Relying on AI outputs: written reliability decisions, named surveyors and dip sampling

The RICS AI standard requires a written, named decision on whether an AI output can be relied on — and randomised dip sampling for high-volume use. What the written decision must contain.

Can you rely on what the AI produced? Under section 4.2 of the RICS AI standard, that question gets a formal answer: for AI outputs with material impact, members and firms must apply professional judgement — knowledge, skills, experience and professional scepticism — and document the reliability decision in writing.

What the written decision must contain

• any relevant assumptions made,
• key areas of concern about reliability, including the underlying datasets,
• the reason for each concern,
• whether anything could be done to lessen each concern, and
• the impact of the concerns on overall reliability — concluding whether the output can reasonably be used for its intended purpose.

And the accountability clause that gives it teeth: the written decision must be prepared by, or under the supervision of, an appropriately qualified and named surveyor who accepts responsibility for its use. Not “the team checked it” — a name.

When the answer is no

If the conclusion is that an output cannot reasonably be used for its intended purpose, the standard requires that conclusion to be communicated to the client in writing, with the reasoning or a summary. That letter is the natural companion to the advance disclosure notice — transparency on the way in and on the way out.

High-volume use: dip sampling

Where AI automates an output or produces outputs in volume, the standard is pragmatic: scrutinising every output is “generally neither necessary nor proportionate”. Instead, firms must run randomised dip samples at regular intervals — because accountability for each output remains with the firm even when review-by-exception is the method. A recorded sampling plan (what percentage, how selected, who reviews, what happens on a failure) turns that sentence into evidence.

Does every AI output need a written reliability decision?

Material-impact outputs do. For automated or high-volume use, the standard substitutes regular randomised dip sampling for per-output decisions — but the firm remains accountable for every output either way.

Who counts as 'appropriately qualified'?

The standard doesn't define a grade — it requires the person to be appropriately qualified for the work in question and named, accepting responsibility. In practice: the responsible surveyor for the instruction, applying the same judgement they would to a junior's work.

This article is general information, not legal or professional advice. ComplyQS is not affiliated with or endorsed by RICS. Related guide: /guides/project-record/.