When analyzing geographical risk for auto rating, the focus is primarily on company experience, which can lack credibility and be slow to react to changing conditions. To address this, companies typically incorporate a complement of credibility, such as data from a statistical organization or competitors. Even though this helps address the credibility issue, it still suffers from only providing a limited view of the complete geographical picture and remains slow to respond to changing conditions. This session will describe how the use of machine learning applied to detailed road segment data provides a more accurate and dynamic view into the geographic risk associated with auto insurance.