People & Exec SupportFree
Meeting Notes and Action Items
Separates the decisions a meeting actually closed from the talk that sounded like one, and gives every commitment a named owner.
River's meeting notes and action extraction reads the transcript as a sequence of speakers rather than a block of text, which is what makes attribution possible. It marks every statement that carries an obligation, then sorts those into decisions that closed, commitments with an owner and a date, talk that closed nothing, and questions nobody answered. The four lists come back separately, because the interesting one is usually the third, and it is the one a summary quietly folds into the others.
Unlike a transcription service's own summary, which compresses the meeting and inherits every word it mis-heard, this one treats a suspect word as suspect. A name rendered three different ways is flagged rather than picked between, and a number two speakers disagreed about comes back as a question instead of a fact. Unlike a generic chat prompt, it will not invent an owner for a commitment nobody in the room accepted, and it says so on the line rather than leaving the slot blank.
Chiefs of staff and executive assistants run this on leadership meetings where the same decision keeps reappearing every month. Programme and operations managers run it on anything carrying an action register between sessions. It pairs with the executive calendar audit when the meetings themselves are the problem, and with survey comment synthesis when the transcripts are research sessions rather than decision meetings. Recruiters run it on hiring debriefs, where interview scorecard synthesis then takes over.
What gets lost between a meeting and its notes
The common complaint about automated meeting notes is that they miss things. The more expensive failure is the opposite. They include everything, and a sentence somebody floated at minute nine arrives in the notes carrying the same weight as a decision the room actually made. A reader who was not there cannot tell those two apart, so the notes get treated as a record of commitments, and a good share of them were never commitments at all.
Ardent Freight ran one 58-minute ops review through this: 892 cues, six speakers, and 41 statements that carried some obligation. Twelve of them closed a decision and nine were commitments with an owner. Fourteen, more than a third, read exactly like commitments and closed nothing. Six were questions the meeting never came back to. Of the nine real commitments, only five were spoken in the first person by the person taking them on, which is the one case where the file hands you the owner without anybody having to interpret it.
Two of those nine carried no date at any point in the transcript, and three had a relative one that only resolves if you know the meeting's own date. The last was assigned to somebody who was not in the room, so nobody present could accept it, and it belongs in the register marked unconfirmed rather than as an action. Three mis-heard words were flagged instead of carried forward: a headcount, a carrier name, and a deadline.
How it works
Hand over the transcript
Paste the VTT, SRT or plain export, and say what the meeting was deciding
Speakers get separated
Each cue is attributed to whoever spoke it, which is what makes ownership checkable
Four lists come back
Decisions, owned commitments, talk that closed nothing, and the questions the meeting left open
Work the register
Ask for a chase list, a version for somebody who missed it, or one owner's actions
What you get
- Decisions the meeting actually closed, kept separate from the talk that only sounded decisive
- Every commitment with a named owner, taken from who spoke it rather than guessed
- Relative deadlines resolved against the meeting's own date, and missing ones flagged as missing
- Commitments assigned to somebody absent, marked unconfirmed because nobody present could accept them
- Suspect transcription flagged word by word, so a mis-heard number never becomes a fact
- An action register row per commitment, ready to carry into the next meeting
Common questions
Which transcript formats does it read?
WebVTT from Zoom or Teams, SRT, and plain text exports from Otter or Fireflies. VTT is the most useful, because the WebVTT specification carries a voice span naming the speaker on each cue, so ownership comes out of the file rather than out of inference.
How does it know a commitment is real rather than someone thinking out loud?
It looks for a named person accepting it, in their own words, in the same stretch of transcript. A proposal nobody answered stays in the list of things that closed nothing. You can loosen that in the form if you would rather see every proposal with the unowned ones marked.
What does it do when the transcription is obviously wrong?
Flags the word and shows both readings rather than choosing. In the Ardent Freight run a headcount came through as fifty where two speakers had said fifteen, and a carrier's name appeared three ways in one file. Those went into the notes as open questions, which is where they belong.
Can it handle a meeting where nobody assigns anything explicitly?
Yes, and that is a common and useful result. The output is then mostly decisions and open questions, with an empty commitments list that is itself the finding. A meeting that closed four decisions and produced no owned action is worth knowing about before the next one.
What happens to the register between meetings?
This tool fills a register; keeping one honest afterward is a different job. The board meeting packet template carries it forward under an identity that has to reconcile every time, and a decision register with a dissent field shows which topics keep returning to the agenda. For commitments to someone outside the company, a briefing that ages every open one flags what no later meeting raised.
Does it work on a transcript with no speaker labels?
It works, with a caveat it states on the page. Without speaker attribution the decisions and open questions come out fine, but a commitment can only be reported as one the room made, not as one person's. Ask your recording tool to export with speaker separation where you can.
How is this different from asking an AI assistant to summarise the call?
A summary optimises for brevity, which is exactly wrong here, because the commitments a meeting drops are short and easy to compress away. This keeps the four categories separate and reports counts, so a third of the obligations closing nothing is visible instead of smoothed over.
Meeting Notes and Action Items
Fill in the form and your workspace opens with the work already underway.