Key blurb because the title is unclear: > In this case, Allstate’s model seemed to determine how much a customer was willing to pay —or overpay—without defecting, based on how much he or she was already forking out for car insurance.
They didn’t determine how much the customer was willing to pay, they modelled how much they could change the renewing price and still retain the customer.
In general insurance, there is a large cliff for rate change, where if you change the price higher or lower by more than a certain percentage (different in each case, and for different products) you are much less likely to retain the policy.
Actuarially you have a ‘technical price’ which is the real cost of insurance, and this includes things like acquisition cost and other operating costs, in addition to the risk costs.
When you update your risk model you often have policies that jump in price (if rated as ‘new business’, on technical price) so modelling the retention of the entire portfolio is extremely important. Your price elasticity is in part based on what percentage of the renewing policies you expect to retain with your new rates.
Implementing a ‘capping and cupping’ strategy to limit the amount of rate change delivered in a single year is extremely common, and allows this estimate of retention to be more accurate.
You need policies to eventually get to the technical price (which for large policies, like a fleet of cars, is literally based on prior claims experience) but you also need to sell and retain policies. Capping the amount of rate change applied in one year is a compromise between the two.