Predictive Analytics

Claims

Pinnacle was approached by a regional insurer that wanted to develop a predictive model that estimated the time that a claim would be open based on what is known at the first notice of loss. The company felt like this would allow them to more effectively manage their caseload and handle claims more proactively. Pinnacle, through the use of predictive modeling, assisted the carrier in designing a model which predicted cycle time based on the FNOL. Not only did this give the company a better understanding of its claims, but also assisted them in understanding their claims data better and improving their data collection.

Commercial Lines

Pinnacle was approached by a national insurer that wanted to develop a more sophisticated commercial automobile rating program. Their current commercial automobile plan was a traditional rating approach and did not take full advantage of driver, credit scoring or vehicle characteristics and the company felt they could take advantage of a significant opportunity in the market. Pinnacle, through the use of predictive modeling, assisted the carrier in designing a new rating and tiering structure, which included modifications of the rating plan, the introduction of underwriting scoring, and new territory definitions. This new structure allowed the company to have more precise rating, more adequate and yet competitive rates for a broader spectrum of risks.

Territory / Summit Analysis

The insurer had not reviewed or adjusted their territorial boundaries in several years; and their current territories were not based upon an analysis of their underlying loss experience.

Pinnacle began with an evaluation of available insurer and external industry data at the ZIP code level as well as their current and future policy processing capabilities. After all of the experience data was adjusted to a common base level, we used our Summit® software product to smooth the data to develop an initial adjusted pure premium by ZIP code to reflect both experience in a ZIP code and in neighboring ZIP codes. The number of additional neighboring ZIP codes used was dependent upon attaining a sufficient exposure level.

The smoothed data was then clustered using Summit® on both a contiguous and non-contiguous basis. The insurer needed to decide whether to maintain relatively contiguous definitions or use more granular and theoretically accurate non-contiguous definitions. Using statistical measures, we identified and graphed optimal definition sets. Non-contiguous definitions were selected by the client. To minimize the number of territories with a limited geographic area or only a few ZIP codes, some manual combination of “outlying” ZIP codes with territories (clusters) with similar adjusted pure premiums was performed.

Finally, we assisted the insurer in developing the necessary filing support material to gain approval. The new program is generating new business in areas of historical profitability but previously less than average competiveness.