Turning meeting notes into follow-ups with AI

How the bot works

Nine steps. Reading the transcript and chasing the owners is ordinary code; the model is asked only to find the commitments and draft the words.

1. It starts from the notes you already have. Zoom, Microsoft Teams, Google Meet, the note-taker you already pay for, a recording off a phone, or notes somebody typed. We don't replace your note-taker — it is good at transcribing, and the part that gets dropped is everything after.

2. It pulls out every commitment, with the sentence that made it. “I'll send the revised scope Thursday” becomes a commitment with the sentence quoted and a pointer back to where in the meeting it was said. It does not paraphrase a promise into something cleaner than what was actually agreed.

3. It names an owner, or says there isn't one. One person per item, matched against your team and the attendee list. “Someone should look at the pricing” is reported as unowned rather than assigned to the team, because an item owned by everybody is the one that never gets done.

4. It separates decisions from ideas from questions. What was decided, what was only floated, and what is still open and waiting on somebody else. Three short lists, in your order. A recap that runs to two pages gets skimmed, which is the same as not sending it.

5. It drafts the internal recap and the client-facing one separately. They are not the same document and never should be. The internal one names owners and says what is at risk; the client-facing one confirms what you agreed and what you will do next, with nothing in it that was said after the client left the call.

6. A person approves anything that leaves the company. The client-facing recap goes to a human with the transcript beside it. They send it, edit it, or decide it shouldn't go. This is the step the tools treat as optional, and it is the one that keeps a half-heard commitment out of a customer's inbox.

7. It puts the tasks where your team already works. Into your own tracker, your project tool or your CRM — one task per commitment, with the owner, the date, the quoted sentence and a link back to the meeting. If an item already exists it is updated rather than duplicated.

8. It follows up with the owner, on your schedule. A quiet nudge before the date, a direct one after it, and an escalation to whoever you name if an item goes past the point where somebody else needs to know. Chasing is the work nobody enjoys and the part that makes the whole thing worth having.

9. It writes the record. One row per meeting and one per commitment: the source transcript, the quoted sentence, the owner, the date, the task it opened, every nudge it sent, who approved the client-facing recap and what they changed, the model and prompt version, UTC timestamps, and what the run cost.

What you configure

It can be configured to run in any business. This is what that actually means — the list of things that are yours, not ours.

Your recording and consent rules

Which meetings are in scope, who has to be told what before anything is captured, and which meetings are never processed at all. Your rules and your jurisdiction's — the bot works from whatever your tool is already allowed to produce.

Your definition of a commitment

What counts: a dated promise, an agreed next step, a request somebody accepted. And what doesn't: thinking out loud, a maybe, an idea nobody owned.

Your people and your tracker

Who is on the team, how names in a transcript map to them, and which tool the tasks land in — your project tracker, your CRM, or a list in the system you already use.

Your recap formats

Two templates, yours: the internal one and the client-facing one. Length, order, tone, what is always included and what is never mentioned outside the company.

Your approval rules

Who approves a client-facing recap, whether an internal recap needs approval at all, and what happens to a recap nobody has approved by the end of the day.

Your chase schedule

When a nudge goes out before a date, when one goes out after it, how many there are, who gets escalated to and at what point. Set per client or per project if you want.

Your channels

Where the recap and the nudges land: email, Slack, Microsoft Teams, or a comment on the task itself.

Your retention

How long transcripts and recordings are kept, who can open them, which meetings are excluded entirely, and what is deleted once the tasks exist.

The words it uses

The versioned prompts behind the extraction and the drafting — yours to read, yours to tune, and replayed against meetings you have already recapped before a change goes live.

Transparent, and configurable by you

The dark band below is what the system will not do. This is what you can see and change — and it is how we build everything, not something invented for this page.

You see what it did, and why

Every commitment carries the sentence that created it and a pointer back to its place in the transcript. You can read the line somebody actually said before you agree that it was a promise — and an item it could not attribute says so rather than guessing.

Versioned prompts you can tune yourself

The prompts behind the extraction and the drafting are yours to read and yours to tune. A change is a reviewable diff, replayed against meetings you have already recapped so you can see what it would have pulled out differently, and rolled back the same way.

A human in the loop wherever judgment actually matters

Automation handles the reading, the drafting and the chasing. Anything that leaves the company is approved by a person, and a commitment the bot could not tie to an owner is reported as unowned rather than assigned to somebody. That boundary is a property of the system, not a promise about how carefully we'll use it.

Every run logged, along with what it cost

One row per meeting and one per commitment: the source, the quoted sentence, the owner, the task, every nudge, the approver and the timestamps — plus token usage and cost for that run.

It can be configured to run in any business. The rules and the wording are yours: your definition of a commitment, your recap templates, your chase schedule, your approvers, your tools. Meeting follow-ups on this page are one configuration of it, not the limit of it.

Works with your tools

This runs on top of the meeting tool and the tracker you already pay for. We're not selling you another note-taker.

• Zoom
• Microsoft Teams
• Google Meet
• Otter, Fireflies, Fathom or Granola
• Asana, Jira, ClickUp or Monday
• HubSpot
• Salesforce
• Notion or Confluence
• Email, Slack or Microsoft Teams
• A transcript file, if that's all there is

Works well for

• Financial advisory firms — Review meetings produce promises about paperwork and dates, and the file has to show what was said.
• Law firms — A client call ends in next steps, and months later the sentence that created each one is what matters.
• Recruiting and staffing agencies — Intake calls set expectations on rate and start date, and the recap is what the client holds you to.
• Insurance agencies — Renewal meetings turn into coverage changes, and none of those should leave the office unread.
• Consulting firms — Every session ends with actions on both sides, and the client-facing recap is a different document.

