In this article
Whoever takes the minutes isn't really part of the conversation. In a meeting of six, one person is mostly typing. AI minutes solve that: the technology works well in Dutch and English, you don't need an implementation project and you see the result straight away.
What almost everyone skips is the legal side. If you record meetings as a matter of routine, you are processing personal data of employees and customers. That requires a legal basis, transparency and a retention period.
What do AI meeting minutes deliver?
- A transcript of what was said, word for word.
- A summary per topic.
- Decisions and action points, with good tools including a name and a date.
- A searchable archive: finding what you agreed with a customer three months ago.
That last one is often underestimated. Over time, a searchable archive of customer conversations is worth more than the time you save per meeting.
Where does it go wrong?
Names and jargon
Names, company names and technical terms are regularly misheard. Most tools let you provide a word list with the names of employees, customers and products. Do it.
Nuance disappears
This is the real risk. Someone says: "we could consider switching". The summary says: "decided: switch". A language model summarises, and summarising means leaving out nuance. A wrongly recorded decision that circulates unchecked takes on a life of its own. More on this kind of error in preventing AI hallucinations.
Poor audio
One microphone in a room with eight people produces an unusable transcript. In physical meetings the microphone determines the quality, not the tool.
Sending without reading
Sending the summary round immediately is tempting. Put one check in between: the chair reads it before it goes out.
The legal side
This is the difference between "we use a handy tool" and "we've arranged it properly".
In the Netherlands, recording a conversation you take part in is not a criminal offence. But that says nothing about your obligations as an employer. As soon as you systematically record and process conversations with employees and customers, the GDPR applies. Arrange the following:
- A legal basis. Usually legitimate interest, with the balancing test written down. Employee consent is shaky, because of the employment relationship.
- Transparency. Participants must know they are being recorded and summarised. In customer calls, say so at the start, not in the small print.
- A retention period, enforced technically too. For example: delete audio and transcript automatically after thirty days and keep only the approved minutes.
- A data processing agreement with the tool's supplier.
- An entry in your record of processing activities.
- Employee participation. Routinely recording conversations at work can fall under the consent rights of the works council (ondernemingsraad, OR). Involve them early.
Put this in your AI policy rather than agreeing it team by team. See drawing up an AI policy and the AI Act and the GDPR: what you need to arrange.
Take extra care with performance and absence reviews, conversations involving special categories of personal data and conversations covered by a confidentiality agreement with a customer. For those, the standard agreement is usually: don't record.
How to choose a tool
Transcription quality is close between the main providers. Choose on these points:
- Processing location: does the audio stay within the EU?
- Training on your data: is it off by default, and is that in the contract?
- Retention and deletion: can you set automatic deletion?
- Integration: is it part of your existing meeting environment, or a separate tool with its own account?
- Word list: can you add names and technical terms?
- Speaker recognition: needed as soon as you want to link action points to people.
- Management: can you see centrally who uses it and revoke access?
The first three matter most, and that's where providers differ most. A separate tool that employees sign up for themselves is a different matter for data processing than the built-in feature of your meeting environment. What's possible in Microsoft 365 is on Microsoft 365.
How to roll it out
- Legal first: legal basis, information notice, retention period, data processing agreement and, if relevant, the works council. Only then the tool.
- Start with internal meetings without sensitive topics. That way you find the quality problems without risk.
- Agree who checks. By default, the chair. Without a fixed reviewer, unread summaries do the rounds.
- Expand where allowed. Customer calls only with notice in advance, and with the exceptions explicitly excluded.
From minutes to action
A minutes tool produces a record. If you also want action points to land in your project system or CRM, or the follow-up email to be ready, that's a step further. How an agent handles that is in AI agent use cases for SMEs.
The technology is quick to set up. The legal basis, transparency and retention period take more thought, but only once. Do that thinking up front. A business that has to explain afterwards why two years of recorded staff conversations sit on a server outside the EU has a bigger problem than the time it saved.
Sources: the Dutch Data Protection Authority on monitoring employees and on algorithms and AI. This article is general information, not legal advice. Have the legal basis and employee participation reviewed if you plan to record conversations routinely.
Frequently asked questions
Can AI take minutes of a meeting?
Yes. Transcription works well, and summarising a transcript into decisions and action points is where language models are strong. Quality depends mostly on audio quality. Always have the summary checked before it goes round.
Can you record a meeting for AI minutes?
In the Netherlands, recording a conversation you take part in is not a criminal offence. As an employer, though, you are processing personal data: you need a legal basis, must inform participants and must set a retention period. With routine recording, the works council may have a say.
What goes wrong with AI minutes?
Names and technical terms are misheard, nuance disappears so a proposal is recorded as a decision, and summaries are sent round unread. That last one is the biggest risk.
Where are the recordings stored?
That differs per provider and is the most important question when choosing. Built-in features of your meeting environment usually keep data within it. Separate tools send audio to their own servers, sometimes outside the EU. Ask about location, retention and training on your data.
How long can you keep meeting recordings?
The GDPR sets no fixed period, only that you keep data no longer than needed. For minutes, the purpose is met once the record is approved. A workable rule is to delete audio and transcript automatically after thirty days and keep only the minutes.
