For the first three years of my career as a product manager, I had the same ritual after every meeting: close my laptop, find a quiet corner, and spend 30 to 45 minutes reconstructing what had just happened.
Not because I forgot. Because I was too busy running the meeting to write anything down properly.
I’d have fragments. “Jake said something about the API deadline.” “Someone pushed back on the pricing tier — was that Sarah?” By the time I had coherent notes, I’d already lost the texture of the conversation. The action items were buried. The context was gone.
The Real Cost Nobody Talks About
PMs are supposed to be the source of truth. We write the PRDs, we own the roadmap, we close the loop. But if your notes are a mess, your follow-through suffers. I’ve seen entire sprint cycles derailed because an action item from a stakeholder call never made it into a ticket.
The problem isn’t attention span. It’s divided attention. When you’re facilitating, synthesizing, and deciding — all at once — you literally cannot also be transcribing.
What I Tried First (And Why It Failed)
I tried note-taking apps with templates. Better, but still manual. I tried having a dedicated note-taker join calls. That added coordination overhead and people talked differently when they knew someone was writing everything down.
I tried Otter.ai for transcripts. Great raw material — but a 90-minute transcript is not a summary. Reading it takes almost as long as the meeting itself.
What I needed wasn’t more words. I needed the right words, structured the right way, automatically.
The Setup That Actually Works
About six months ago I put together a simple pipeline:
- Record every meeting via Zoom’s built-in recorder
- Transcribe using Whisper (fast, accurate, runs locally)
- Process the transcript with an AI model using a structured prompt
- Deliver the output to Slack and Notion automatically
The prompt is the key part. I ask the AI to extract: a one-paragraph summary, a bulleted list of decisions made, a list of action items with owner and deadline, and any open questions that need follow-up. That’s it. Same structure every time.
The whole pipeline runs in under three minutes after a meeting ends. By the time I’m back at my desk, notes are already in Notion and the channel has been pinged.
What Changed
The obvious win: time. I get back 30 to 40 minutes per meeting. Over a week with eight or nine meetings, that’s nearly a full workday.
The less obvious win: quality. Because the structure is consistent, stakeholders know exactly where to look. Engineers stopped asking me “what did we decide about X?” because the answer was always one search away. Our sprint retrospectives got sharper because we had reliable records of what had been discussed.
What I Learned as a PM
Automation doesn’t remove judgment — it relocates it. I still review every output before it goes out. I still rewrite the summary when the model missed the point. The AI handles the mechanical reconstruction; I handle the interpretation.
The other thing I learned: start with one meeting type. I began with weekly syncs, got the prompt right, then expanded to design reviews and stakeholder calls. Each context needed slightly different structure. Trying to do everything at once is how you end up with a tool nobody trusts.
If you’re a PM drowning in post-meeting cleanup — this is worth an afternoon to set up. You’ll spend that time back within a week.