Every insurer tries to spot potentially fraudulent claims and refer them to a special investigation unit (SIU). Every SIU tracks the results of its investigations. And various reports are sent to trade groups and state agencies. However, putting a dollar value on either the amount of “normal times” fraud or “hard times” fraud is surprisingly difficult.

This is because the aggregate of these referrals and investigations is not the same as the total amount of insurance fraud that occurs. There are two kinds of leakage: fraud the adjustment or SIU processes do not discover, and fraud that is discovered but not pursued. Undiscovered fraud results from poor adjustment practices or technology. Unpursued fraud results from more or less deliberate policies that weigh the negative consequences of very aggressive antifraud measures against the impact on claimant satisfaction, compliance with mandated claims practices, and damage to the insurer's image and market position.

So, even after an insurer acknowledges–perhaps to itself, perhaps very quietly–it will not identify and pursue all fraudulent claims, there still are many things it should do to reduce fraudulent payouts. And there are many ways technology can support those measures.

Creating automated scores for fraud potential. Any insurer writing personal lines workers' compensation or health insurance should provide its claims adjusters with scores that indicate the likelihood of fraud.

Updating fraud potential scores whenever new information becomes available. As a claim is adjudicated, more information becomes available, e.g., police reports or third-party involvement by certain physicians and attorneys. Scores and profiles should be recalculated to reflect new information.

Selective and intelligent use of external data. Checking the people, places, and things involved in a claim against external data sources (e.g., known bad actors, vehicles involved in prior claims, etc.) is a powerful tool for identifying potential fraud. This will involve extra costs, so an insurer must establish guidelines or rules for determining when the cost-benefit relationship is favorable.

Making indicated referrals to SIUs. This might seem achingly obvious, but it is not unusual for a relatively small number of adjusters to account for a large proportion of referrals to the SIU. Conversely, many adjusters make few referrals to an SIU. When this occurs (after making allowance for differences in adjusters' case mixes), this pattern indicates a serious problem. Workflow, rules, and business activity management (BAM) technology can automate such referrals and enhance claims managers' ability to monitor actual performance.

Avoiding making too many referrals to SIUs. The number of false-positive SIU referrals should be minimized. Fewer false-positive referrals make the SIU more efficient–and fewer false-negative referrals will push the loss ratio down.

Improving the SIU process. The SIU investigation process (receive referral, investigate, take appropriate action) mirrors the claims process itself. Many of the same issues apply to both: making correct initial assignments to adjuster and investigators; deciding what external information to obtain or resources to deploy; deciding when and how to complete the process (pay/deny claim, find/not find fraud, and pursue/not pursue restitution or criminal complaint).

There are a number of specific technologies that support these best practices.

Predictive modeling. This is the use of various statistical methods (such as regression, or classification and regression trees) to establish the probability a given claim is fraudulent. A trained analyst or statistician will use several techniques to find which model or models provide the greatest level of predictive power. An important benefit of predictive modeling is transparency–it is easy to identify and describe the factors and the weights that contribute to the outcome.

Neural networks. This also is a statistical method in which programmed processing agents work in parallel to develop a fraud score. They have been widely used to determine fraud in the credit card industry. Neural network analysis can be used with large amounts of structured or unstructured data (e.g., adjuster notes). Neural networks may do a better job at finding fraudulent claims that prior fraud detection methods missed. A disadvantage is it is difficult to explain how neural nodes work to an adjuster, an SIU investigator, or a state Department of Insurance regulator.

Profiling. Although this term has become politically charged in recent years, it remains a legitimate and powerful technique when used properly. Profiles are characteristic patterns of behavior and circumstances. For example, a profile of a valid claim could include a distraught claimant at the time of the first notice of loss and a more calm and cooperative claimant on subsequent contacts. A profile of a fraudulent claim could include a calm claimant at the first notice of loss, followed by several aggressive contacts in rapid succession. Profiles also are very important for understanding typical and atypical patterns of injury (various injuries resulting from a specific trauma) as well as patterns of care (two doctor office visits vs. 10 doctor office visits).

Claims databases. There are several commercially available databases of claims that have been made over many years to many insurance companies. One indicator of potential fraud is an association of elements in a current claim (such as a claimant, a third party, or a specific vehicle) with earlier claims. Claims databases allow an adjuster (or an automated function within a claims system) to query these large databases of prior claims to discover whether there are shared elements with the current claim. One such association or more can indicate a somewhat elevated potential for fraud.

Identity matching. A person's identity has multiple dimensions: name, address, phone numbers, Social Security number, occupation, physical appearance, etc. There is no claim-related reason for a person making a valid claim to change any element of identity. However, people with a pattern of making fraudulent claims frequently will change various elements in their identity. The goal of identity-matching technology is to create a complete record of specific individuals' interactions with all insurers. A person of interest could be a claimant, an injured party, a witness, a medical or rehabilitation provider, or an attorney. For practical reasons, people almost never change all the elements of their identity. Robert Stone may become Rodney Silver, but he will keep the same cell phone number.

Link analysis. This technology examines the connections among people, places, and things involved in multiple claims. Initial recognition of recurring connections can be established by the automated use of search formulas (algorithms). Findings usually are displayed graphically, giving an analyst the ability to start with a single element (for example, an attorney) and see what links exist with physicians, accident locations, witnesses, and vehicles. Link analysis is well suited for discovering or investigating patterns of organized fraud.

Other technologies. Business rules or business process management solutions can be used to initiate the calculation or updating of fraud scores and the subsequent actions adjusters should take. Additionally, case management tools can be used by managers and staff in SIUs. Much like modern claim adjuster desktops, they can provide a digital repository of all relevant information as well as workflow design, control, and monitoring.

Even in these hard times–or perhaps especially in these hard times–insurers are looking at the returns of using technology to reduce fraud.

Donald Light, based in Celent's San Francisco office, is a senior analyst in the firm's insurance group. His research focuses on claims, underwriting and policy administration, reinsurance, product life cycle management, business process and rules solutions, and accounting. He is the author of reports on policy administration, claims, underwriting, product development, insurance accounting solutions, and fraud mitigation technology. He can be reached at [email protected].