The protection prompt: what AI still needs to get right

The Mills Review has prompted plenty of discussion across the mortgage industry. While much of the focus has been on the finding that a quarter of UK adults would trust tools such as ChatGPT with financial decisions, another point in the report is equally significant.

The FCA identifies both the advice gap and the protection gap as areas where AI could help improve access. Currently, only 9% of consumers use traditional regulated advice, and just 30% hold life insurance or income protection.

That second figure is particularly interesting. The regulator isn’t just focused on the advice gap; it’s also clearly identifying protection as an area where AI could help improve outcomes. So, what might that look like in practice?

Mortgage-related questions tend to start with a clear customer need – rates, affordability, or whether it’s the right time to remortgage. Protection conversations rarely start that way. The FCA found as much in its own protection market study: most people never buy cover, not because it’s too expensive, but because nobody ever put the question in front of them.

And once they are asked, things can get complicated fast – health disclosure, exclusions, what ‘unable to work’ means for their specific job.

It’s rare for someone to start a conversation by asking whether they need income protection. Buying protection has never been the kind of thing people wake up in the morning wanting to do. That’s where advisers continue to play an important role. They introduce the conversation, explore a customer’s circumstances and surface risks they may not have considered.

As journeys become more digital or AI-supported, it’s worth considering how those conversations continue to happen, and who takes responsibility for prompting them.

Compared to other products like debt advice, the Mills review found that mortgages are currently one of the financial products consumers are least likely to rely upon AI to help them with. The immediate risk then isn’t that AI muscles the adviser out of the mortgage appointment tomorrow.

It’s slower than that, and arguably sneakier: the FCA expects reliance to build as AI systems get more capable and more embedded, and the protection prompt is exactly the kind of thing that quietly stops happening once a journey speeds up, long before anyone notices it’s gone missing.

INFORMED ISN’T THE SAME AS RIGHT

Then there’s the customer who’s done their own research. Every adviser knows this person. They used to arrive with a printout from a comparison site; now they’re just as likely to have spent twenty minutes asking a chatbot about their cholesterol, their job title, or the type of cover they might need.

The challenge is that while AI is making customers more informed, it carries the risk that they’re more easily misinformed where a chatbot isn’t accurate or hasn’t personalised the customer’s needs.

The Mills Review isn’t blind to this itself. Consumer groups told the FCA directly that people over-trust AI outputs because they’re delivered with such confidence, and struggle to spot when they’re wrong without checking elsewhere.

Get an AI-generated mortgage rate wrong and a broker fixes it in the first five minutes of the call. Get a protection recommendation wrong, an underestimated cover amount, a skipped disclosure question, a customer convinced they already know what they need before underwriting even starts, and the mistake doesn’t show up until a claim does. That’s a much worse moment to discover it.

The Review itself suggests AI agents could help close the protection gap by prompting people at the right moment, rather than waiting to be asked. That shifts the conversation from whether AI belongs in the mortgage journey to how protection discussions remain part of it as the process becomes quicker and more automated. If those prompts aren’t built into the journey, they’re unlikely to happen by chance.

Mortgage advisers and specialists have spent years being the ones who ask the questions the customer didn’t think to bring up. The FCA is now betting that AI can do some of that asking for them.

Advisers would be wise not to assume it can, until someone has built it, tested it and shown it can do the job consistently.