The FCA published research on 27 August showing that 56% of 18- to 40-year-olds who hold or are considering investments say they trust AI tools, ahead of television, the press and social media influencers, the last of those on 29%.
Four in five of the least experienced investors had already used AI for help with a decision.
Further down the release, 44% believed AI-generated financial information is regulated, and 32% expected to be compensated by the FSCS or the Ombudsman if it turned out to be wrong.
That survey covered investments rather than mortgages, and at 666 respondents it is small. I would still read it as a fair guide to where borrowers are heading.
THE CLIENT ARRIVES WITH AN ANSWER
Some of our clients now find us through AI in the first place, which is deliberate on our part, and they arrive with a lender already in mind. More than once that lender has been the wrong one for them.
A recent case: a client on a visa, buying at an LTV that the suggested lender does not accept from visa holders.
The concepts the model had explained were sound. What a fixed rate is, how the term works, what the LTV does to pricing, all of it fine.
The application to that particular borrower was not, and the adviser spent the first part of the appointment unpicking product guidance rather than giving advice.
INFORMATION IS NOT ADVICE
The industry reflex is to treat this as encroachment, and I think that misreads what has happened.
Information was always the most copyable part of the job. Criteria, product structures, how loan-to-value drives pricing, what an offset actually does: all of it is now explained patiently and mostly accurately to anyone who asks at eleven at night. That is a good outcome for consumers. It is also a poor thing to have built a proposition on.
Information tells a customer what is possible. Advice tells them what is appropriate for them, and that requires knowing things the model was never given.
Whether they intend to move in two years. How secure the income actually is. Whether a baby is coming. What a payment moving £200 does to that particular household rather than to a spreadsheet. The context is where the advice lives, and the client does not know to type it in.
MODELS PRODUCE ANSWERS. ADVICE IS BETTER QUESTIONS
A client who opens with “ChatGPT said I should take the two-year” has given the adviser a better start than a blank page. The first twenty minutes no longer go on explaining what a fixed rate is.
They go on why they think it is right, what it assumes about rates falling, and what happens if they move in eighteen months. Models produce answers. Advice is mostly better questions.
The weakest response available to an adviser is “don’t trust ChatGPT”. The strongest is to take what it said seriously enough to test what it assumed.
That only works if the adviser knows enough to spot the confident answer that happens to be wrong. Without the domain expertise to push back, you accept a plausible-sounding conclusion, and the model agrees with you faster than a colleague would.
The expertise is what makes the tool safe to use, which is the opposite of the way this is usually framed.
AI WILL EXPOSE MEDIOCRE ADVICE
The obvious objection is a fair one. If the information half of the job can be automated, then a good deal of what some advisers do all day can be automated.
I would not argue otherwise. AI is unlikely to make advisers irrelevant by knowing more than they do.
It is more likely to make mediocre advice irrelevant, by stripping out the part of the role that never required much judgement and leaving the part that does.
Firms whose proposition is answering questions and filling in forms should read the FCA’s numbers as a warning.
Firms whose advisers carry the accountability and take the cases that do not fit a rule should read them as an opening.
THE ADVISER CANNOT ARRIVE MANUAL
It would be a strange arrangement for the client to arrive AI-enabled and the adviser to arrive manual.
The same technology should be reading the documents, flagging what is missing and preparing the affordability scenarios before anyone sits down, so that the collecting is already done and the appointment is spent deciding.
What it does not survive is being bolted onto a process that was already poor. You cannot automate a bad process into a good one, and plenty of firms are about to try.
ACCOUNTABILITY IS THE DIFFERENCE
The FCA’s Mills Review in July put a fifth of UK adults, around 11 million people, as likely to use AI that acts autonomously within goals they have set. The client who arrives with an opinion is the easy version of this.
The harder one arrives when the client turns up with an agent that has already run the sourcing and wants to know why we disagree. My instinct is that the profession is better placed for that than it thinks, because the thing a model still cannot do is be accountable for the recommendation.
The FCA has been clear that accountability stays with named humans.
That is the ground advisers stand on and it is firmer than the mood in the trade suggests.
