Can AI prepare the annual accounts?

AI can prepare an annual account up to the point of professional judgement, and Dutch practices are already using it, mostly for the wrong tasks.

It can do the preparation: check the ledger for miscodings, reconcile bank, VAT and payroll, draft the notes and the first version of the statements in your own template. The accountant keeps the judgement, the going-concern question and the signature. This guide says which parts of the compilation file AI takes over, what the NBA expects, and what a practice pays.

By Tom Joseph · Last updated: 14 September 2026

Free consultation Pricing

In short

AI can prepare an annual account up to the point of professional judgement, and Dutch practices are already using it, mostly for the wrong tasks. 78% of Dutch accountancy firms use AI, but only 24% in compliance and control work, the heart of the practice, against 37% for text and communication (Exact MKB Barometer, Q1 2026). The NBA and NOREA published a joint guideline on AI in the accounting and IT-audit practice on 1 June 2026. In the sector overall, 39.8% of specialist business-services firms with 10 or more employees used AI in 2024 (CBS AI Monitor 2024). What the client signs is still the accountant's: the system prepares, the professional judges.

Who does what

The compilation file, split between AI and the accountant

Step	AI	Accountant
Ledger review	Flags miscodings, duplicate invoices, unusual journal entries, missing VAT codes	Decides the corrections
Reconciliations	Bank, VAT return, payroll, intercompany and fixed-asset reconciliations, with differences listed	Explains the differences that remain
Provisions and estimates	Proposes from history and policy: bad debts, warranty, holiday pay	Judges the estimate
Draft statements	First draft of balance sheet, profit and loss and notes in the firm's template, in the RJ or micro/small format	Edits, applies the standard, signs the compilation report
Going concern and continuity	Shows ratios, covenants and cash trends	The judgement, entirely

Where it fails

What AI cannot do in an annual account

It cannot exercise professional judgement, and under Standard 4410 the compilation report is the accountant's. It cannot know what is not in the books: the verbal agreement with the bank, the dispute with a customer, the director's plans. And it produces a plausible note as easily as a correct one, which is why every figure in the draft traces to a ledger line or a document and the reviewer sees the trace, not a text. The NBA's core values apply unchanged; the tool changes how fast the file is ready for judgement, not who is responsible for it.

The systems matter: the AI layer reads from and writes to Exact Online, AFAS, Twinfield, Unit4 and your compilation software, or the assistant retypes.

Cost

What it costs a practice

Crux Digits scopes it in three fixed steps: a €2,500 audit on twenty of last year's compilation files, which shows where the hours go per file; a €20,000 Production-ready MVP that runs ledger review and reconciliations for one client segment against your own files; and production from €50,000, integrated with your bookkeeping and compilation software, with the reviewer's sign-off in the workflow. Excl. VAT, all code handed over, and the firm's own data never leaves the firm's tenancy. See AI for accountancy.

Buyer guides

Compare further

Excel automation: out of spreadsheets A private AI assistant over your documents How to automate time tracking and invoicing

FAQ

Frequently asked questions

Can AI sign or replace the compilation report?

No. Under NBA Standard 4410 the compilation report is the accountant's professional statement, and the NBA and NOREA guideline of June 2026 keeps responsibility with the professional. AI prepares the file; the accountant judges and signs.

Which part of the annual account does AI do well?

The mechanical part: ledger review for miscodings and duplicates, bank, VAT and payroll reconciliations, proposals for routine provisions, and the first draft of statements and notes in the firm's template, with every figure traced to its source.

Is client data safe in an AI system?

Only when the system runs in the firm's own tenancy, does not train on client data and logs every access; a public chatbot is neither. The Dutch Data Protection Authority reported dozens of breaches in 2025 from staff entering data into public AI tools.

Does it work with Exact Online, AFAS or Twinfield?

Yes. The AI layer reads the ledger and writes proposals back into those systems and into your compilation software; integration is scoped in the audit.

What does AI cost an accountancy practice?

At Crux Digits: a €2,500 audit, a €20,000 Production-ready MVP for one client segment, and production from €50,000, all excl. VAT, with the code handed over.

Want to know where the hours go in a compilation file?

Book a free consultation. Bring three files from last season and we will tell you honestly which steps AI takes over and which it should not.

Book a free consultation