I’m sure you know the feeling: You hang up from a 60-minute kickoff call and you’ve got three pages of somewhat-legible notes, 10 different tabs open in Chrome, and you’re already pretty sure you missed something. Six weeks later you find out what it was, when the client says “I thought we agreed the redesign included the blog templates” and you’re staring at a SOW that says otherwise.
Running a good discovery call and taking down all the action items and notes at the same time is no small feat. Every minute you spend typing is a minute you’re not actually listening, not asking the obvious follow-up, and not catching the thing the client said sideways at minute 38 that turns out to be pivot point of the whole project.
So I stopped taking notes on calls. You read that right: I don’t take notes on client calls anymore. I use a tool that records and transcribes everything, and I turn that transcript into a clean requirements doc or SOW in just a couple of minutes. There’s no prompting expertise involved and nothing to configure. If you’ve been waiting for a low-risk place to start with AI, this is it.
Here’s what the workflow actually looks like, what it costs in time and money, and the before-and-after on what comes out the other end.
This is the easiest AI win you’ll ever get
This is the crawl stage, the first level of the Crawl, Walk, Run, Fly framework I use with agency owners. At this stage you’re not “doing AI” in any way that would impress someone on LinkedIn. You’re installing a tool that has AI built into it and letting it run. Zero learning curve, nothing to prompt, payoff on your very next call.
I tell every agency owner to start here for the same reason. It’s the lowest-risk, most relatable entry point, it fits into how you already work, and it quietly builds the data foundation that everything in the later stages depends on. You get an immediate win today and you set yourself up for the bigger ones later.
The setup is the whole barrier to entry
I’ll save you the comparison shopping: use Granola. I’ve tried a handful of these and I’ve been a Granola-only user for awhile now. There are a few reasons it came out on top for me:
It never joins your call. Most tools (Fathom, Otter, Fireflies) dial into the meeting as a bot, which means there’s a little robot sitting in the participant list. Granola doesn’t do any of that. It captures the audio straight off your computer, so nothing shows up on the call and nobody’s performing for the recorder. This also means that it works the same across Zoom, Google Meet, Webex, a Slack huddle, or sitting across a table from someone in person. It only keeps the text, too. The audio gets deleted as it transcribes, which is a clean answer when a client asks where the recording lives.
The interface is just a notepad. It looks like Apple Notes: a blank page and nothing else. You can still jot some of your thoughts down during the call if you want, and Granola will weave it into the generated notes after the meeting is over. There’s no dashboard to learn and no setup to fight with, which makes adopting a new tool even easier.
It speaks MCP. MCP is the emerging standard for connecting AI tools directly to your data, and Granola ships an official connector for it. In plain terms: when you’re ready, Claude (or ChatGPT, or Cursor) can pull your meeting notes on its own, with no copying and pasting at all. That’s a Walk-stage move, so just file it away for now. It’s the reason I’d build on Granola specifically, because the data you start collecting today is already wired to plug into everything you’ll do later.
If you’re already on Fathom or Otter and happy, don’t switch for the sake of it. But if you’re starting from zero, I’d start with Granola.
So what do you do during the call itself? Almost nothing. It records, transcribes, and drops a summary plus a list of action items and decisions in your lap the second you hang up. No new process to sell the team on, nobody hunched over a laptop typing while the client talks.
The one habit that makes it pay off long-term: get organized from day one. Set up a folder per client, name every file with the date of the call, and put them somewhere you’ll actually look. It takes ten seconds per call and it’s the difference between “I know we discussed scope somewhere” and pulling up the exact transcript where the client signed off.
From a transcript to a requirements doc
The full flow, start to finish, looks like this: client kickoff call → automatic transcript → a structured requirements doc you can actually hand to your team.
Most of that happens for free. The tool gives you the transcript, the summary, the action items, and the decisions without you lifting a finger. The one small step I sometimes add is to take the transcript, paste it into Claude (ChatGPT works the same way), and ask it to reshape the call into the doc format I actually use: including things like scope, deliverables, open questions, and technical constraints. This means that in just a couple minutes, I’ve got a first draft of a requirements doc or SOW that I can send back to a client instead of just a transcript.
The before-and-after is the part that sold me. My kickoff notes used to be scattered and partial, written while I was half-listening, with the important details either missing or buried under stuff that didn’t matter. Now I get something complete, searchable, and structured, every time, without trying.
Why it’s worth it before you’ve saved a single hour
You’ll definitely notice some time savings once you get this whole thing running, but they’re not even the best part. Right away, you’ll notice you can be more fully present on calls, you’ll stop missing details or things mentioned in passing and you’ll impress your clients by turning around polished, accurate documents even faster.
Every one of those ties back to client value, not just to the novelty of using a new tool. Better-scoped projects and fewer disputes are worth real money to an agency, and they show up immediately.
Another benefit: this work compounds
The value isn’t really in any single call. Instead, it’s that six months of transcripts turn into a searchable knowledge base for every client you have. Every decision, every offhand comment, and every “let’s circle back on that” is captured and findable.
That’s the main reason to start now even if today’s payoff feels small. You’re not just saving time on one requirements doc. You’re building the pile of organized data that makes the next stages, prompting AI directly and setting up per-client Claude Projects, dramatically more useful. The people who get the most out of AI a year from now are the ones who started collecting clean data today.
Bonus: the upsells hiding in your call transcripts
There’s money sitting in calls you’ve already had. Clients mention problems, goals, and frustrations in passing that never make it onto a SOW, because you were busy writing down the stuff that did.
Once you’ve got a stack of organized transcripts, you can dump a batch into Claude and ask what upsell opportunities you’re missing across all of them. That’s a Walk-stage move, so I won’t break down the whole workflow here. For now, just remember that the data you’re starting to collect is worth even more than just the requirements doc it produces.
It’s not only upsells, either. I’ve written up a related version of this idea over on Claude Code for Marketers, mining those same call transcripts for social content instead of letting your best off-the-cuff explanations disappear the second you hang up. Same transcripts, different payoff.
Where to start
Before your next client call, install Granola and set up your folder structure. That’s the entire commitment. And if it feels too simple to count as “adopting AI,” good. That’s the point of the crawl stage. You’re not behind, you’re starting in exactly the right place.
Want help figuring out your next step?
If you want a second opinion on where you actually are with AI and what the next practical move is for your agency, that’s what I do. No hype, no fifteen-tool stack you’ll abandon by Monday, just the next useful thing.
Book a free intro call or email me directly at [email protected].