PAI: Can you tell in general what problems insurance companies are facing at the moment?
DP: It’s my understanding that, worldwide, insurance fraud is becoming more and more of a problem for all companies. Organized fraud rings in the U.S. have recently received significant media coverage regarding their success in perpetrating fraud. This success is now being replicated in other parts of the world. It’s grown in the U.S. to the point that virtually every insurance company above a certain size now has a fraud prevention program, and people dedicated to investigating fraud. Government has intervened to the point where some states actually mandate the structure of such departments, and require fraud plans and reporting. Europe and the Pacific Rim countries have insurance companies who openly acknowledge that fraud is an issue that has historically been ignored—one that it is growing and deserves attention.

Most insurance companies also see a need to identify claims that can be pushed through the claims process with more efficiency and better customer service. This may, in fact, be more valuable to most companies than detecting and preventing fraud.  

PAI: What sorts of solutions did those companies come up on their own?
DP: Most insurance companies handle fraud detection issues as a business process, training adjusters and encouraging fraud referrals publicly. The sad truth is that adjusters, by and large, are motivated to take steps necessary to reduce their claims inventory. Identifying claims that are suspicious and then taking the steps to avoid paying them is incredibly more time consuming and risky. Companies that have taken an automated approach to finding more referrals are creating a rules engine approach which typically provides far more false positives than is acceptable. These engines also often require time and energy from IT to make any changes to them.

PAI: Are there any differences between U.S. and European companies?
DP: The main differences that I’ve seen are the value of the claims and the data available to help discover suspicious aspects. Claims in the U.S. are typically more valuable because of the way attorneys are involved and are compensated. Attorneys often drive up the perceived value of a claim by encouraging claimants to be over-treated for minor or non-existent injuries. In Europe, public data sources are generally more restrictive, so investigations may become hampered.

PAI: How can predictive analytics help improve customer satisfaction, decrease costs, and detect more fraud in insurance claim handling?
DP: Predictive analytics allows claims to be segmented into three categories. One category is claims that represent very little risk of being fraudulent or of high exposure to the carrier. These claims can be settled quickly with very little work. The second category is claims that are blatantly suspicious (although not necessarily fraudulent). These types of claims require special attention. The third category is claims that warrant the standard amount of processing.

Robust predictive analytics solutions allow for frequent re-evaluation as often as pertinent new information becomes available. A decrease in costs can be achieved by not paying claims that shouldn’t be paid, as well as by reducing the number of people that must be involved in processing claims. Another decrease can be achieved when you can engage special investigators much earlier in the claims process. Discovering early in the process that a claim is either valid or invalid shortens the time it takes to close the claim. Detecting more fraud is inevitable when an objective scoring system is in place.

PAI: Can you tell me more about how the process works?
DP: The most effective predictive analytics vendors take a customized approach to every insurance company. However, there are commonalities in how predictive analytics can be used for fraud detection.

Predictive analytics uses proven technologies known as business rules and predictive modeling, which analyze historical claim data to predict claimant behavior and identify both known and new fraud risks. Business rules capture the suspicious fraud indicators, or behaviors, for the company. Business rules can be updated and new ones added to account for specific events or new indicators. Once identified, these business rules are combined with predictive modeling, which uses sophisticated techniques that enable organizations to analyze claims and detect forms of suspicious behavior that may indicate fraud.

Predictive modeling enables organizations to enhance their business rules by discovering forms of suspicious behavior which may have been occurring, but have gone undetected, and—more importantly—by discovering new forms of suspicious behavior as they occur. The most effective predictive analytics solutions use real-time decision optimization, which immediately assesses claims at the point of submission and assigns a score based on the potential fraud risk associated with that particular claim.

This combination of analytics and decision optimization enables companies to settle legitimate claims quickly—often in as little as one phone call or one online session—and to detect potentially fraudulent claims before they increase your costs and impact other critical measures of performance.

PAI: How is predictive analytics implemented?
DP: It begins with an in-depth study of the needs of the company.  First, interviews with the company’s professionals should be performed and the company’s experiences validated against general industry data. Then, business rules should be developed which are meaningful and can be used to create predictive models from company data. The scores generated by these models will be the result of a comprehensive analysis of all that is found and evaluated.

When a customer makes a claim—whether through a call center, online form, or in-person visit—the claim information is analyzed against a combination of customer information, claim type characteristics, business rules, and fraud risk data. Based on this analysis, each claim is assigned a score that reflects its level of exposure and fraud risk. Each score is backed up by understandable descriptors such as “accident happened at night, there were no police at the scene,” so claim handlers or investigators know why a claim received a particular score. Low-risk claims, for example, are scored for fast-track approval, payment, and closing. If a claim is scored as medium-risk, or is missing critical information, the agent or claim handler can be prompted with “smart” questions during the initial call or session. High-risk claims are scored for referral to an investigative team or SIU.

PAI: What is the most important data to consider gathering? How detailed should you make this data?
DP: All relevant claims data, policy data, and payment data is critical. In general, the more detail the better. It’s preferable to begin with a very broad set of data, and then discard those pieces that don’t add value, rather than attempt to build models based on limited information.

PAI: What will the outcome be?
DP: They will see more referrals to their SIUs than they had seen before. These referrals will be based on the factors they identified as likely predictors of fraud, and they’ll see claims referred earlier in the life cycle of the claim.

PAI: What do you improve?
DP: We improve claims processing and suspicious claims detection.

PAI: What is the average return on Investment?
DP: If an insurer can adequately investigate all of the referrals that deserve to be investigated, they should be able to at least triple or quadruple the amount of money they avoid paying.

PAI: How quickly is your investment likely to pay off?
DP: A definitive return on investment (ROI) for insurance fraud is somewhat difficult to quantify. A starting point is to look at the value associated with the claims that were not paid due to fraud. It’s important to gain consensus early on regarding what the value of a claim is. I believe most insurance organizations see a return on their investment within 3-6 months, easily. 

PAI: Where and how do these projects most often fail?
DP: I personally have yet to be involved in an unsuccessful implementation. None of our customers have expressed dissatisfaction with the results of their predictive analytics solution.