Back to the blog

The NBA strategy for 2027-2030 on AI: what an accounting firm can settle now

· By Bee Flow · 5 min read

The Dutch accountancy body NBA made AI one of three themes in its strategy for 2027-2030. The choice and introduction of tools stays with the firm. What the strategy says, what the June guidance already asks, and five things to check this month.

A case file with a magnifying glass over one highlighted line

On 6 November 2026 the NBA, the Dutch professional body for accountants, published its strategy for 2027 to 2030 under the title Vertrouwen en veerkracht (trust and resilience). It is one of the documents for the members' meeting of 14 December in Nieuwegein. There is no vote on it that day, Accountant.nl reports. The board explains the strategy and members discuss how to fill it in. One of its three themes is called "Accountant in het AI-tijdperk", the accountant in the age of AI.

For a firm of ten or forty people the strategy holds no new obligation. It does say who has to do the work.

What the strategy says about AI

The NBA chooses, within the wider field of technology, to focus on AI, and says it will help accountants in every field of work to use AI and AI-generated data responsibly.

Two passages matter for a firm. The first is about responsibility. The strategy lists what can go wrong: wrong or incomplete output, privacy, information security, legislation, bias and a lack of transparency. Its conclusion is that accountants may not simply rely on AI technology or on data made with it. They stay responsible for using it correctly, for the quality of their work and for their own professional judgement.

The second is about who does what. Practical applications and help with introducing specific systems are, in the strategy's words, in the first place the responsibility of accountants themselves, their organisations, the training institutes and the service organisations. The NBA provides the shared base: overview, interpretation, dialogue, standards and the exchange of knowledge. Where needed it will draw up professional frameworks.

So nobody is going to tell your firm which tool to use or how to set it up. That decision, and the reasoning behind it, is yours.

The practical demands are already on paper

The strategy stays at the level of direction. The detail is in a document that has been out since June: AI-leidraad 2: AI Toegepast, written by NBA Accounttech and NOREA, the body for IT auditors. It is guidance for members and no law, and it is concrete. Chapter 2 of the document lists the conditions. Four of them reach straight into the daily work of a firm.

Confidentiality comes first. Client data and personal data may not just be entered into public AI models, and the guidance names the free versions of ChatGPT and Gemini as examples. That data has to stay in a secured, closed-off environment.

Then policy. A firm has to put its requirements for the use of AI into its quality management system. The choice for an AI application has to be argued, and every application gets its own risk analysis. The guidance adds a sober note here: a plain script is regularly cheaper and more effective than AI.

Sources are the third. An AI application should only draw on authorised, current and reliable sources. The guidance mentions retrieval from your own documents as a way to secure that, and asks that answers refer to the sources used, so the professional can open the main ones and check whether the content was rendered correctly and nothing material was left out.

The fourth is the file. The prompts used, the sources consulted, the verification steps and the conclusions go into the file, so that someone can later trace how AI output was assessed and used. The firm keeps a catalogue of validated prompts.

One warning in the guidance deserves a place on the office wall: language models are not calculators and can get simple sums wrong. A calculation that matters gets checked.

Five things to check this month

  1. Which AI tools are in use today, including free versions on private accounts. Ask it without blame, or you will not hear the answer.
  2. Where client data goes when someone uses those tools, and whether there is a data processing agreement with each supplier.
  3. Whether your AI policy is written down and part of the quality system. A single page is enough to start with. We described one in a one-page AI policy for a small company.
  4. Whether a colleague can open the source behind an AI answer. If not, the check the guidance asks for cannot be done.
  5. What goes into the file when AI was used, and who checks the calculations.

The AI Act also asks employers to make sure staff who work with AI know enough about how it works and where it fails. The guidance repeats that duty. Our post on AI literacy under Article 4 explains what it comes down to.

We would not wait for the frameworks the NBA may still draw up. The guidance is concrete enough to work with now, and the questions above will be the same ones a reviewer asks later.

This post is no legal advice and no explanation of the professional rules. For those, read the NBA documents themselves.

Where Bee Flow fits

Bee Flow is an AI workspace for a whole office that runs on your own server or on Bee Flow Cloud in the EU. You decide which model does the work, including a model hosted in the EU or one that runs locally on your own hardware. Privacy Shield blocks personal data before a message reaches the model, in every licence. A knowledge base answers from your own documents and shows its sources as cards that open the original document, which is the check the guidance describes. An agent holds one agreed way of working with fixed instructions, so colleagues share an approved prompt and do not each invent their own.

None of this takes over the accountant's judgement, and a tool does not make a firm compliant by itself. It does give you a closed-off place to work in and answers you can trace. The page for accounting firms shows how that looks in practice, privacy and security describes where data goes, and you can plan a call if you want to walk through your own situation.

#accountancy #ai-policy #netherlands