AI for accountants mainly speeds up the repetitive work in bookkeeping and the writing around advice and reporting, while the accountant stays responsible for every figure that leaves the firm. The gains are in invoice processing, bank reconciliation and first drafts; the risk is in figures and sources that a model plausibly makes up.
For an accounting firm, that difference matters more than in most professions, because the output falls under professional rules, anti-money laundering legislation (in the Netherlands, the Wwft) and the firm’s quality management system.
What is AI for accountants?
AI for accountants is software that recognises patterns in financial data and text, and on that basis suggests entries, matches or drafts for the accountant to review. It takes over routine steps, such as reading invoices or suggesting a general ledger account; the judgement and final responsibility stay with the accountant.
Two kinds of AI come together in a firm. The first is built into the accounting software and works on structured data. The second is a language model such as ChatGPT or Claude, which writes, summarises and answers questions, but does not post entries and has no access to the records unless you supply them yourself or set up an integration. Claude has connectors on all plans and ChatGPT has plugins with connected apps, which give a chatbot access to the systems you link.
How is AI changing accounting?
AI is changing the work in accountancy mainly by shifting time from data entry to review and advice. According to a study by Choi and Xie in the Journal of Accounting Research (2026), with a survey of 277 accountants and field data from 79 small and medium-sized businesses, more AI use is associated with a shift of about 9% of accountants’ time from routine data entry to higher-value work, and with a month-end close that is 7.5 days shorter.
Those figures are correlational, not a controlled experiment; the researchers themselves leave open whether AI is the cause. They do show where the effect lands: in the turnaround time of recurring work, not in replacing the accountant.
The Dutch professional body for accountants, the NBA, takes the same line. Since 1 January 2026, the NBA has named AI as one of three development themes for continuing professional education (PE) in the period 2026 to 2028, alongside the accountant’s DNA and risks. The theme is explicitly not an additional mandatory PE topic; accountants apply it or explain in their PE portfolio why not.
How do you use AI in bookkeeping?
To use AI in bookkeeping, you let the software suggest the repetitive processing and review the doubtful cases yourself. AI in bookkeeping delivers results fastest, because the work is structured and repetitive and the system learns from earlier processing. Four tasks lend themselves best to support.
Processing invoices automatically
Processing invoices automatically means the system reads the supplier, amount, invoice number and VAT rate from a document and prepares that data in the accounting software. The gain is less manual data entry and fewer typing errors. An invoice is a source document that must remain verifiable, so a low-confidence reading is passed to an employee for review.
Reconciling bank statements
Reconciling bank statements with AI matches incoming and outgoing payments to open items or general ledger accounts, based on amount, description and earlier matches. For a payment that cannot be matched unambiguously, the system makes a suggestion rather than a match. You review those doubtful cases yourself, because a wrong match throws off the whole reconciliation.
Categorising transactions
Categorising transactions with AI produces a suggested general ledger account, even for suppliers the system does not know yet. The more history a set of records has, the more reliable those suggestions become, and the more scattered the data, the weaker the suggestions. A misclassified transaction carries through into the tax return and the annual accounts, so you check the suggestions on a sample basis.
Flagging anomalies
Flagging anomalies is where AI goes further than fixed posting rules. The system compares new entries with what it expects based on earlier periods and marks what stands out: a duplicate invoice, an unusual amount, a missing contra entry. This gives you a targeted checklist instead of a full dataset; reviewing each flag remains your job.
Which accounting software uses AI?
Most accounting packages used by Dutch firms already have AI features built in, such as invoice recognition and automatic bank matching. A separate AI product is usually not needed for the bookkeeping itself.
The table below compares four packages widely used in the Netherlands, with their AI feature and the audience they are most often used for.
| Package | AI feature | Typical for |
|---|---|---|
| Exact Online | Invoice recognition, automatic matching of bank transactions | SMEs in general and accounting firms |
| Twinfield | Invoice recognition and processing across multiple sets of records | Accounting firms |
| Yuki | Highly automated processing, checked by the firm | Firms that receive many client documents |
| Moneybird | Reading invoices and receipts, focused on simplicity | Freelancers and small businesses |
The choice depends on the size and type of client portfolio, not on the number of advertised AI features. More important than the package is that the automation leaves room for a checking step, because each of these packages makes suggestions that someone approves.
What can you do with an AI chatbot in an accounting firm?
An AI chatbot in an accounting firm supports the writing and research around the figures, not the bookkeeping itself. An AI chatbot such as ChatGPT or Claude writes, summarises and answers questions based on what you supply; client data only belongs there if the firm has recorded which account may be used and under which agreements.
Drafting documents and reports
Drafting documents and reports goes faster with a chatbot as the starting point. The model produces a first version of a cover letter, meeting notes or an explanatory note to the annual accounts, which the accountant then checks and refines. The time saved comes from removing the blank page; what the firm sends remains the accountant’s text and responsibility.
Researching technical questions
Researching technical questions with a chatbot is useful but also the riskiest use in a firm. A model can summarise a complex scheme or give a first direction on a tax question, provided it is steered well. Good output starts with how you give an LLM context: the situation, the source and the desired format. You always check an answer about laws and regulations against the source itself.
