Definition

Evidence-led AI accountancy, defined.

Evidence-led AI accountancy is the practice of using AI to surface and organise the evidence behind financial statements, on the condition that every material figure remains traceable to its source and a named, qualified human reviews and signs the result. The machine does the labour of review. The human does the owning. In one sentence: AI can surface the evidence; the partner still signs.

The term exists because the market split into two camps and both are half right. The sceptics say generative AI produces plausible output rather than correct output, so keep it at arm’s length. The enthusiasts say the models can already draft the accounts, so adopt or fall behind. Both skip the question a regulated profession actually turns on: not whether the machine can produce the answer, but whether the answer can be traced back to the evidence, and who takes responsibility when it cannot.

Evidence-led AI accountancy answers with a standard rather than a mood. Its mechanics, as we run them inside Jack Ross:

The Review Gap

For a decade the profession automated preparation (bookkeeping, bank feeds, coding, draft accounts) and left review, the conversion of prepared work into signed work, almost untouched. That unaddressed half is where lock-up, write-offs, quality variation and the partner bottleneck live. Evidence-led AI aims squarely at it.

The Trace-to-Source Standard

Every number in a file occupies one of five states: CLAIMED (asserted), CITED (attributed), PINNED (tied to the exact file, sheet and cell), PROVEN (tested against the underlying records and the framework, FRS 102 and the Companies Act for most UK files), and SIGNED (a named, qualified person owns it). The machine’s job is to climb numbers up the ladder. Only a human can take the last step, because only a human can be accountable.

The one-question test

For any AI touching professional work: can you point to the cell? A finding that cannot be traced to source is not a review point. It is a guess with good grammar, and it is your name on it, not the software’s.

What it is not

It is not automated sign-off; accountability does not transfer to a tool. It is not a copilot beside a tired reviewer who is free to trust it as much as they like, which on a busy Friday means too much. And it is not anti-AI; used inside a governed workflow, the technology makes numbers more trustworthy, not less, because the evidence trail becomes the output.

Common questions

Who developed the approach?

The standard described here was developed by Umar Memon, Managing Partner of Jack Ross Chartered Accountants (Manchester, established 1948), and is set out in full in The Signed Review.

Does it slow the work down?

It relocates the labour: the machine carries figures up the ladder around the clock; the human hours move to judgement.

Does it reduce headcount?

No; it changes the apprenticeship. Trainees learn judgement by interrogating findings rather than grinding preparation. We did not lose anyone to AI. We moved them up to it.

The full standard, an operating model and a 90-day implementation plan are free in The Signed Review. Weekly notes in The Partner Still Signs.