In my last post, I described how the insurance fraud landscape is shifting toward a greater focus on organized ring activity. In order to deal with the changing nature of insurance fraud exposure, the insurance industry needs to shift its thinking. But there are multiple hurdles to overcome in order to drive this fundamental change in the way fraud is addressed.

Where’s the data?

One of the biggest challenges insurance companies have when trying to become more proactive in detecting organized fraud activity is gaining access to the right data sources. Many companies are saddled with multiple legacy claim applications, systems that are segregated by line of business, disparate homegrown SIU case management systems and vendors that house critical data points like medical billing information. Looking to aggregate this information and combine it with policy, HR, and other internal systems is a daunting task. And then there’s the opportunity to incorporate information from third party data providers which adds a whole new layer of complexity.

Many projects fail at this early stage because they are perceived as massive IT projects with a huge price-tag. While these projects can be complex, narrowing the scope of the effort and leveraging robust data integration and data quality tools can make this happen much faster and more cost-effectively.

Each claim must be investigated on its own merits

Most insurer Special Investigation Units (SIU’s) were originally designed to handle investigations of single suspicious claims. In the early days, many SIU’s focused on investigating suspicious auto theft claims. SIU’s gradually expanded to conduct investigations on other types of claims but the approach was the same. This single-claim approach – which paralleled the claim adjudication process – became the foundation for future SIU growth and the driver of the business processes, metrics and organizational structures still in use today.

But times have changed and SIU’s need to keep pace. While a large percentage of SIU work still involves investigating single opportunistic suspicious claims, organizations need to shift (or add) more of their resources to focus on the growing threat from organized activity. The existing structure – where investigators carry a caseload of individual claims – is not conducive to detecting or investigating organized fraud ring activity. While legal, regulatory and contractual obligations still warrant a thorough individual claim investigation; a new complementary structure must be designed to address the new fraud risk. This structure should focus more on data analytics and investigation of suspicious providers, attorneys and claim participants.

Detecting organized fraud risk is different

SIU’s have long relied on lists of red flags to help claim adjusters spot suspicious claim activity. But the over-reliance on these lists has cultivated a mindset that fraud can be spotted with business rules. With organized multi-claim frauds, this is simply not the case. A single claim adjuster does not have enough information to rely on red flags to spot these types of fraud scams. The false sense of comfort provided by red flag lists has done more to hurt insurers’ ability to detect organized fraud exposure than practically anything else.

Insurers need to use new methods to detect organized fraud exposure. While red flag business rules can be useful when automated to look across large data sets, alone they still produce undesirable false positive rates. In order to effectively detect and prioritize these types of cases, a hybrid approach to fraud detection that leverages multiple detection technologies is critical. Scammers are getting increasingly creative and suspicious rings are not necessarily easily spotted by something as simple as a known attorney/provider relationship. These rings have multiple layers and advanced social network analysis is necessary to uncover those complex relationships.

The value calculation has changed

Ask ten insurers how they measure SIU results and you’ll likely get ten different answers. But at the end of the day, the primary focus is always on how many claims were impacted and the dollars saved or recovered as a result of the investigation. While affirmative litigation in organized provider-level investigations may result in some dollars recovered, the ultimate value of these cases is based on future exposure averted.

Unfortunately, the value of future exposure is much harder to quantify. Not having a concrete ROI calculation makes it much harder to justify investments in organized fraud detection and investigation programs. This is another factor that causes investigation programs to fail before they even get started. Leading insurers need to develop metrics for this type of savings and consider the long term value of organized fraud detection programs.

We already have a predictive model, and it stinks

Over the last decade, many insurers have heard about “predictive modeling” and many have even tried building models for fraud detection either with inside resources or by purchasing a vended black-box solution. While it’s fair to say that results have been mixed, many insurers have expressed disappointment with these early predictive modeling efforts. Why haven’t insurers recognized all of the value promised by predictive modeling?

There are several reasons for this. First, building a model is not overly difficult. But, there are several factors that influence the ultimate usefulness of the model. As always, the devil is in the details. Consider these factors:

• Some “predictive modeling” solutions aren’t really predictive models at all. Many simply rely on a bevy of red flag business rules. The tool may produce a “score” but it’s really just a number based on a big library of rules. And business rules alone typically yield lots of false positives.
• Traditional supervised predictive models use historical results to predict future events. Even the best models only identify the types of fraud that have already been identified. Some insurers do not have robust fraud history to build from or they are expecting the model to find “new” frauds, but traditional supervised predictive modeling is not designed for this purpose.
• While many large insurers have the in-house expertise and technology tools to build a predictive model, it may take up to a year or more to develop a model that yields tolerable results. But, what many insurers lack is the technology to operationalize the model. Results are often shipped out in spreadsheets via email, quickly becoming cumbersome and unmanageable. Proper operationalization often takes an additional 12 to 18 months after the model is built.

It is absolutely possible for an insurer to develop a fraud detection solution internally. However, most woefully underestimate the time and resources needed to do it properly. Unfortunately, the result is often a lackluster model that produces lots of false positives and is difficult to manage. While this has soured some investigations and executives on predictive modeling, all hope is not lost. There is a better way.

Solving the problem

A hybrid approach to fraud detection can help address all of the issues identified above. As insurers continue to make investments in their SIU program and adjust resources to address the growing exposure to organized fraud activity, they need to make similar changes in their detection and prioritization technology to support the analysts and investigators in the field. The insurers who recognize this and take quick action will be leading the pack while those who fail to adapt will quickly become soft targets to increasingly sophisticated organized insurance fraud rings.

James Ruotolo is the Principal for Insurance Fraud Solutions at SAS. Connect with him on Twitter or LinkedIn.