How a rollout works

Nothing is sent on day one. A dry run over meetings you have already recapped by hand comes first, and it is where your own numbers come from.

1. 01Discovery

   The quote call, then a working session. We read your recording and consent rules, two or three real transcripts, the recaps your team wrote from them, where your tasks live and who approves what goes to a client. You get the extraction rules, the two recap templates and a written scope.

2. 02Dry run

   The bot runs over meetings you have already recapped. Nothing is sent, no task is created, nobody is nudged. We compare its commitments with what your team actually wrote down, item by item — every miss is a rule to fix and every extra is a definition to tighten. This is where your own numbers come from, on your own meetings.

3. 03Assisted mode

   Live meetings. The bot extracts, drafts and opens the tasks; a person approves every client-facing recap before it goes. Internal recaps and nudges can run on their own once you are happy with them.

4. 04Optional standing recaps

   Optional. For an internal weekly where the shape never changes, the recap can send itself with every hard stop in force and the kill switch. The client-facing one stays approved by a person — plenty of teams stop at step 03, and that's a good place to stop.

What we don't claim

• No numbers we haven't earned. No percentage of action items completed, no extraction-accuracy figure, no time saved. We haven't run this on your meetings. The dry run is where those numbers come from, and they'll be yours.
• No customer claims on this page. No names, no case studies, no logos.
• It can only work from what was said. A meeting where nobody agreed who was doing what produces a list of unowned items, which is an honest output and an uncomfortable one. The bot will not invent the owner the conversation never named.
• It isn't a note-taker, and it isn't a project manager. Transcription is already solved and you already own a tool for it. This is the part after, and it does not replace the person who runs your projects or decides what matters.
• Fully autonomous client communication isn't on offer. The design assumes a person approves anything that leaves the company. That's deliberate, and it's why the boundary is written down.

Questions people ask

Can AI turn meeting notes into follow-ups and action items?

Yes - and the useful version is not a summary. An AI agent can read the transcript your meeting tool already produces, pull out every commitment somebody made with the sentence that created it quoted rather than paraphrased, name one owner and a due date for each, separate what was decided from what was only floated, draft the recap, open the tasks in your own tracker and then chase the owners. What it does not do is send a client-facing recap that nobody read, or assign an item the conversation never actually assigned.

Does it replace our note-taker?

No, and it is cheaper not to. Whatever you already use - Zoom, Microsoft Teams, Google Meet, Otter, Fireflies, Fathom, Granola, or somebody typing notes - produces the input. Transcription is solved and you are already paying for it. The part that gets dropped is everything after the transcript: who owes what, whether it happened, and what the client was told you would do.

Will it email our client a summary on its own?

No. The client-facing recap is drafted and goes to a person with the transcript beside it, and they send it, edit it or stop it. That is deliberate: a recap is a written record of what you committed to, and a half-heard sentence in a customer's inbox is expensive to take back. Internal recaps and nudges can run on their own once you are happy with them, which is a different risk entirely.

What about recording consent?

It stays your decision and your rules. The bot does not join calls, start a recording or process a meeting you have excluded - it reads what your existing tool is already permitted to produce, and a meeting with no permitted transcript simply has no follow-up. Which meetings are in scope, who must be told what beforehand, how long transcripts are kept and who can open them are all configuration, and we will not advise you on your jurisdiction's law.

How does it decide who owns an action item?

By the sentence, matched against your team and the attendee list - and it reports an owner it cannot establish rather than inventing one. "I'll send the scope Thursday" has an owner and a date. "Someone should look at the pricing" is reported as unowned, because an item owned by the whole team is the item that never gets done. An unowned item goes to whoever chairs the meeting as a question: who is this?

Does it chase people, or just make a list?

It chases, and that is most of the value. A quiet nudge before the date, a direct one after it, and an escalation to whoever you name once an item is late enough that somebody else needs to know. The schedule is yours and can differ per client or project. Every nudge is recorded, so there is no argument later about whether somebody was asked.

Will it put the tasks in our project tool?

Yes - one task per commitment, with the owner, the date, the quoted sentence and a link back to the meeting, in Asana, Jira, ClickUp, Monday, Notion, HubSpot, Salesforce or whatever your team actually opens. An item that already exists is updated rather than duplicated. If your tracker is not one of those, the question is only whether it offers an API or an import; most do.

What stops something said in a meeting from making it do something?

Words spoken on a call are treated as data and never as instructions. Every model call runs as structured output against a fixed schema, untrusted text is kept out of the instruction block, and anything that reads like an instruction is quarantined, flagged for a human and logged. Saying "send this one straight to the client" out loud does not send anything, and the approval rules sit outside the model entirely.

How long does it take to get running?

The first milestone is a dry run, and it needs three things: two or three real transcripts, the recaps your team wrote from them, and a decision on who approves what goes to a client. We will not quote you a timeline before we know which tools are involved and what access they allow - you get a real one in your quote, once Ozzie's call has gathered the details.