Last year, the ISO team surveyed commercial lines writers, large and small, to learn more about their near-term and long-term predictive modeling initiatives, abilities, and needs. What we gleaned will inform our goals as we ramp up our commercial lines modeling efforts in the second half of 2012.
In general, insurers’ commercial lines modeling efforts lag their personal lines efforts. Some insurers told us they don’t use predictive models at all; others use them for underwriting guidance and schedule rating but not class plans. Insurers that have used sophisticated modeling to develop rates and rating factors have either positioned the models on top of their current rating plans or, more rarely, as completely new rating plans.
Who develops the models? The answer we received correlates with insurer size. Small and medium-size insurers are more likely to work with consultants. Sometimes that’s a short-term partnership in which the consultants build the models and the companies maintain them; in other cases, it’s a longer-term relationship in which the consultants continue to maintain the models. Large companies, however, are more likely to do it all themselves.
Insurers consistently mentioned several constraints to building predictive models, though the importance of each constraint varied from insurer to insurer. IT system impacts were a major concern as were insureds’ premium swings. Respondents also mentioned the needed buy-in from both management and agents and — not least of all — the need to assemble a technically proficient staff with combined expertise in data management, statistical modeling, and insurance.
Which commercial lines of business are ripest for modeling? Among the “lowest-hanging fruit” are commercial auto and businessowners, two lines where insurers can repurpose some results from personal lines models. Insurers also expressed interest in workers compensation and general liability models. A major driver of interest was how much data the insurers have. Most don’t have enough to build robust businessowners policy models, especially considering that they would most likely need separate models by peril and business segment. The problem is even more acute in general liability modeling.
Finally, what types of data are insurers including in their models or looking to include? From what we heard, it seems there’s no such thing as too much data. Companies are mining their own data for variables related to a policyholder’s payment history and selected payment plan, variables related to the agency that issued the policy, and variables related to the length of time the insured has been with the company. They’re also seeking to tap third-party vendors for other business information, such as data gleaned from MVRs, credit reports, claim history reports, and detailed property-specific information.
ISO is now formulating plans for building new commercial lines predictive models. Some of the work is already under way; other aspects are still being mapped out. Upcoming blog posts will detail these initiatives. We welcome your comments. You can send e-mail to firstname.lastname@example.org, or use the form below to share your comments with our other blog followers.