Field note · Advice & compliance
Can an IFA use AI to write suitability reports without breaching Consumer Duty?
The question is not whether the model can write it. It obviously can. The question is whether you can show, eighteen months later, who decided what — and that is a records question, not a model question.
The short answer
Yes — but the model must draft, never decide. A defensible setup has four things: the adviser signs the recommendation, every source the draft used is retrievable, client data stays inside controls you can evidence, and the file shows what was changed between draft and sent. Absent those, you have not saved time. You have moved the risk.
Every adviser has now been shown the same demo. A fact-find goes in, a suitability report comes out, and the room does the arithmetic on paraplanner hours.
The figure usually quoted is four to six hours per report. It comes from vendors selling suitability report generators — it is a vendor claim, not independent research, and we could find no regulator or trade body figure to corroborate it. Treat it as the seller's number. It is still probably the right order of magnitude, which is exactly why the question deserves a serious answer rather than a demo.
What is the actual regulatory risk?
Not that a model writes badly. Models write fluently, which is the problem.
A suitability report is a record of a decision. Consumer Duty asks you to evidence that the outcome was good for the client. If the reasoning in the file was generated rather than reasoned, and nobody can now say which parts the adviser actually held a view on, you do not have an evidence problem in the future — you have one already.
What has to be true before it is defensible?
1. The adviser signs, and the file shows they changed something
A file where every AI draft went out untouched is a file that says nobody read them. Keep the diff between draft and sent. It is the cheapest possible evidence that a human was in the loop, and it costs nothing to retain.
2. Every source is retrievable
If the report says the fund's ongoing charge is 0.62%, you need to be able to produce the document that said so, on the date it said it. A draft that cannot show its working is a liability wearing the clothes of an efficiency.
3. Client data stays inside controls you can evidence
This is a data protection question before it is an FCA one. Where does the fact-find go, who processes it, where is it stored, for how long, and can you show that to a client who asks? "It goes to a model" is not an answer to any of those.
4. The failure mode is visible, not silent
The dangerous output is not the one that breaks. It is the one that reads perfectly and is wrong about one number. Your check has to be on the facts, not on the prose — because the prose will always be fine.
What does a sensible first step look like?
Not the whole report. Take the section that is genuinely clerical and has a checkable output — the factual summary of current arrangements — and automate only that. It is the part with the most re-keying and the least judgement.
| Section | Judgement load | Sensible to draft? |
|---|---|---|
| Current arrangements summary | Low — it is transcription | Start here |
| Objectives as stated by the client | Low, but must be verbatim-faithful | Yes, with the source quoted |
| Product / provider comparison | Medium — selection is a view | Draft the table, own the shortlist |
| Attitude to risk conclusion | High | No |
| The recommendation and why | High | No |
| What was considered and rejected | High — and this is what gets tested | No |
Is anyone actually being fined for this?
Not that we can find, and we looked. That is not reassurance — it is early. The absence of enforcement in a new practice tells you the sample is small, not that the practice is safe.
The defensible position is the boring one: use it where the work is clerical, keep the human decision explicit, keep the trail, and be able to explain the setup in one page to someone who is not technical. If you cannot write that page, you are not ready to use it on client files.
Questions people actually ask
Can AI write a suitability report for a UK financial adviser?
It can draft one. The adviser must still select the recommendation, judge suitability, and sign the file. A defensible setup keeps the human decision explicit, retains the sources the draft used, keeps client data inside evidenced controls, and preserves the difference between the draft and what was sent.
How long does a suitability report take to write manually?
Four to six hours per report is the figure most commonly quoted. It originates from vendors selling AI suitability report generators rather than from independent research or a regulator, so treat it as a seller’s number.
What is the biggest risk of AI-drafted suitability reports?
A fluent report that is wrong about one fact. It will not look like a failure to a reviewer, because the prose is fine. Checks therefore have to be on the underlying facts and their sources, not on the readability of the output.
Sources
- AdvisoryAI, AI Suitability Report Generators for UK Financial Advisers: 2026 Buyer’s Guide, 18 May 2026 — the four-to-six-hours-per-report figure. — vendor selling the fix; uncorroborated re-checked every 6 months
- FCA Consumer Duty, in force 31 July 2023 — the four outcome tests that make this a records question. source ↗ — regulator, primary re-checked every 6 months
Checked 8 September 2026. Next scheduled check 7 March 2027. Numbers that move — leaderboards, live indices — are re-checked every 30 days; annual datasets and rules in force every six months; dated research once a year. If something here has gone stale before we got to it, tell us and we will correct it and say what changed.
Want the clerical half automated and the judgement half left alone?
We scope the sections, build the checks on the facts rather than the prose, and hand you the one page you can show a compliance reviewer.
Talk to us about this