A product manager's calendar in discovery season is customer calls, almost without gaps between them, and the job on the other side of each one is turning an hour of a customer describing their problem in their own words into something specific enough to argue for in a roadmap review. Typing through the call gets a rough approximation. Recording it gets the actual words, including the ones that mattered more than they seemed to at the time.
The bot most customers have already opted out of noticing
Otter and Fireflies handle the write-up by putting a named participant into the call, and for a customer who has had a dozen vendor calls already this quarter, another named bot is barely remarkable. The sharper cost shows up somewhere else: the recording, the transcript and the summary are all produced on that vendor's own servers, which means a customer's product complaint — sometimes about a competitor, sometimes about their own internal process — sits somewhere neither the PM nor the customer chose together. Granola avoids the bot but is still a cloud subscription end to end, so the same question applies to it too.
Decisions
- Deprioritise the export feature; the real blocker is permissions
Actions
- Send a follow-up with the permissions roadmap by next week
Open questions
- Would a read-only role solve this without full SSO?
What the app does with a customer call
Coii AudioNotes records the microphone and the system audio a Mac is already carrying, as two separate tracks, and transcribes both on that same machine while the call is still running. A language model bundled inside the app turns the transcript into a summary once the call ends — decisions, action items, open questions — so the specific phrase a customer used to describe their problem is still there to quote back to engineering, rather than paraphrased from memory by the time the roadmap review happens.
- 00:14:02CustomerThe export just times out once we're past about ten thousand rows.
- 00:14:10PMDoes it fail the same way every time, or only on certain reports?
- 00:14:17CustomerOnly the ones with the custom fields turned on, we think.
Telling a customer's voice apart from the PM's own
A bot inside a call separates speakers using a per-participant feed from the meeting platform. Reading the Mac's own audio instead means that feed doesn't exist, so speaker separation works by voice print — matching each voice against the others already in the recording. A one-on-one discovery call is exactly two voices, which is the case voice-print matching handles most reliably.
What it deliberately doesn't do
There is no integration with a roadmap tool, a ticketing system or a CRM — nothing here files a piece of feedback against a feature request automatically. There is no shared workspace a designer or an engineer can browse without being sent the file directly; each recording lives on the Mac that made it, across up to three of a PM's own machines on one licence. It runs on macOS only, so a PM on a Windows laptop needs a different tool for the same job.
Why the exact wording outlasts the summary
A synthesized theme like "customers want better permissions" is useful in a roadmap deck and useless in an argument with an engineer who wants to know exactly what broke. The transcript underneath the summary is where "it times out past ten thousand rows, only with custom fields on" lives — the sentence that turns a vague ask into a reproducible bug report, and the sentence that gets lost first when notes are typed live instead of recorded and reviewed afterward.
A quarter of customer calls, searchable together
One call's transcript is useful for the roadmap review it feeds. A quarter of them, all searchable on the same Mac, is a different kind of asset: "how many customers mentioned SSO unprompted this quarter" is a question a PM can actually answer by searching transcripts rather than by remembering which calls felt SSO-heavy — without any of those calls having passed through a server other product teams' customer data also sits on.
Who this genuinely isn't built for
A product org that already routes every customer call through a shared insights platform, where a whole team tags and searches feedback together across dozens of PMs, gets real value from that shared infrastructure — value this tool doesn't attempt to replace. This is built for the PM whose calls are mostly their own to record and synthesize, not for a workflow that depends on a shared, cross-team feedback repository from day one.
The convenience a manual recorder gives up
A bot that joins every calendar invite automatically captures a call without the PM remembering to start anything, which is genuinely useful on a day with back-to-back discovery calls and no gap to think about recording. A tool that starts by hand shifts that responsibility onto the PM — a real cost worth naming rather than a reason to pretend a manual recorder has no downside next to an automatic one.
Writing a spec that survives contact with engineering
A feature spec built on "customers want better permissions" gets questioned line by line in review, because it's a conclusion without the evidence underneath it. A spec that quotes the actual transcript — the customer's own sentence about the ten-thousand-row export timing out — gives an engineer something to design against directly, and gives a skeptical stakeholder a reason to trust the prioritization that isn't just the PM's judgment call. The transcript is what turns a roadmap argument into a citation.
