The Meeting Was Great. Now What?
How to use AI to stop losing what your team is learning
TL;DR: Most organizations lose more learning than they keep. Not because people don’t care, but because there’s no scaffolding to hold it. AI can be that scaffolding. We can use AI not to replace human judgment, but to keep the raw material accessible (and connected!) so that when you do sit down to make sense of things, you’re not starting from scratch. After all, we’re “only” human and we’ve got a lot going on right now!
“I feel like we keep having the same conversation”
Here’s a scenario most of us know too well. Someone on your team leaves. But, they were the one who remembered why the board shifted strategy two years ago, what came out of that one convening with partners last spring, and which funder conversations led to the current program design. They didn’t write most of it down (who has time when we’re moving so fast?), and what they did document lives in email threads, Google Docs with names like “Notes - FINAL v3,” and a mental map that walked out the door with them.
Or, maybe no one leaves.
Maybe your team meets every month to discuss strategy, and every month you have a version of the same conversation because no one can quite remember what was decided last time, what you tried, or what you learned from it.
This isn’t a failure of commitment or intelligence. It’s a systems, structures, and behaviors problem. The natural constraints of human memory (recency bias, forgetting information as time passes, and the sheer volume of things competing for our attention these days) mean that even thoughtful, well-intentioned teams often lose more institutional knowledge than they keep.
My experiences managing a team for years were enough for me to see this problem close up (and feel directly responsible to make it better!). But I see some version of it across all teams and orgs that I work with. So, I’ve been thinking about this for a long time, and I decided to experiment with it directly over the past year and a half.
What is institutional memory? (And, why should I care?*)
When I say “institutional memory,” I’m not talking only about record-keeping or compliance documentation. I’m talking about something more alive than that: the accumulated learning of a group over time. The questions that evolved. The decisions that were made and why. The patterns that emerged across conversations, across projects, across seasons. The things people said in a meeting that connected to something someone else said three months earlier (but no one noticed because no one could hold both in mind at once).
These are things that documenting and record-keeping enable, but they are the sparks and insights of connection that give us both a-ha moments for our work and enable us to feel connected to an organization’s pulse over time.
For mission-driven organizations, this is particularly important. Our work is cumulative, sometimes chaotic, and often very context-based. We’re trying to understand really complex problems, build relationships over time, learn from what we try, and make better decisions as we go. Often, we’re doing all of this on limited budget and bandwidth. When institutional memory is fragile, as it so often is — held in one person’s head, scattered across platforms, biased toward whatever happened most recently — we undermine our own ability to learn. We end up reinventing, repeating, and losing the very insights that should be compounding.
Where AI fits (it sits**)
Over the past year and a half, I’ve been stewarding an interest group (80ish members, meeting monthly, with regular attendance around ~15 members) and purposefully experimenting with using AI to maintain the group’s institutional memory.
What started as simple documentation turned into something more like a living, queryable knowledge base that grew with every session.
Here’s what the practice actually looked like:
After every meeting, I saved the transcript and chat, de-identified them (removing names and sensitive details) and stored everything in a Claude “project”.
I set up a process (now a “skill”) to help me draft a post-meeting summary based on the meeting transcript and chat. Note: I refined this process over time, trying to balance providing a recap for those who attended while also providing a summary that was more useful than a chain of events. I received feedback that many who couldn’t attend due to scheduling often looked forward to the summaries and found them accessible and useful.
Over time, that project accumulated transcripts, meeting summaries, survey data, planning documents, and retrospective artifacts. I could ask it questions like: What themes have come up across our recent meetings? When did we last discuss XYZ topic? What topics have members raised that we haven’t gotten to yet? How have the group’s questions evolved since we started?
It’s important to note that I am one person, leading this group as a volunteer, within a group of mission-driven professionals dedicated to embodying principles of emergent learning in their work (so…a pretty good audience to work with!).
Those queries informed how I designed the next meeting. They helped me connect threads across sessions that I wouldn’t have caught relying on my own memory. And when we did a year-end retrospective, the AI could compare our current state against what we’d said we wanted to accomplish twelve months earlier (with the actual evidence, not just my recollection of it).
The AI was incredibly valuable to me, as a steward of this group, to hold the knowledge in one large bucket that I could scoop from as necessary with my own questions and observations. I’ve used the analogy of Dumbledore’s pensieve before, and it always comes up when I refer to this work. All of this information is swirling about and I know things are in there, but it’s up to me to reach in and pull out the threads.
