Your AI notetaker has never been in a customer's kitchen
AI notetakers like Otter, Fireflies, and Fathom work by joining video calls as a participant or capturing computer audio. Field sales conversations happen in person, so there is no call to join and no bot in the participant list: the entire category structurally cannot see the meetings where in-person businesses make their money. Meeting intelligence for the real room requires phone-based, consent-first capture instead.
A category built for the desk
The AI notetaker boom followed remote work, on an assumption that was entirely reasonable at the time: if every meeting is a video call, a bot in the call can hear everything. Every major product in the category is built on it. Otter, Fireflies, Fathom and Gong send a participant into the call. Granola takes a different route and captures your Mac's audio while you type notes alongside it.
Both routes need something specific to exist. The bot needs a calendar invite, a join link, a participant list and a host to admit it. Granola needs an open laptop and a person typing on it.
That assumption holds beautifully for software sales and fails completely for the trades, home services, medical device reps, real estate, and every business whose revenue conversation happens across a table. The result is a strange asymmetry: a company's cheapest internal stand-up is transcribed, searchable and summarised, while its $18,000 kitchen-table close exists only in one rep's memory.
The limits you hit even when you improvise
Most of these apps have a phone recording mode, so the obvious workaround is to use one as a general voice recorder in the field. It is better than nothing. It also runs into the pricing model, which was designed around calls and quietly punishes field work.
The caps below are from each vendor's own pricing page, checked in August 2026. Two things matter for in-home sales. First, the per-recording ceiling: a first visit on a big-ticket job routinely runs past ninety minutes, and a cap always truncates the end, which is exactly where price, objections and the decision live. Second, the monthly minute allowance: a rep running three appointments a day at roughly 75 minutes generates about 4,500 minutes a month, which exhausts most mid-tier plans in under a week.
The second one does more damage than it looks, because it does not stop the rep, it changes them. Someone rationing minutes starts deciding which appointments are worth recording, and the ones they skip are the routine-looking visits. Those are precisely the visits that reveal why a lead source underperforms.
| Tool | How it captures | Limit that bites in field sales |
|---|---|---|
| Otter | Bot joins the video call | 4 hours per meeting on Business; 1,200 min/month and 90 min/meeting on Pro |
| Fireflies | Bot joins the video call | 3 hours per recording on Business, 2 hours on Pro |
| Fathom, Gong | Bot joins the video call | No call to join at an in-home appointment |
| Granola | Captures Mac audio while you type | Needs an open laptop and someone typing |
| Plaud | Dedicated hardware recorder | 1,200 min/month on Pro; a device to buy, charge and remember |
| Knack | Phone on the table, on device, consent-first | No minute cap and no per-meeting cap on any plan |
What the room actually needs
In-person capture has different physics all the way down. No join link, so the phone on the table is the recorder. No mute button and no bot in a participant list, so consent has to be explicit, spoken and social rather than implied by a name appearing on screen. No reliable connectivity in a basement, an attic or a rural job site, so capture has to be on-device first and upload later. Multiple overlapping voices at conversation distance rather than one person in a headset, so the transcription pipeline has to be tuned for rooms.
There is a legal difference too, and it is not a footnote. Recording colleagues on a video call is largely handled by the bot being visible in the participant list. Recording a homeowner inside their own home is the textbook case of a reasonable expectation of privacy, and around a dozen US states require all-party consent for it. A product built for calls has no reason to have solved that, and mostly has not.
And the output has to leave the app. A field rep will not review a transcript library at nine at night. The meeting has to become the follow-up email, the CRM update and the booked next visit on its own, or it becomes another thing that does not get done.
What to do on Monday
Work out how much of your revenue conversation your current tooling can actually attend. Count last month's appointments, then count how many had a join link. For most home services businesses the second number is close to zero.
Check the caps on whatever you already pay for, both per recording and per month, against a real rep's volume rather than an average.
Decide the consent policy before the tooling, not after. Always-ask is correct in all fifty states and removes a judgment call from a tired rep standing in someone's kitchen.
And judge any option by what it produces rather than what it captures. A transcript is not the goal. The follow-up sent, the CRM updated, and the pattern across a few hundred appointments are the goal, and a tool that stops at the transcript has left the expensive part to a human who is already out of hours.
Sources
Why can't AI notetakers record in-person meetings?
Because the mechanic they are built on needs a video call. A bot requires an invite, a join link and a participant list; Granola requires an open Mac and someone typing. An in-home appointment has none of those, so what you fall back to is a generic voice recorder rather than the product they built and priced.
Can I just start a voice memo and upload it to an AI later?
You can, and it beats nothing. It leaves consent implicit, it breaks on long recordings, and it produces a transcript instead of finished work: no follow-up sent, no CRM updated, no memory across meetings. It also relies on the rep remembering to do it every time, which is the assumption that fails first.
Do any call bots handle in-person meetings?
Several offer a phone recording mode as a side feature. The difference is the centre of gravity. Tools built for calls treat the room as an edge case, which shows up in the minute caps, the consent handling and what happens after the recording exists.
What is the minute cap problem, concretely?
A rep doing three appointments a day at around 75 minutes generates roughly 4,500 minutes a month. Otter Pro allows 1,200 and Plaud Pro allows 1,200, so the plan runs out in about six working days. The real damage is behavioural: a rep rationing minutes stops recording the routine-looking visits, and those are where the patterns live.
Why does a per-meeting length cap matter?
Because of where it falls. A long first visit goes discovery, walkthrough, options, price, objections, decision. A cap truncates the end, which is exactly the part that decided the outcome. Fireflies caps a recording at 3 hours on Business and 2 on Pro; Otter at 4 hours on Business and 90 minutes on Pro.
Is this a criticism of those products?
No. They are good at the job they were built for, and if your pipeline runs on video calls they are the right choice and usually cheaper. The point is narrower: the category was designed around a meeting that has a join link, and a large part of the economy sells in rooms that do not.
What should in-person teams look for instead?
Capture that works on a phone and survives no signal, no monthly or per-meeting caps, consent built into the flow, recaps you can verify against the audio, and output that leaves the app as a sent follow-up and an updated CRM rather than another archive to read.