Starting with the calls that don't matter yet
The first week goes better started on a routine check-in rather than the call with the account that's about to churn. A PM who tries the app on an ordinary customer call first gets a feel for how the transcript reads and what the summary produces, before relying on it for the call where getting the exact wording actually matters. That ordering costs nothing and means the habit is already built by the time the important call comes around.
What a designer or engineer can and can't see
A customer call's transcript lives on the PM's own Mac, not in a shared folder the whole team can browse. If an engineer wants the exact wording behind a bug report, the PM sends that specific transcript excerpt — the same way they'd forward a quote — rather than granting the team standing access to every customer call the PM has ever recorded, most of which have nothing to do with that bug.
The version of this that fails quietly
The realistic risk isn't a leak — it's a PM who means to record every discovery call and simply forgets on the one call where a customer says something that would have reshaped the next quarter's roadmap. Nothing here reminds anyone to press record before the call starts; building that habit on ordinary weeks, before it's the call that mattered, is the only real fix.
Setting it up is smaller than the decision to switch
The first launch asks for two macOS permissions — microphone access and permission to record system audio — the same prompt any Mac app requesting audio capture triggers, and that's the whole of it. No account to register, no workspace to configure, no CRM to connect before the next customer call gets recorded. The larger part of moving off a bot-based habit is usually the habit itself: remembering to press record instead of trusting a bot that used to join automatically.
A customer call from a conference floor
A discovery call taken from a trade-show booth or a customer's own office, on whatever Wi-Fi happens to be available there, is exactly the situation where a cloud notetaker's write-up depends on a connection that may be weak or absent. Because transcription and the summary both run on the Mac itself, the notes get produced the same way regardless of what the network looks like that day — a difference that shows up rarely and matters completely on the day a bad connection would otherwise have meant no notes at all.
What actually changes over a year of discovery calls
Nothing about how a customer call runs is different with this in the background — same platform, same questions, same conversation. What changes is what a PM has available a year later, when a roadmap decision gets questioned and the answer depends on whether the original customer call is a rough memory or an actual transcript, searchable on the Mac rather than reconstructed from whichever notes survived that quarter.
What a stakeholder review actually asks for
A skeptical stakeholder in a prioritization review sometimes asks whether a request came from one loud customer or a real pattern. A searchable archive of customer calls turns that into an answerable question — how many calls actually raised this, in what words — rather than a PM's recollection of how common the complaint felt, which is exactly the kind of claim a roadmap argument needs to survive being challenged.
The question a customer eventually asks
A customer who notices a recording indicator on a discovery call sometimes asks where the conversation goes. "Nowhere but the Mac that recorded it, and nothing is sent anywhere to write up the notes" is a complete and checkable answer — a shorter and more honest one than pointing a customer toward a vendor's privacy policy they are unlikely to read before the call ends. It's also an answer a PM can give without checking anything first, which matters on a call where the question comes up unannounced.
The habit that outlasts any one tool
None of this replaces the judgment a good PM already brings to a customer call — knowing which comment to chase and which to let pass. What it changes is whether that judgment is backed by an accurate record of what a customer actually said, rather than a summary shaped by however clearly the call happened to be remembered by the time there was a moment to write it down.
What it costs, what it runs on
Coii AudioNotes is $19, paid once, for three of a PM's own Macs, with a 30-day trial that needs no card and no account. It runs on macOS 13 Ventura or later, Apple Silicon or Intel. Against Otter's $8.33–$16.99-a-month Pro tier or Fireflies' comparable subscription, that's $19 total rather than a monthly line item for as long as the job lasts; against Granola's $14-a-month Business tier the price gap narrows, but a customer's product complaint still goes to a cloud service to produce the notes either way. Full comparisons here, here and here. What happens when a customer call runs somewhere without reliable Wi-Fi is covered in offline transcription for Mac, and the wider case for a Mac-native meeting notes app and what an AI notetaker without the bot actually changes both cover ground this page doesn't repeat. The same shape of problem, one level up, shows up for founders on investor calls and for engineering managers running one-on-ones with the team building what the PM's calls describe.