Predictive Modeling’s Role in P&C Industry

San Diego — Health insurers have been known to use predictive modeling within a number of business processes—claims, fraud detection, underwriting, etc. However, new advances in predictive modeling will play an increasingly important role in the property/casualty insurance business in the next decade, panelists told the Casualty Actuarial Society (CAS) Predictive Modeling Seminar in October.

“There is a growing recognition in cognitive psychology, behavioral economics and business that predictive models across the board in many different industries, including property/casualty insurance, help human experts make decisions more accurately, objectively and economically,” said plenary session moderator Jim Guszcza, the national predictive modeling lead for Deloitte Consulting LLP.

Guszcza noted that predictive models have enabled insurers to build underwriting models with significant segmentation power and are increasingly being applied in such areas as claims modeling, agency analytics, customer segmentation and target marketing, and price optimization.

Key new rating variables that are being incorporated into insurers’ predictive models include homeowners rates by peril, homeowners rating by building characteristics, vehicle history and usage-based auto insurance, according to CAS.

Citing the example of usage-based auto insurance, Robin Harbage, senior consultant, EMB America LLC, said that predictive modeling allows both commercial and private passenger auto insurers to develop more reliable rates.

“For a long time, we have collected roughly 40 static pieces of information on drivers and most of that information has nothing to do with how they operate the actual vehicle, except for points and violations,” he explained.

Standard auto insurers, specifically, can benefit from predictive modeling, according to a March 2008 study from Conning Research & Consulting Inc. titled "The Nonstandard Auto Insurance Market: Evolutionary Challenges."

"As predictive modeling has become more prevalent in auto insurance underwriting, the standard auto market has expanded to include and price risks that would once have been thought of as nonstandard," says Alan Dobbins, analyst at Conning Research & Consulting. "As a result, the new nonstandard market is undergoing a dramatic shift in risk profile."

Predictive modeling and newly emerging claims technologies should continue to provide leading insurers a significant advantage in the nonstandard segment, according to the study. Smaller and/or unfocused companies may try to replicate these capabilities, but then continue to struggle as more sophisticated models emerge.

At the CAS seminar, Glenn Meyers, chief actuary, ISO Innovative Analytics, observed that there is a lot of opportunity for predictive modeling applications in the area of insurance claims, particularly in fraudulent claims detection. “We cannot explicitly identify fraudulent claims,” he said. “Quite often the individual information on the claim itself is not sufficient to identify a fraudulent claim. But many fraudulent claims are potentially organized, so one valuable approach is to look at relationships over multiple claims.

“You have a claimant in an accident who shares a telephone number with the witness in another accident,” Meyers continued. “Or an insured in one accident shares a social security number with an insured in another accident. These are the kinds of connections that cannot be identified on an individual claim. New techniques such as data visualization can now be used to inspect the data and see the potential links.”

Highmark Inc., though it provides health insurance, is an example of an insurer embracing predictive modeling for fraud detection. It implemented a robust case management system, which enables it to scan all documentation so that it is available electronically, Tom Brennan, director of special investigations for the healthcare insurer told INN in October. The carrier also is a proponent of rules-based tools, which are being widely adopted. "If you have an issue, you can build an algorithm, go into your data and find out just how bad you've been had," Brennan said. Yet, Highmark felt the need to go beyond business rules, and thus embraced predictive modeling, which uses statistical analysis to establish the probability of a claim being fraudulent.

Sources: Casualty Actuarial Society and INN archives

Exclusive content only available on InsuranceNetworking.com