Meeting transcript example: three speakers, decisions and actions

This is an extract from a weekly operations meeting, clean verbatim with timestamps. Three speakers, one interruption, one decision and one action. Read it, then the notes underneath.

[00:12:40] TOMAS BERG: The second warehouse slot is the thing I want to settle today, because we keep pushing it.

[00:12:47] AUDREY LIN: We have pushed it three times.

[00:12:49] TOMAS BERG: Three times, yes. So — Nadia, what does the volume actually look like for November?

[00:12:56] NADIA HAQ: November is up about a fifth on last year, and December is going to be worse. The current slot does not cover it. We either take the second slot or we start turning orders away in the last week of November.

[00:13:18] AUDREY LIN: [overlapping] And the cost of the second slot is—

[00:13:19] NADIA HAQ: [overlapping] —about the same as the first, give or take.

[00:13:22] AUDREY LIN: Then I do not think it is a difficult decision. Turning orders away in December costs more than the slot.

[00:13:31] TOMAS BERG: Agreed. We take the second slot. Nadia, can you get the contract to Audrey before Thursday so she can look at the termination terms?

[00:13:41] NADIA HAQ: Before Thursday, yes.

Look at 00:13:18 and 00:13:19. Audrey starts a question and Nadia answers it before she finishes. Both lines are marked [overlapping] and the em dashes show where each one was cut off and picked up.

This is the single place automatic transcription most often fails in meetings. Most systems assign an overlapping stretch to whichever voice is louder, which here would have given Audrey's half-question to Nadia and made the exchange read as one person talking to themselves.

The practical fix: after generating a transcript, find every point where a turn is under three seconds or ends mid-sentence, and listen to those spots specifically. In a one-hour meeting that is usually a dozen places, not the whole hour.

If people dialled in, expect worse results for them than for people in the room. Compressed call audio loses exactly the frequencies that speaker separation relies on.

A transcript is a record, not a summary, and almost nobody reads one end to end. The decision and the action in that extract are at 00:13:31 and they are the reason the meeting happened. They should appear twice: in the transcript where they were said, and in a short block at the top.

A decision block records what was decided, who decided it and the timestamp: "Take the second warehouse slot — agreed by Tomas Berg at 00:13:31." The timestamp is what makes the transcript worth keeping, because anyone who disputes the decision later can hear it.

An action needs an owner and a date, both taken from what was actually said rather than what you assume was meant. Here: "Nadia Haq to send the contract to Audrey Lin before Thursday." If the audio does not contain an owner or a date, the action block should say so instead of inventing one — that gap is useful information about how the meeting ran.

Small talk before the meeting starts and after it ends is usually cut, with a marker showing where the record begins and ends. Be consistent about it, because selectively cutting the beginning of some meetings and not others is the sort of thing that looks deliberate in hindsight.

Someone asking for a section to be off the record should get it, and the transcript should show that a gap exists: "[00:24:02–00:26:40 not transcribed at the request of the speaker]". A silent gap and a marked gap are very different documents.

Do not tidy what people said to make them sound better. Clean verbatim removes fillers, not opinions. If a transcript will be read by the people in it, tell them that before the meeting rather than editing afterwards to avoid the conversation.

Recording law varies by jurisdiction and some places require every participant's consent. Announce the recording at the top and keep that announcement in the transcript.

Record the room rather than relying on a call platform's own transcript where you can — a single good microphone in the middle of a table usually beats compressed per-participant call audio for speaker separation.

Then transcribe, check the short turns, write the decision and action block, and circulate. The video below covers the transcribe-and-download step.