Traditional vehicle rating is based on grouping similar vehicles together with some adjustment for loss experience. This session will show two complementary concepts of how detailed vehicle information can be used to develop accurate predictions of insurance loss. For each concept, we discuss the types of data available, data preparation, the modeling process, and the results on actual industry loss data.

The first part focuses on an advanced analysis of vehicle characteristics on a make/model/year/trim level. We look beyond simple VIN decoding approaches, and introduce text-mining tools in order to bridge disparate formats of key information. The second part considers CARFAX vehicle history data on unique vehicles (i.e. the history of your specific car). Through pattern extraction and advanced analytics we recreated a CARFAX-based score that adds substantial incremental segmentation power above and beyond the rating elements used today.