Bottom line: AI notetakers are excellent at transcripts and useless at judgement, and after 90 days of using one on every single sales call I run, the biggest cost wasn't the £20 a month subscription, it was what happened to the conversation once people knew the bot was listening. If you're going to use one, do it deliberately, tell people up front, and never let it near the calls where trust matters most.
Why I started recording every call with an AI
Back in September I got fed up with my own memory. I'd come off a call with a prospect, scribble three words on a Post-it, and by the time I sat down to write a follow-up email two days later I'd forgotten whether they said their budget was £3,000 or £30,000. That's not a small mistake to make in a proposal.
So I turned on an AI notetaker for every call. Every single one, sales, discovery, client check-ins, even a couple of podcast interviews I did as a guest. I used a mix of Otter.ai and Fathom over the three months, partly to compare them, partly because Fathom's free tier does full transcription and summaries with no cap on minutes, while Otter's free plan caps you at 300 minutes a month, which sounds generous until you're on four calls a day.
I wasn't testing this to write a review of the software. I wanted to know something more basic: does having an AI in the room change what happens in the room. It does. More on that in a minute, because that's the bit most people writing about these tools skip entirely.
What the AI got right
Let's start with the good, because there's a lot of it.
- Action items. Across roughly 60 calls, the tools correctly pulled out follow-up tasks about 85 percent of the time, in my rough count. "Send the case study by Friday", "check the pricing for the three-month package", that kind of thing. That alone saved me from at least a dozen dropped balls.
- Numbers and dates. If a prospect said a figure, it was in the transcript, word for word, timestamped. No more guessing whether it was £3,000 or £30,000.
- Speed. A 45-minute call turned into a usable summary in under two minutes. That's fast compared to me typing notes from memory an hour later, half asleep, at 9pm.
- Searchability. When a client mentioned in March that their financial year ends in July, and it mattered again in June, I could search the word "financial year" across three months of transcripts and find it in seconds.
If you're the kind of business owner who does most of your selling and client management on Zoom or Teams, and you're not technical, this is exactly the sort of low-effort, high-payoff AI use I've written about before in my piece on how non-technical business owners can use AI without needing to understand any of it. You do not need to know how a large language model works to get value from a good transcript.
What it consistently missed
Here's where it got interesting, and where I stopped trusting the summaries on their own.
The AI is brilliant at capturing what was said. It's terrible at capturing what was meant. Three examples from the 90 days:
- A prospect from a manufacturing firm in Yorkshire said, laughing, "oh we've got loads of budget for this, don't you worry." The AI summary flagged this as "budget confirmed, high." The tone was sarcastic. She had, in fact, no budget, and told me as much fifteen minutes later. The bot missed the joke entirely and I nearly built a proposal around a lie a nervous person told to save face.
- On a discovery call, a prospect went quiet for about eight seconds after I mentioned a price. That silence told me more than anything he said for the rest of the call. It doesn't show up in a transcript at all. There's no line for "long, uncomfortable pause that meant he was doing mental maths and didn't like the answer."
- Someone described their current supplier as "fine, I suppose." Two words that, said with a sigh, are a massive opening. The AI summary rendered this as "satisfied with current provider." I nearly didn't follow up on what was, in the end, a warm lead who switched to me six weeks later.
This is the bit worth being blunt about, because most write-ups of AI meeting tools treat the summary as gospel. It isn't. It's a transcript with confidence, not judgement. If you run your follow-ups purely off the AI summary without listening back to the tricky bits yourself, you will misread your best prospects.
The part that changed how people behaved on the call
Now the uncomfortable bit, the one nobody selling you a notetaker subscription wants to put in their marketing copy.
Once people knew a bot was transcribing the call, some of them changed how they talked. Not all of them, but enough that I noticed a pattern within the first month. A few prospects got noticeably more guarded the moment the little "this call is being recorded and transcribed by AI" notice popped up. One client, mid-conversation, said "hang on, is that thing writing down everything I say?" and went quieter for the rest of the call. Another asked me point blank to turn it off before she'd discuss a sensitive point about her business partner leaving. I did, obviously, and the conversation opened right back up the moment the red recording dot disappeared.
That tells you something real: people self-censor in front of a machine in a way they don't in front of a human, even one they've just met. If your business depends on people telling you the truth about their problems, and mine absolutely does, that's a cost worth taking seriously. I stopped using the notetaker for the first ten minutes of any call where I sensed a prospect was nervous, and I turned it off completely for anything involving a client complaint or a difficult renewal conversation. The transcript isn't worth the honesty you lose.
There's a second version of this problem that's harder to spot. A couple of my sales calls started to feel like performances for the transcript rather than conversations with a human. I caught myself, on at least two occasions, saying things in a slightly more polished, quotable way because I knew there'd be a written record. That's not a disaster, but it's not nothing either. Selling well is often about the messy, unscripted bit where you say the wrong thing and the prospect corrects you and you both laugh. A recorded, transcribed call nudges everyone toward being careful, and careful is not the same as trustworthy.
The data question nobody asks until it's too late
Here's a question worth asking before you turn any of these tools on: where does the transcript go, and who can read it later.