In other words: AI didn’t do the learning. It held the ingredients. It enabled me, and the group, to begin our conversations with our past learnings in mind. The meaning-making (deciding what mattered right now, which insight to surface and which to hold for later, how to bring a pattern to the group in a way they could actually engage with) was entirely human.
Is this something that good old-fashioned note-taking could have enabled? Maybe. But even in the best note-taking circumstances, I find that it can be difficult to bring in all of the elements that make this memory repository useful. It’s hard to bring in Zoom chats into notes, for example. But notes are often from the perspective of the note-taker, who may be biased by their role or expertise. This method collects everything, such that conversations threads that didn’t grab my attention months ago may connect to things happening in the group now.
I love using cooking metaphor when I talk about this work: AI offers mise en place. It preps your kitchen. It lowers the activation energy so that when you sit down to do the real work, you’re not starting from scratch. But it does not cook the meal. It doesn’t envision the culinary experience. You’re still the chef. You’re just ready to focus on what you do best.
Mise en place for meetings: store, query, and connect your ingredients
If you want to start building institutional memory with AI, here are my simple tips. It’s lightweight on purpose, because the barrier is often building discipline around doing these things consistently, and not really anything related to the technology. [Though, to be clear, the tool you choose and practices you put in place to support the psychological safety of the group, confidentiality of information, and accessibility of information are all things you should consider and get explicit about.]
Store consistently. Save your meeting transcripts, chat logs, key decisions, and reflections somewhere you can return to. It doesn’t have to be perfect (please don’t let “I need a system first” become the reason you never start). It can be messy. The goal is just to start and to do it consistently. Most major AI tools now support having dedicated project spaces, folders, or containers where you can accumulate related documents over time. Drop things in as you go. (If you want to boost the power of your system, build a practice of taking a voice note or typing up your reflections after a meeting and use that as additional fodder for the repository.)
Query intentionally. This is where the value, and your expertise, actually lives. Don’t just store things and forget about them. I hear so often people lament about AI meeting notes no one ever looks at again. Don’t be that person! Drop them in the folder (or automate it!). Build that habit. Then, before your next meeting, or before a planning session or strategy conversation, go to that accumulated material and ask it questions. Not just “what happened last time?” (though that’s fine) but things like: What themes are emerging across the last several conversations? What questions keep coming up that we haven’t addressed? What did we say we’d try, and have we talked about what happened? The questions you ask matter more than the technology you use to ask them.
Connect your expertise. The AI will surface patterns. Always. It will always surface something. Some will be useful. Some will be wrong (I’ll get to that). Your job is to take what the AI surfaces and decide: Does this feel right? Is this the thing the team needs to see right now? How do I bring this back to the group in a way that’s useful, not overwhelming?
This connecting step is where institutional memory becomes institutional learning. Without it, you just have a well-organized archive. But with it, you have a practice that compounds over time. One that, I’d argue (and I’ve felt) enables you to go into those next gatherings ready to connect and be with people and their ideas, knowing it’s all going to get saved and stored and feed back into the group’s shared brain.
Where the AI was “meh”
Of course this experiment wasn’t perfect. After all, the point was to learn!
Over the time I’ve been using this, there have been points where the AI confabulated facts with confidence. It produced wrong tool names, incorrect timelines, and misattributed details that all looked right. Part of this was due to my note-handling (I deidentify lots of things out of an abundance of caution, so it will extrapolate to fill in the gaps). I caught them because I knew the answers. A reader wouldn’t have. It also never flagged which of its claims were drawn from direct evidence versus inferred (everything arrived with equal confidence). I could correct for some of this in my instructions to it, which is something I’m doing in my recent updates.
At one point, it completely missed the group’s “cultural” identity". I asked another AI tool to create a year-end visual summary based on the meeting transcripts. It produced something neon, techie, and corporate-looking. I rejected it immediately: “That’s not us.” Months of transcripts hadn’t given the AI a sense of who we were (only what we discussed). When I brought it back to Claude with my feedback and asked “What’s our version of a year-in-review?”, it suggested something organic, gardening-aligned, and warm. That felt right.
Another time, the AI suggested we split the community into beginner and advanced tracks. I knew this was absolutely not the right move. And it was a good thing I didn’t listen, because in our reflection on year one, the beginner-majority dynamic ended up being one of the group’s greatest strengths. The AI was simply pattern-matching from other communities of practice models and completely missing what made ours work at this moment in time.
The takeaway isn’t “don’t trust AI.” The takeaway is: you are the sense-maker. The AI holds pieces. You decide what they mean and what you do with them.
Ok, so where should I try this?
