Propensity model

In September 2020, the Financial Conduct Authority published the final report of its market study on home and motor insurance pricing: around six million UK policyholders were paying high prices, and they would have saved 1.2 billion pounds in 2018 had they paid the average price for their risk. Propensity models sat at the center of the mechanism. A propensity model estimates the probability that a customer behaves in a given way: lapsing at renewal, accepting the renewal, converting a quote, buying a second product. It is trained on the history of quotes and renewals, with the price offered and its change among the variables, which yields an elasticity: how far the probability of staying drops when the premium rises by 5%. In price optimization, this demand model is combined with the technical cost of risk to find the price that maximizes a margin or volume objective under constraints. The same tool serves more harmless purposes, targeting a sales follow-up or spotting customers about to leave for a retention call. The UK slippage came from how it was used: insurers identified price-insensitive policyholders, often loyal for years, and raised their premium at every renewal. Since January 1, 2022, FCA rules have required that a renewal price not exceed what an equivalent new customer would get through the same channel, as set out in the entry on the price walking ban. Propensity models have not left the UK market. They now weigh on acquisition pricing and on the choice of which customers get a retention discount, which the cap does not forbid. Under the GDPR, such a model is profiling.