An AI colleague for your accounting practice: repeat questions and documents
An accounting practice gets the same questions every month from different people: where do I submit this, what do you still need, has my return been filed. Nova absorbs those repeat questions, reminds clients of missing documents and files things in the right place. What it does not do is assess — that is your profession.
Published on · Last updated on
The problem in this sector
Filing peak: clients keep emailing "is my filing in yet?" No time to call everyone.
| Aspect | For this sector |
|---|---|
| Channels | E-mail · WhatsApp |
| Own knowledge | Each client gets their own files and statuses, strictly separated |
| Deployment | Nova Vault |
| Expected outcome | The client gets the status of their own file back without the accountant in between. Calls only for real questions. |
How far your sector has come
Of all sectors on this site, this one sits closest to the leading group. CBS counts accountancy under specialist business services, which together with financial services uses AI relatively often — clearly above average, though not reaching the 54 percent of information and communication. In practice: your competitors are already working on this, and the question is not whether but for what.
Where the limit sits
Here a constraint applies that other sectors do not have: the statutory tax retention obligation. The Dutch Tax Administration states that the retention period only begins once a record's current value lapses — not on the invoice date. An AI that clears out on the basis of age is therefore structurally wrong. Nova does not clear anything from the administration; it processes messages, and those have their own, shorter retention.
Concretely, these are the things Nova does not do on its own here:
- Giving tax advice or classifying an entry; it flags what is missing, you determine what it is.
- Deleting documents from the administration — the retention period hangs on current value, not age.
- Filing or approving a tax return, not even when everything appears complete.
That limit does not live in a setting someone can flip by accident; it is established up front. That is the difference between a system you know the behaviour of and one you find out about afterwards.
What the anti-money-laundering rules keep out of an assistant’s hands
The Dutch anti-money-laundering act (Wwft) draws a line here that other sectors do not have. Client due diligence under article 3 and identity verification under article 11 belong to a person in the practice. An assistant can request documents and send reminders, but cannot establish that someone is who they say they are.
Heavier still is the tipping-off ban. About a report of an unusual transaction to the Financial Intelligence Unit you stay silent, including towards the client (article 23 Wwft). An assistant allowed to search the whole client file could quote such a report by accident. So the Wwft file stays outside its sources.
And here the conversation is often part of the records itself. Business chats fall under the seven-year retention duty in article 52 of the Dutch tax act, WhatsApp included.
The full working-out sits in AI at a bookkeeping or financial services practice, with a per-message breakdown of what the assistant may and may not see.
The general rules on automated decision-making and transparency are set out in AI and the GDPR for SMEs and the AI Act for SMEs.
What we need from you
This differs per sector, so it sits here rather than in a generic checklist:
- Which systems you use and how clients submit today — portal, email, or both.
- Your submission deadlines per client type, because that is what most repeat questions are about.
- Which questions you deliberately want to keep answering yourself; with advisory questions that is more often than you think.
How a project runs — intake, analysis of your existing communication and a shadow week in which nothing is sent yet — is set out on the approach page.