A few months ago, we shared how Matrix Group chose Fathom as our AI-powered note-taking app after testing multiple options. It was an exciting step in our AI journey—one that promised to make note taking more accurate and efficient, ensuring that key takeaways didn’t get lost in the shuffle.
But as we started using Fathom more regularly, we ran into some challenges. Not because the tool wasn’t working—it was!—but because it wasn’t working for us in the way we needed it to.
Our AI Notetaker Was Helpful, But Not Quite the Right Fit for Us
When we first implemented Fathom, we were excited about how well it captured and summarized our meetings. The AI did exactly what it was designed to do—generate summaries, highlight key points, and pull out action items.
But as we started using it regularly, we realized that Fathom wasn’t quite working for us. Here’s where things got tricky:
- The way Fathom labeled action items didn’t quite align with how our team structures follow-ups. The AI was tagging both “to-dos” and “next actions,” but in a way that didn’t match our workflow, which meant our project managers had to do additional editing to make the notes truly actionable.
- The formatting of the notes wasn’t translating well into our client extranet. This meant extra steps to clean things up before sharing them out with our clients.
- We weren’t experiencing the time savings we expected. Instead of eliminating work, we found ourselves reworking AI-generated notes to fit our processes.
These issues didn’t mean Fathom was a bad tool—just that it wasn’t the perfect fit for how we work. So, we started exploring ways to tailor AI-generated meeting notes more precisely to our needs.
Could a Custom AI Note-Taker Work Better?
We asked the big question: What if we wrote a custom GPT for note taking?
Since we already used custom GPTs for other tasks and workflows, we were confident that a custom GPT could help us process meeting transcripts and generate notes formatted exactly how we wanted. And, it did!
We created a custom GPT that would take transcripts from Zoom, Teams and Google Meet and generate meeting notes that:
- Listed the meeting attendees, grouped by organization
- Organized the notes by topic; the notes were appropriately detailed (Fathom sometimes summarized discussions too much)
- Listed the To Do items at the top and bottom of the notes
- Used limited formatting so the notes could be sent through our company intranet
This new custom notetaker was amazing. We rolled it out to the project managers and cheered when they said this notetaker was better than Fathom.
Removing Friction Still Didn’t Solve the Problem
A few weeks later, we discovered a problem: No one was using our new custom notetaker
Why? The process required extra steps—logging into Zoom, downloading the transcript, uploading it to our custom GPT, and retrieving the notes. It was an extra layer of work that most people weren’t willing to take on.
To make adoption easier, we assigned a team member to handle the AI note-taking process for everyone. The idea was simple: instead of expecting project managers to run the process, we’d do it for them and distribute the notes.
Yet when we followed up, we realized something surprising: PMs still weren’t using the notes.
Why?
- The formatting was still too complex for our extranet.
- The AI was summarizing too aggressively, stripping out details that were important for context.
- Even with automation, there was no real buy-in.
How We Finally Got AI Note-Taking Right
It took five months of testing, tweaking, and getting feedback before we finally got the process right. We adjusted our custom GPT to balance summaries with key details, stripped out unnecessary markup, and made sure the process was as seamless as possible.
But the biggest lesson? Tools alone don’t change behavior.
For AI adoption to stick, there needs to be:
- Ongoing feedback: We had to regularly check in to see what was working and what wasn’t.
- Follow-through: Just setting up a tool isn’t enough. We had to make sure people were actually using it.
- Authority to drive change: Someone with authority over the process needs to champion the change.
- Iteration: Refinement takes time. Expect to make several changes along the way.
Where We Are Now
After months of refining, our AI note-taking process finally works the way we need it to. We’re getting the right level of detail, in the right format, and without creating extra work.
But if there’s one thing we’ve learned, it’s this: AI tools (or any technology for that matter!) are never “set it and forget it.” They require tuning, testing, and continuous feedback to actually deliver value.
We’d love to hear from you. Have you experimented with AI for note-taking? What worked and what didn’t? Let’s compare notes!
