of every insurance company.
Traditional fraud detection methods and systems struggle with the complexity and speed of emerging organized fraud schemes. Insurers need a new approach to monitor customer behavior across multiple claims and lines of business. By combining multiple detection techniques, analytics and reporting, insurers can decrease fraud losses, reduce false positives and improve investigator efficiency.
Today’s fraud landscape
Today’s fraud trends and schemes are not those of 10 years ago. More advanced technology available to both insurance carriers and consumers has given rise to fraud that is much more inscrutable — schemes that are harder to detect with the human eye or by a single detection approach alone. Insurance claim fraud detection techniques are critical to combating the issue, but any one technique alone can cripple an investigation unit with false positives.
Related: The changing face of fraud
Given relatively flat growth and resource constraints, it is important that insurance companies consider multiple analytical detection techniques. Used in concert, they yield both a higher volume and a higher quality of claims requiring further investigation. They also tend to flag claims for further investigation earlier in the claim’s life cycle, increasing the likelihood of fraud mitigation, which positively impacts severity and ultimately profitability.
Multi-thread detection to fight fraudsters
Detection methodologies of the past were single-threaded. That left them quite vulnerable to circumvention. For example, fraudsters quickly understood and adjusted to the industry’s stance on claims filed within 30 days of the policy inception. Does that mean the business rule is not valid? Not at all. But it clearly illustrates that the industry cannot rely on any one rule alone for fraud detection.
To stay a step ahead of fraudsters, an insurance company’s detection techniques must be at least as sophisticated as the fraud being committed. Technology can give insurers the upper hand.
Today’s analytic technologies
Anomaly detection: Anomaly or outlier detection is very adept at identifying suspect service providers involved in claims. Creating peer groups and then applying anomaly detection to the groups within the data provides insurers great insight into their claims data. Coupled with other techniques, investigators can take a more proactive approach to combating provider-driven fraud.
Predictive modeling: Using past results to detect similar fraud in the future via a supervised model is effective, but given the speed at which fraud evolves and internal protocols are implemented, the approach is not without its pitfalls. Today’s fraudsters quickly learn and adapt to changes in insurers’ internal protocols. For example, they quickly learn the dollar threshold for claims that are desk settlements. Once a new fraud scheme is hatched, or a business protocol is changed and/or developed, a supervised model approach can fall behind. That is just another example of why insurers should not solely rely on a single detection technique.
Text mining: It is estimated that over 80 percent of critical insurance claim data is captured in an unstructured format. Even with sophisticated claim systems, critical data elements are entered manually — text typed into a field. Text mining can help uncover fraud in even large data sets using key words within claim notes and/or first notice of loss entries, derived after analyzing historical claims referred for further investigation. Imagine if this same claim also fails a few business rules and involves a service provider who is identified as an anomaly relative to his/her peer group. Chances are that a high-quality referral has been detected.
Social network analytics: Monitoring customer behavior across multiple claims and lines of business via social network analysis is another helpful technique. The ability to not only ingest and cleanse the data received by a customer but to also invest significant time to entity resolution concerns is critical in ensuring such monitoring can be accomplished with great accuracy. Given unstructured data factors in the claims environment, it is imperative to recognize spelling iterations — those done intentionally by fraudsters and unintentional entry errors; resolving if, for example, "James" and "Jim" are in fact the same individual.
Today’s fraud is dynamic and continuously changing. So, too, must be the tactics used to stop it. When these analytical approaches are combined and working in concert, insurers gain the ability to recognize fraud faster and to identify a higher volume of top quality referrals for further investigation.
This enables efficiency gains in workflows, clarifying initial investigative plans and allowing investigators to hit the ground running with multiple leads to either validate or clear. And, of course, the claims lacking any suspicion of fraud can also be processed more quickly — a boon to delivering great customer service.