Most of these tools store your transcripts on their own servers, indefinitely, unless you go into settings and manually delete them. That means a full written record of a client telling you their cash flow is a mess, or naming a competitor by name, or admitting they're about to be sued, sits on a third-party server with its own retention policy that you almost certainly haven't read. If you're in the UK or EU and handling any personal data on those calls, and you almost certainly are the moment a client's name, health situation, or financial detail comes up, you need explicit consent to record and process that, not just a pop-up notice that half your prospects click past without reading.
In practice, for a small business, that means: tell people verbally, not just via a bot notice, that the call is being recorded and transcribed by AI, and let them say no. I now do this on every call. Some people say no. I turn it off, take notes by hand like it's 2004, and it's fine.
The real numbers on cost and time
Here's what it cost me across 90 days, roughly 60 calls:
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
- Fathom: free tier, unlimited calls, summaries and action items included. No cost.
- Otter.ai Business plan: £16 a month per user (paid in dollars, so it moves with the exchange rate), which covers unlimited transcription minutes and better search across old transcripts.
- Time saved writing notes: roughly 20 minutes per call I'd previously have spent typing up scrappy notes from memory, so around 20 hours over the 90 days.
- Time spent double-checking the AI's read on tone or sarcasm: about 5 minutes per call where something felt off, so maybe 3 hours total.
Net time saved: roughly 17 hours over three months. That's real, and worth having. It is not, however, "AI runs your sales process for you." It's a decent assistant with no sense of humour and no ability to read a room.
How I use it now, after the 90 days
I didn't ditch the tool. I changed how I use it.
- I tell every prospect verbally, before I hit record, that the call is being transcribed by AI and they can ask me to pause it any time.
- I turn it off for the first few minutes of any call where someone seems nervous, and for any conversation about money problems, staff issues, or anything a competitor could use against them.
- I never send a follow-up based purely on the AI summary. I skim the actual transcript for the parts where the tone might have mattered more than the words.
- I delete transcripts after 90 days unless there's a live reason to keep them, mostly because I don't want a server somewhere holding three years of everyone's business problems.
If you're weighing up whether to roll this out across a team rather than just for yourself, that's exactly the sort of decision worth getting outside eyes on before you commit a whole sales team to it, which is where working with an AI implementation coach earns its fee, because the mistakes here are the kind you only spot after three months of habits have already formed across a team, not after a demo call with a salesperson from the software company.
I've run enough of these small, deliberate experiments now, tracking a change for weeks and writing up exactly what happened rather than what I hoped would happen, that it's become a bit of a habit on this site. I did the same thing with my own SEO recovery after five quiet years, and the pattern is always the same: the tool does roughly what it says on the tin, and the interesting bit is always the human behaviour change nobody warned you about.
Who this is worth doing for
If you're a one-person consultancy or a small team doing a handful of client calls a week, an AI notetaker is worth the ten minutes it takes to set up. It will save you from forgetting things and misquoting prices. If you're running a sales team of five or more and thinking about mandating it across every call, slow down and think about the trust cost first, not just the licence cost.
And if you're just starting to explore what AI tools are worth your time as a small business owner with no technical background, there's a decent list of lesser-known but useful tools worth a browse in 50 websites you probably didn't know existed, several of which are exactly this kind of quiet, unglamorous productivity tool rather than the flashy stuff everyone's shouting about on LinkedIn.
Frequently asked questions
Do I need to tell people I'm recording a call with an AI notetaker?
Yes, and a pop-up notice from the software isn't enough on its own. Say it out loud before you hit record, and give people the option to say no. In the UK and EU, recording a call that includes another person's personal information without proper consent can put you on the wrong side of data protection rules, and beyond the legal point, it's simply the decent way to run a business call.
Which AI notetaker is best for a small business, Otter, Fireflies, or Fathom?
Fathom's free tier is the best value for a solo operator or very small team because it's unlimited and includes summaries. Otter is stronger for search across months of old transcripts once you're on a paid plan. Fireflies integrates well if you're already deep into a CRM. None of them are meaningfully better at reading tone or sarcasm, which is the actual gap that matters.
Will an AI notetaker make my sales calls better?
It will make your follow-ups more accurate and your action items more consistent, which is a real win. It will not improve your ability to read a nervous prospect, and in some cases it makes people slightly more guarded because they know the call is being transcribed. Treat it as a note-taking assistant, not a sales coach.
Should I use AI notetakers on every single call?
No. Turn it off for sensitive conversations, difficult client issues, and anything involving a third party's private business problems. The transcript is never worth losing the honest, unguarded version of the conversation, and after 90 days of testing this myself, that's the one rule I'd tell anyone to keep.
Related reading: The AI Notetaker in Your Sales Calls Is Quietly Costing You Deals.
Related reading: AI Notetakers Are Quietly Costing You Client Trust. Here's What I Do Ins.
Related reading: AI Meeting Notes: What They Get Right, What They Miss, and When I Switch Them Off and AI Meeting Notes Are Quietly Costing You Client Trust.