Where does AI become unreliable in financial work?
AI becomes unreliable in financial work at the points where an error carries furthest: a wrong figure travels through the tax return into the annual accounts and the advice. A model presents an error with the same confidence as a correct answer, and two pitfalls come up most often.
Made-up figures and sources
Made-up figures and sources are the best-known pitfall of generative AI, and the mechanism behind them is called an AI hallucination. In the period when I used ChatGPT intensively, I consistently found that it cited studies, reports and sources that turned out not to exist when checked.
The large firms run into the same thing. According to the Dutch trade site Accountant.nl (10 August 2026), based on reporting in the Dutch financial daily Het Financieele Dagblad, PwC reports contained made-up footnotes and misattributed claims, and EY and KPMG withdrew reports because of content made up by AI. For an accountant, a simple rule therefore applies: a figure or reference you have not traced back to the source yourself does not belong in the file.
False certainty in anti-money laundering screening
Anti-money laundering screening (under the Wwft in the Netherlands) with AI looks attractive, but gives a misleading sense of certainty. A model that flags unusual transaction patterns lacks context the accountant does have, and recognises new structures poorly because it was trained on earlier patterns. The screening helps focus attention, but does not relieve the firm of its own gatekeeper role under the Wwft.
Who is responsible for AI output in an accounting firm?
The person responsible for AI output in an accounting firm is the accountant who signs off the work, not the model that prepared the figures or the text. An algorithm does not appear before a disciplinary tribunal and does not sign a compilation report. The starting point is therefore simple: AI is never responsible, you are.
That responsibility does not shift to the supplier of the accounting software either. A suggested entry accepted without review is an entry by the firm. The risk grows when employees use tools on their own without the firm having agreed anything about it; that unseen use is called shadow AI, and in a firm with a duty of confidentiality it is a direct file risk.
How do you set up AI responsibly in an accounting firm?
To set up AI responsibly in an accounting firm, you record for each task who uses which tool, with which data, and which check goes with it. In the Netherlands, that record fits directly into quality management under the NVKM, the Dutch quality management standard, which has applied to regular licence holders under the Audit Firms Supervision Act (Wta) since 1 January 2026 and applies to other firms from 1 January 2027.
Since 2 February 2025, Article 4 of the EU AI Act has required organisations that use AI professionally to take measures to support their employees’ AI literacy. Since 27 July 2026, under Regulation (EU) 2026/1744, this has been a best-efforts obligation without a guaranteed level per employee, which has made documentation more important: what you can show is the evidence of the measure. AI training for your whole firm records that measure per employee, with a certificate in their name. Two principles from the financial world translate directly into practice.
Applying the four-eyes principle
The four-eyes principle keeps AI output manageable with a second review before anything leaves the firm. From my years at Rabobank, ABN Amro and Van Lanschot, I bring that habit to AI work: nothing goes out without a second check. In a firm, that becomes a fixed step: AI delivers the draft or the suggestion, an employee approves it, and for higher risk a second person also looks at it.
Risk classification per task
Risk classification per task determines how much checking an AI application needs. An internal summary for your own use requires less than an explanatory note to the annual accounts or tax advice that the client acts on. I use AI widely myself, but the checkpoints differ per task, and the same logic works in an accounting firm.
A firm that records which risk applies to each type of task, and which check goes with it, prevents tax advice from getting the same light check as a rough note, and immediately has the overview the NVKM asks for. In the AI Literacy Course, you learn to turn these agreements into an internal AI policy with checkpoints, which you then align with your professional body’s rules (in the Netherlands, those of the NBA).
Where do you start with AI in an accounting firm?
You start with AI in an accounting firm by taking stock of the AI use that already exists, not with a new tool. Employees sometimes already use a chatbot without any agreements; the first question is which tasks that touches, with which client data and under which account.
Then pick one task with a clear gain and low risk, for example invoice processing or a first draft of a standard letter, and build a fixed way of working with a checking step around it. If that works, the next task follows. Start with an inventory, per tool, of what is allowed.
AI for accountants delivers most when the whole firm follows the same way of working rather than each employee having their own approach. With AI literacy training for teams, everyone takes the same course, with a certificate in their own name.
Frequently asked questions about AI for accountants
If you take AI for accountants seriously, you will also have to deal with the rules and publications of your professional body. The answers below relate to the Dutch professional body, the NBA.
Does AI training count towards your PE as an accountant?
For accountants in the Netherlands, AI training counts towards your PE when it fits a learning objective from your own work and you record it in your PE portfolio. The NBA’s 2026 PE regulations set no minimum number of hours and no list of recognised activities, but require learning objectives, completed activities, supporting evidence and a reflection on what you have learned for each calendar year. Before 31 January, you declare whether you met the requirement for the previous year.
Does the NBA have guidance on AI use by accountants?
Yes, the NBA has guidance on AI use by accountants: in June 2026, AI guidance 2, AI Toegepast (AI Applied), was published by NBA Accounttech in collaboration with NOREA. The guidance discusses preconditions and validation of AI use and gives practical examples for the SME accountant, such as bank reconciliation and compiling annual accounts. A separate chapter covers examples for the external auditor and IT auditor, ordered by auditing standards.