This practice isn’t specific to interest groups or communities of practice (that’s just where I happened to build it). Here’s where I think it could matter for mission-driven organizations more broadly, in all kinds of meetings and projects:
Leadership teams tracking strategic decisions. Imagine being able to query: What were the key arguments for and against this direction when we discussed it last year? What assumptions were we making? Instead of reconstructing from memory or scattered notes, you have the actual conversation to return to.
Program teams connecting learning across cycles. If your team runs programs in cohorts or cycles, each one generates insights. But how much of what you learned in Cohort 3 actually informs the design of Cohort 5? A queryable repository of reflections, debriefs, and participant feedback could make those connections visible. You could ask things like: What were the biggest pieces of feedback from participants? How did program staff think the program could improve in future iterations?
Cross-organizational coalitions. If you’re part of a coalition or learning network where multiple organizations come together periodically, the continuity problem is even more acute and tricky. People and organizations may rotate in and out, even though you’re all focusing on the same wicked problem(s). Conversations might happen months apart. Having a shared institutional memory (with appropriate privacy protections) could be the thing that keeps learning building rather than resetting. It can also increase accessibility of information and widen the queries to include multiple stakeholders if it is presented in a shareable form (such as NotebookLM).
The common thread is this: wherever you or your organization’s work involves learning over time (and whose doesn’t?), something that can hold the pieces across conversations, across months, across transitions is genuinely valuable for the human(s) in the lead.
Start small
If any of this resonates, here are three micro-moves. Pick one!
Record your next three team meetings and save the transcripts. That’s it. You don’t need to do anything else yet. Just start accumulating the raw material. Transcripts are what I think of as the “primary sources” of organizational learning (they’re the closest thing you’ll have to a running record of what was actually said, discussed, and decided). If you want my best practices on how to use AI notetakers and proper “etiquette”, you can get them here.
After your next important meeting, record a two-minute voice note. Reflect on what you heard, what you noticed, the dynamics in the room (things that don’t show up in a transcript). Drop it alongside the transcript. You’re adding the human layer the AI can’t capture. I promise it’ll be very valuable down the road!
After you have three or four meetings saved, ask one question. Put the transcripts into an AI project and try: What themes are coming up across these conversations? What keeps coming up that we haven’t resolved? See what it produces. Check it against your own sense of things. Notice where it’s useful and where it’s off. Experiment with adding supporting documents to its knowledge by giving it information about the members of the group, the goals of the group, or the group’s main charge.
Those are starting points. Not a complicated system. Just a practice amplified by intentional documentation.
Questions this use surfaces
Throughout all of this experimentation, it did raise some of the harder questions that I’m still considering as I continue to use this approach to steward the group.
For example, when AI synthesizes a group’s learning and presents it back, is the meaning-making happening when the AI synthesizes or when the humans discuss the synthesis? How do humans need to discuss/consume the synthesis so that it furthers their learning? I don’t think it’s totally separable, but the distinction matters for how we think about what “learning” actually is in this context. And how we think about organizational or collective learning may be different than individual learning.
There’s also an accessibility tension: AI makes thoroughness easy. You can generate comprehensive analyses, detailed syntheses, exhaustive reports. But “thorough” and “useful” aren’t the same thing. If a thorough analysis produces a 130-page report because AI makes it possible to dive into the information in so many ways, has it served the work or overwhelmed the audience? What does “fit-for-purpose” mean when AI removes the natural constraint of human memory and processing time?
These are the kinds of questions I think are worth considering rather than rushing to answer. And they are probably worth considering with the humans in your group to understand their needs and desires. These questions are at the intersection of how we use these tools and how we understand learning itself, but I think it’s easier to begin to tangle these answers apart when we’re actively experimenting and reflecting on what the tool is/isn’t offering.
I believe that the organizations that will learn most effectively in this moment are the ones that build practices (not just tools, not just documentation habits, but practices) for holding their knowledge across time and making it available for the humans who need to make sense of it. Whether they use AI or not is a choice, but I do think there are some genuinely useful ways to leverage it to augment group learning and I find that super exciting.
*Hi, fellow Bluey/Unicorse fans!
**AI isn’t my big orange cat, but if it was…
I’d love to hear from you. Has your organization found ways to hold institutional knowledge that actually work? Or does this hit a nerve because you’ve watched collective learning languish?
And if you know someone whose team could use a second brain for their institutional memory, I’d be grateful if you’d share this with them.
💡Did you know?🌱 Helping mission-driven leaders, teams, and organizations thoughtfully and responsibly adopt AI is what I do! If you want to learn more, please reach out!
To stay up-to-date on my writings across platforms and access additional guides and resources, please join my occasional mailing list (no spam, ever!).