Personalised price, or quiet value extraction?

Since 1 January 2022, UK home and motor insurers have had to quote renewing customers a price no higher than they would quote the same person as a new customer. The FCA called it the equivalent new business price rule (PS21/11). Everyone else called it the end of price walking: the habit of nudging loyal customers up a little every year because they were unlikely to shop around.

It was a good rule. It was also a narrow one. It fixed one signal, tenure, and left the rest of the pricing model alone.

The admission in the Mills Review

The FCA’s Mills Review, published on 6 July 2026, contains a sentence that every personal lines pricing team should pin above their desk:

“AI-enabled pricing may make it harder to distinguish benign personalisation measures from extraction where consumers pay more without receiving additional value.”The Mills Review, FCA, 2026

That is the regulator saying, in plain terms, that it may not be able to see the problem. Elsewhere the same review warns that hyper-personalisation “could help better match products to needs, but also enable bias, opaque pricing and personalised manipulation”.

Our view: the next fairness fight is inside the model

A pricing model with thousands of inputs does not need a field called “will not shop around” to learn who will not shop around. It can find the pattern in payment method, channel, time of day, device or any of the proxies that sit in a modern quote journey. Nobody has to intend it. The model only has to be rewarded for margin.

We are not saying UK insurers are doing this. We are saying that, as the review itself puts it, it is getting harder to prove they are not, and that the burden of proof sits with the firm. Under the Consumer Duty, firms must be able to show their products provide fair value. “Our model is too complex to explain” is not a fair value assessment.

The quieter risk: everyone prices the same way

The review flags a second problem that gets less attention. Where firms “use similar models for credit underwriting, investment recommendations or pricing, synchronised decision-making could amplify market movements, create procyclical effects, or lead to coordinated withdrawal of services from particular customer segments.”

In insurance terms: if several large insurers buy pricing tools built on similar models and similar data, a postcode or a customer group can become expensive everywhere at once. No cartel, no meeting, no intent. Just correlated software.

What we would ask for

• Segment-level fair value evidence. Not one average for the whole book, but outcomes by customer group, published at least to the board and on request to the FCA.
• A plain answer on renewal. Customers should be able to ask which broad factors moved their price this year, and get a straight reply.
• Proxy testing as standard. Before a new rating factor goes live, test whether it is mostly predicting loss or mostly predicting loyalty.

The price-walking ban showed the FCA will act when a pricing practice is clearly unfair. AI makes “clearly” much harder to reach. Insurers that build the evidence now will be the ones still trusted when the next rule arrives.