AI file review tools for UK advice firms: what they check, what they can't
AI file review tools score what a file already contains against a fixed rubric — and flag the gaps. What they cannot do is replace the human judgement on what the file should contain.
What a rubric-scored tool checks: the 11 check areas — suitability reasoning, vulnerability consideration, cost disclosure, alternatives, capacity for loss. Each gets a Pass / Amber / Fail with the rule behind it, line-referenced.
What it cannot do: decide the vulnerability judgement, write the rationale, or own the recommendation. That is human — and the file must show a human signed off. The tool flags; you judge.
The honest summary: AI makes every file a review at a price a day-rate consultant cannot match — but the evidence it produces is only as good as the human who reviews the flags and signs the file.
What the FCA says about oversight of file review
The Consumer Duty makes the firm accountable for the outcomes clients get, not just the products it sells. The FCA's Consumer Duty pages set out the four outcome areas and their monitoring. A review tool sits inside that accountability — it scores files, but the firm owns the outcome.
The FCA's advice market survey reports that 88% of retail clients receive ongoing advice — advice at a scale that runs on process. A rubric makes the review repeatable: the same checks, the same bands, every time.
What a rubric-scored tool actually checks
| Check area | What the tool looks for | The typical flag |
|---|---|---|
| Suitability reasoning | A rationale tied to the client's stated objectives. | Past performance quoted instead of client objectives. |
| Risk and capacity for loss | Both recorded, separately, and consistent with the advice. | A single “medium risk, medium capacity” line. |
| Cost disclosure | Total costs in pounds, initial and ongoing. | “Fees apply” with no figure. |
| Vulnerability consideration | Considered, with outcome and reasoning. | No mention where the circumstances suggest it. |
Worked example: the same recommendation, two files
The gap is rarely the advice; it is the file that fails to show why it was right.
Before
“Recommended a growth portfolio to meet the client's objectives.” That is the whole rationale — no risk test, no capacity check, no alternatives, no charges.
After
“Objectives: growth over ten years, capital preservation secondary. Risk profile 5 of 10; capacity tested against a 15% fall, income and savings drain shown. Alternatives: cash and a lower-risk fund, rejected for these reasons. Total first-year cost £1,240, 0.6% ongoing, impact on returns illustrated.” The tool flags two Ambers; the human resolves them, and the file carries a dated sign-off.
The tool makes the gap visible; the human decides whether it is closed.
Common mistakes in AI-assisted file review
- Scores shipped without human review. A tool output is not a sign-off; the file must show the human signed.
- A full Pass treated as “the file is done”. The rubric only sees what is in the file — a file can pass every check and still be wrong for the client.
- Flags without a linked rule and line. A flag without the rule and line cannot be actioned or defended.
- Un-anonymised client data uploaded. Data protection happens before the tool does — the anonymisation protocol keeps both the review and the data safe.
Related reads
The tool is one step in a loop. See the file review and suitability hub for the full loop, the anonymisation protocol for the gate before a file meets a tool, and the 12-point file-review checklist for turning flags into a signed file.
Frequently asked questions
Will an AI tool replace the human reviewer?
No. The tool flags what the file does and does not contain; the judgement stays with the human, and the file must show a human signed.
What does a rubric-scored tool actually check?
It scores the file against a fixed rubric — reasoning, risk and capacity, cost disclosure, alternatives, vulnerability — with a rule reference behind every Pass, Amber or Fail.
Is it safe to put client files through a tool?
With the anonymisation protocol in place, yes: identifiers are removed, amounts rounded, and the reviewable content survives.
Your next step
Run one anonymised file through a rubric before you buy anything. Proven Duty scores files, line-references every flag, and records the sign-off. Start a free trial or see pricing.