The Technology

In this case study, we worked with an MGA that writes non-standard auto insurance. We fed our AI system a set of actual insurance policy and claims data, then asked it to 'score' every submission the MGA receives, live and in real-time, returning responses on each policy in under one second each. This algorithm has been running live for over two years now and has scored over 10,000 policies in that time. Each policy's score represents our system's estimate of the expected profitability that policy would achieve, on a policy-by-policy basis. We then sorted those policies top-to-bottom based solely on the assigned score and split those policies into ten equal-sized buckets, best scores to worst. We then calculated the actual, eventual loss ratio for each bucket. The graph below displays the result -- the Soteris scoring system demonstrates a clear ability to predict which policies are likely to lead claims, and which ones are not.

Our software has raised this customer's policy profitability **over 250%**.