Predicting the Future.
Intuition and gut feelings often prove valuable in life. When it comes to insurance, however, they are undoubtedly fallible—and potentially costly. Rules-based systems use only a few factors, propped up by conventional wisdom and hunches. The emergence of predictive modeling, which applies disciplined, statistical techniques to large bodies of data, holds the promise of more objectivity in order to provide improved handling of claims.
Predictive modeling in insurance began in personal lines and has only recently gained a foothold on the commercial underwriting and pricing side, and more recently in claim handling and fraud detection. It is grounded in complex math, but one need not have a Ph.D. to grasp predictive modeling’s potential. Tom Barger, Director, Zurich Services Corporation, Zurich in North America Claims, likens the process to juggling: just as expert jugglers can keep more balls in the air at once, predictive modeling allows insurers to contemplate more factors simultaneously, producing a clearer picture for all involved.
By providing comprehensive risk analysis, Zurich’s modeling technique and multivariable linear regression help improve insurance processes. One universal truth about claims: the more time between an incident and the notice of the loss, the more expensive the claim tends to be. In addition to the single factor of reporting lag time, multivariate modeling considers many variables concurrently, such as the state in which the claim occurred, the type of injury, the age of the worker, the industry, and so on.
“When you do [multivariate analysis], you see that the relationship between the reporting lag and the claim cost is not nearly as direct—in fact, it’s much weaker than you’d get from univariate, or single variable, analysis,” says Zurich’s Barger. “Multivariate analysis is a much better tool for understanding what’s driving the cost. The benefit is that, once you see the real picture, you can see the true cost drivers rather than the mirage.”
This rigorous approach to analyzing large bodies of data produces models and, by extension, behaviors and business practices that utilize available information rather than imposing preexisting ideas. Whether on an individual claim or a collection of claims, this improved identification of the factors driving up costs makes it easier to apply the right resources appropriately and proactively. “Our focus is not only to use the data to determine, for example, what the likely outcome of the claim will be, but also to determine how to best handle the claim based on that information,” says Steve Hatch, Chief Claims Officer, Zurich in North America. For risk managers looking to get the most bang for their buck from limited resources, the value should be clear.
Zurich’s Barger expects predictive modeling to become routine at some point soon in the insurance industry’s future. Eventually, the inclusion of “big data”—text, voice, media, social media, weather, demographic data, customers’ business data, and so forth—may well provide an even more complete picture of what is driving costs. At the moment, not all firms or lines of business are equally well-suited to the process of predictive modeling. Its techniques work best across high volume business lines, due to the huge amounts of data required. Larger carriers, such as Zurich, have an advantage in combining claims data with underwriting data (which third-party administrators and smaller carriers may not have), as well as industry and census data, for starters. To avoid the dreaded “garbage in, garbage out” syndrome, historical data also needs to come from a system that has remained stable over the years—for example, data defined as “litigated” today should be based on the same definition as it was five or ten years ago, or needs to be adjusted to account for any shifts.
Many companies promote their use of predictive modeling, and identifying industry leaders with robust resources and expertise apart from the rest is an important but potentially challenging task. Zurich’s commitment, both in terms of resources and people, is clear: the Zurich Center of Excellence, for example, has more than two dozen dedicated modelers building models for workers’ compensation, general liability, and other product lines.
Barger and Hatch suggest a number of questions to ask potential insurers, to get at the core issues surrounding predictive modeling:
• How are you using predictive modeling?
• How is it changing your processes?
• How does predictive modeling fit into your strategy?
• What are your future plans?
• How are you building your models? Do you employ external resources?
• How many claims are in your data? How were the models validated?
• How big of an investment have you made, in terms of dollars or hours?
Zurich developed its first model with strategic investments of time and resources. The positive impact to customers should prove well worth the effort over both the short and long term.