The FCA has opened something called the AI Input Zone, and most mortgage brokers will never hear about it. That is the problem.
It is not a consultation on draft rules. There is nothing to object to, no proposal to push back on. The regulator is doing something quieter and, in its way, more consequential. It is asking the industry to send in real examples of artificial intelligence working well and working badly, so it can anonymise them and publish a “good and poor practice” guide later this year. The window closes on 19 June 2026. Three questions: what has helped firms use AI safely, what has stopped them, and what you want the eventual guidance to cover.
Here is the part that should make every brokerage principal sit up. The FCA has gone out of its way to say it wants to hear from small firms, the ones with fewer than a hundred staff. That is not a footnote. That is almost the entire mortgage intermediary sector. And small firms are precisely the ones least likely to respond, because they are busy, because nobody is paying them to write submissions to regulators, and because there is a comfortable assumption that this sort of thing is for the banks to worry about.
Think about what happens if that assumption holds. The submissions that land on the FCA’s desk will come overwhelmingly from large institutions. Firms with model risk teams, with data scientists, with compliance functions that have a line item for “responding to regulatory calls for evidence”. Their idea of good AI governance is shaped by the resources they have. When the FCA distils all of that into a good and poor practice publication, the picture of what good looks like will be built from the working practices of organisations that look nothing like a five-person brokerage in Stockport.
And then it gets applied to everyone. That is how this always goes. Guidance written from the evidence of the well-resourced becomes the standard the under-resourced are measured against. Not through any malice, simply because the regulator can only work with the examples it is given. Silence is not neutral here. Silence is a decision to let other people describe your work for you.
So what would a good answer from the mortgage space actually look like? Start with the thing brokers already do instinctively, which is keep a human in the loop. The defensible use of AI in an advice firm is not the machine making the recommendation. It is the machine doing the triage, structuring the fact find, chasing the documents, running an affordability pre-check, and then handing a qualified adviser something better organised than they would have had otherwise. The suitability decision stays with a person who is accountable for it under the Senior Managers regime. That is not a limitation to apologise for. It is the entire reason the model is safe, and it is worth the FCA hearing it described in plain terms by people who do it every day.
Then there is Consumer Duty, which is where this stops being abstract. The honest version of good practice is mapping each AI use case against the four outcomes before it goes anywhere near a client, and writing down where the foreseeable harm sits. An AI tool that drafts a suitability letter touches consumer understanding. A chatbot that fields a first enquiry touches consumer support, and the moment it meets someone in financial difficulty the stakes change completely. Brokers who have thought this through have something genuinely useful to tell the regulator, and the regulator has explicitly asked to be told.
The barriers are just as worth describing, because they are real and the FCA needs to understand them rather than guess at them. Regulatory uncertainty has produced a wait-and-see freeze, where firms that could safely use a tool to cut admin and spend more time advising clients hold back because they cannot tell where the line is. Network principals have responded to that uncertainty with blanket bans on appointed representatives using any AI at all, which is a blunt instrument that destroys legitimate productivity to manage a risk that better controls would handle. Professional indemnity insurers are nervous about anything with AI near it, and cover gaps are starting to appear. None of this comes through unless someone who lives it writes it down.
There is a proportionality point underneath all of it that the mortgage sector is unusually well placed to make. Brokers do not build AI. They buy it. Almost every tool a brokerage uses is a third-party product, which means the realistic question is not “how do you validate your model” but “what does proportionate due diligence on a vendor look like when you are a small firm with no data science function”. If the good practice publication answers that question using the standards of an institution that builds its own systems, it will be useless to the people who most need guidance. If it answers it with input from firms that actually operate this way, it might be the most helpful thing the FCA produces on AI all year.
The FCA describes its own approach as evidence-based rather than prescriptive. That phrase carries a quiet implication that is easy to miss. Evidence-based means the output is only as good as the evidence that goes in. The regulator is not pretending to know what good looks like in a brokerage. It is asking. Whether the answer reflects the reality of small intermediary firms depends entirely on whether those firms can be bothered to say.
There is a version of the next eighteen months where the mortgage advice profession looks at a piece of FCA guidance on AI and recognises almost nothing of how it actually works, and spends years explaining to the regulator why the standard does not fit. There is another version where some of that guidance exists because brokers took an afternoon, before 19 June, to describe what they had already learned. The deadline is the easy part to note. The harder question is whether an industry that complains, fairly, about rules that do not fit its shape will take the one invitation it has been given to help draw them.