Overview of Detection and Investigation Techniques

David J. Rioux | September 14, 2009

• Announcing the CI Legal Expert Panel
• Another Twist to Consider—The Allstate Case
• Competition is Fierce—Restrictive Covenants in the Shafron Case
• Internet Use (To Catch Abuse)
• Facebook Litigation Grows
• Proof—Roles of Adjusters and Lawyers in Fidelity Recovery
• Best Practices for Successful Subrogation Recoveries
• State of Securities Class Actions
• Professional Liability and Securities-Related Claims Trends
• Misguided Critique of Covert Surveillance
• Employee Termination—A Risk You Can Manage
• Using Data Analytics to Maximize Company Performance
• Give It Up—Vehicle Fires & Frauds
• Overview of Detection & Investigation Techniques

Back to main page

Most people have an image of the stereotypical insurance fraud investigator: hiding in the bushes camcorder in hand, waiting for the unsuspecting claimant.

And despite all the advancements in technique, technology and proficiency, fraud investigators still relies on these old methods of detecting deceit.

While most insurers (and investigators) have invested in reengineered and automated processes, the primary responsibility for detecting and documenting fraud continues to rely on good old-fashioned detective work—detective work that relies on specialized training, industry experience and a solid appreciation of red flags.

That means that no matter how technologically advanced fraud detection becomes, there will still be a need for the experienced investigator.

Predictive Modelling

One major trend is the use of data mining and information intelligence. For example, predictive modelling is a front-end tool with the capability of flagging questionable claims at first notice of loss. As such, there are two types of predictive models: supervised and unsupervised.

Supervised methods use an historical database of known fraudulent and legitimate claims from which to construct a data model. This data model used to detect fraud types that have previously occurred. Unsupervised methods seek to identify those claims that are most dissimilar from the vast number of other claim types. Unsupervised models do not require claims with known fraudulent outcomes. Claims identified as outliers from the rest are a basic form of nonstandard observation.

In either case, predictive models are built from a company’s historical claims data. New claims data is run through the model and scored. If the score hits a predefined threshold the claim is then flagged. Indicators of potential fraud may be subtle or obscure from common recognition. It may be the absence or inclusion of certain information in the claim, or the relationships and combinations of attributes present on a claim that the data model derives.

The potential benefits of predictive modelling can be significant. It provides a global and more consistent screening process across the claims organization. It also helps to identify claims with the greatest propensity for fraud much earlier in the claims lifecycle. Claims that are scored and flagged then investigated for fraud typically have a much greater success rate and produce a higher return in the claims fraud mitigation process.

Data Analytics

Information may come from many different data sources, pulled together by technology tools.  Unlike predictive modelling, these tools are typically claims back-end solutions used by data analysts and are transparent to a claims adjuster. They support the investigative process when fraud is suspected and some offer proactive fraud detection capability on the backside. Data analytics ability to mine millions of claims and other records can have a significant impact on a company’s fraud fighting efforts and results.

One major obstacle in developing actionable intelligence is ‘meaningless information.’

Meaningless information is data that lacks causal relationship and relevance. It’s data that serves as a processing function where no common connections are typically made.

For example, contact information i.e. a name, address and phone number independently may be meaningless to fraud detection, but indexed across other source data may reveal evidence of suspicious activity.

The industry is faced with too much data and not enough information.  Turning data into knowledge in a timely and relevant way is crucial to fraud detection.

The answer to the problem of meaningless information is data visualization or link analysis technology. These solutions can pull together seemingly unrelated pieces of data from disparate sources, analyze and transform them into meaningful information. Visual analysis software can be used to reveal patterns, trends and relationships contained within complex data sets. Unlike statistical analysis, which deals mostly with aggregated results and reports, proactive and reactive analysis can explore direct and indirect connections in data. These patterns and relationships emerge from the data and are presented in a graphical representation.

One area data analytics is a major help is in the use of public records. Many companies use public records and other information database sources in the investigation process to develop predictive models or to determine metrics that will help in the claims and rate setting process.

By using external data, carriers and claims professionals supplement their fraud detection and investigation process and add value to a company’s ability to detect and investigate fraud.

While the use of external data with link analysis technology has been around for some time, introducing and determining the value add of external data in the predictive modelling process is still very much in its infancy stage. For that reason a company should always first leverage and perform analysis on internal data before seeking external sources, and carefully evaluate the use and type of external data for effectiveness.

Antifraud Solution

Fraud technology, due to recent advancements, can significantly expand the ability to identify and investigate potential fraud over manual methods (some still in use today). Yet, fraud technology is not a solution by and of itself. A quality fraud investigator is essential to ensure that technology and knowledge/experience are used in conjunction for maximum advantage.

Aggressively combating insurance fraud makes good business sense because it impacts the bottom line, competitive position in the marketplace and the policyholders. A strong antifraud program can serve as a deterrent against those individuals looking to commit insurance fraud and lead them elsewhere. Even small incremental changes in the fraud identification and investigation process can have a large impact to the fraud mitigation rate. For example, a quarter percent increase in the number of fraud cases referred to and investigated by a special investigations unit (SIU) could result in millions of dollars more in migrated losses. Less than a 10% improvement to identifying better quality SIU referrals could result in millions of dollars more.

Today, an effective fraud solution requires a holistic approach to the problem, including support at all levels of the organization, as well as a skillful blend and integration of technology, people and the right business processes. Using the right types and mix of technologies in concert with human talent like claims adjusters, investigators and analysts is critical to a successful fraud program. It leads to better information’s intelligence that delivers results.

David J. Rioux is vice president and manager of the Corporate Security Department for Erie Insurance, and president of the International Association of Special Investigation Units. david.rioux@erieinsurance.com