AI for statutory accounts and accounts production

Key facts

As at 25 October 2026.

Where AI helps

• Mapping the trial balance. Bookkeeping charts of accounts differ between Xero, QuickBooks, Sage, FreeAgent and spreadsheets. AI can read account names and codes and suggest the line item: trade debtors, other creditors, administrative expenses. A person confirms each suggestion.
• Drafting notes and reports. Accounting policies, the directors' report and, for medium-sized and large companies, the strategic report are mostly wording around facts. AI can draft them from the company's activity, the figures and last year's text, as editable drafts.
• Reading documents. Last year's accounts as a PDF can be read for the comparatives, which are then checked against the filed accounts.
• Spotting inconsistencies. A note that does not match the statements, a missing disclosure for the size class, or a balance in the wrong direction.

Where AI should not decide

• Amounts. Totals, subtotals, tax, comparatives and the notes' figures should come from deterministic calculations that give the same answer every time, so the accounts reconcile and can be reviewed.
• Rules. The company size, the framework, the formats and the statements above the signature follow from the Companies Act and the standards, not from a language model's guess.
• Judgements. Going concern, provisions, impairment, valuations and events after the year end are decisions for the directors. AI can set out the facts; it cannot take the responsibility.
• Approval. The board approves the accounts and a director signs the balance sheet. No software can do either.

How StatutoryAccounts.io uses AI

In StatutoryAccounts.io the AI suggests the line item for each nominal account and writes editable drafts of the notes, the directors' report and, where required, the strategic report. The figures come from fixed calculation rules, not from the language model. You confirm the mapping, edit the drafts and approve; every change and approval is recorded. Your data is stored in London (Supabase, eu-west-2), and the AI features send only the data they need to Anthropic's commercial API, which by default does not use API data to train its models.

Checks before you approve AI-drafted accounts

1. Does the trial balance balance, and does it match the bookkeeping at the year end, after all year-end journals?
2. Is every nominal account mapped to the right line item, including the director's loan account, VAT and PAYE balances?
3. Do the comparatives agree with last year's filed accounts?
4. Does every note describe what the company actually does, rather than generic wording?
5. Are the going concern note, the size class and the statements above the signature right for this company and this year?

The small company accounts example shows what a reconciled set looks like.

Related pages