Use case assessment
Are AI meeting summaries worth building?
Genuinely AI, and almost never worth building. This is a buy decision with a consent problem attached.
Real AI, well solved by existing products, and the interesting question is not whether to build but who is being recorded and whether they agreed.
What it usually means
Transcribing a meeting and producing a summary, decisions and action items.
Is it genuinely AI?
Yes, on both counts. Speech recognition across accents, overlapping speakers and poor audio is a model problem. So is deciding what mattered in an hour of conversation.
It is also a solved problem with mature products, which changes the question from should we build this to should we buy this, and what happens to the recordings.
What the simpler version looks like
There is a version worth naming before any tooling: someone writes down the decisions at the end of the meeting.
That sounds glib and it is not. Teams that adopt summarisation tools to fix meetings where nothing is decided usually end up with an accurate record of nothing being decided. The tool is a good answer to a recall problem and a bad answer to a facilitation problem, and the two get confused reliably.
What it costs to run
Per-seat pricing on a mature product, which is predictable and rarely the issue. The cost that surprises people is storage and retention: transcripts of every internal meeting are a substantial and searchable body of records that carries data protection obligations, and that becomes discoverable in a dispute.
Deciding a retention period before rollout is much easier than deciding one after two years of accumulation.
Where it sits under the EU AI Act
Not an Annex III high-risk use. But two things do apply.
Article 50 disclosure where the system interacts with people — and enforceable since 2 August 2026. A meeting bot that joins a call should identify itself.
Emotion inference is a prohibited practice in workplace contexts, in force since February 2025. Several meeting products offer sentiment or engagement analysis. That feature, in an employment context, is on the wrong side of the line, and it is usually switched on by default.
The GDPR question is separate and larger: recording colleagues, and external participants, requires a lawful basis and usually notification.
When it is worth it
When meetings have genuine decisions that genuinely get forgotten, when participants know they are being recorded, and when someone has decided how long transcripts are kept.
Not as a substitute for someone owning the outcome of a meeting.
Common questions
Should we build our own meeting summarisation?
Almost never. This is a mature product category with strong existing options, and the differentiating work is not in the model. The real decisions are about consent, retention and which features you switch off.
Do AI meeting assistants need to disclose themselves?
Yes. Article 50 requires disclosure when a person is interacting with an AI system, and that has been enforceable since 2 August 2026. A bot joining a call should identify itself.
Is sentiment analysis in meeting tools allowed?
Emotion inference in workplace contexts is a prohibited practice under the EU AI Act, in force since February 2025. Several meeting products offer engagement or sentiment scoring, often on by default, and in an employment context that is on the wrong side of the line.
Scoring 40 AI use cases in a regulated enterprise — the kind of default-on feature that turns a low-risk tool into a governance problem.