Everyone’s suddenly got an AI note-taker sitting in on their meetings. Good. But if all it’s doing is handing you a tidy summary at the end, you’ve bought yourself a very expensive notepad.
The note was never the valuable bit.
I made that point in a keynote at Jellis Craig last week, and it still holds: what matters is the compounding.
One meeting captured is a record. Meetings captured over months, searchable, sitting in the one place, start showing you things you cannot see from inside a single conversation.
Where the team is actually strong. Where it’s weak. Which conversations keep happening because nobody ever closed them the first time.
I heard Josh Phegan talk about his daily directions meeting rhythm at a Domain event recently: align everyone in the morning, check progress at 3 pm, never leave a meeting without clear next steps.
Solid discipline. In plenty of agencies, it’s still running on memory and whatever got scribbled in a notebook, and the only person who remembers what was agreed is whoever wrote it down fastest.
Give that same discipline a memory layer, and it stops being a meeting habit and starts being an asset.
You didn’t hear half of it, and that’s normal
Here’s the bit people underestimate. I interview people for a living, and what still catches me out is listening back to a conversation from a day or two ago and thinking, wow, I don’t remember them saying that at all.
Not because I wasn’t listening – but hearing and listening are often different things. When you are sitting in front of a vendor, are you even a little bit distracted by the school recital or the fire that’s about to go off in your inbox? (We’ve all been there!)
A recording doesn’t have that problem. It surfaces the things and the nuances you missed the first time round, not because the AI is smarter than you, but because it wasn’t thinking about the recital.
That’s the real value of a tool like Granola (which I use), or a transcript dropped into ChatGPT or Claude. Not the record. The recall. And recall is only worth anything if you go and ask it questions, which is exactly where most people stop.
The before, during, after
If you want tomorrow morning’s stand-up to be AI-first, here’s the shape of it. Nothing complicated, nothing you don’t already have.
Before. Get consent. Not implied, not assumed, actually let the team know why you think this is the best way forward. (That’s step zero whether it’s your team or a customer!)
Then whoever’s running the meeting looks back at what the actions were last time and chases them before anyone sits down. Don’t wait for the meeting to find out nothing happened.
Look at my notes from the last daily directions meeting. List every action that was assigned, who owns it and what the deadline was. Then flag anything that hasn't been mentioned since, so I can chase it before we start.
During. Talk about the things that actually need talking about. Directions, blockers, wins, whatever’s live. The AI’s job is to disappear into the background and capture, not to become the meeting.
After. Owners, deadlines, next contact, captured properly rather than scribbled on whoever’s notepad was closest. Then feed it straight back into tomorrow’s “before” step. I do this after any meeting, not just the stand-up: open a tool that can see your notes and your CRM and ask it directly.
Here's today's meeting. Give me four lists, one line each: every commitment made and who made it, anything that needs updating in the CRM, any new meetings or appraisals to book, and anything raised that we said we'd come back to and didn't.
If you’ve got consent and your office allows it, connect the tools so that comes back in one instruction instead of three separate logins.
Consent, review, discuss, capture, follow up, repeat. The value isn’t in any single step. It’s that the last step feeds the first, every day, so nothing gets to quietly die in a notebook.
Who owns it? Whoever’s running the meeting owns both ends of the loop, the review before and the capture after. If that rotates in your agency, the ownership rotates with it. What shouldn’t move is that someone is always named against it.
Consent isn’t a compliance box, it’s a strength
The OAIC’s guidance on AI and privacy sits: consent shouldn’t be assumed just because people were told a meeting might be recorded. You ask, properly, before you record anyone. And if the tool picks up something sensitive that wasn’t necessary to the meeting, that gets dealt with, not filed away.
I’d rather treat asking as a strength than a formality. Said out loud in a team meeting, it isn’t a disclaimer. It’s a leader telling their team they take the job seriously enough to remember every commitment made in the room. Frame it that way and you’ll get a yes, because it isn’t surveillance and everyone in the room can tell the difference.
Day 1, day 30, day 90
Nobody’s run a controlled study measuring real estate teams at day one, 30 and 90 of using an AI note-taker, so I won’t pretend there’s a formula. What I can give you is the rhythm I’m currently aiming for. Treat it as a practical benchmark for your own rollout, not a proven threshold.
Day one is a consent-and-capture day, not a data day. Nothing sensitive gets recorded until every participant has actually said yes.
Thirty days in, you should be searching, not just recalling. Enough loop-backs logged that you can check who closed their actions last month versus who just talks about closing them.
Across all my team meeting notes from the last 30 days, show me every action item that was raised more than once. Group them by person. I'm looking for the things that keep getting re-committed to and never closed.
Ninety days in, you’ve got enough for a coaching conversation built on evidence rather than a single incident. One missed deadline is a bad week. The same gap across three months is a pattern, and now you can have that conversation properly.
Look across the last 90 days of meeting notes for [name]. Where do they consistently deliver, and where does the same issue turn up more than twice? Give me the evidence for each with dates. Don't tell me what to do about it.
What actually compounds
There’s real research behind why a shared record beats any single note. Work on transactive memory, essentially a team’s shared system for knowing who knows what and being able to retrieve it, has found a moderate positive link to team outcomes across dozens of studies (Fausett et al., 2026). Other work is more cautious, finding the effect leans more on cohesion than on hard performance numbers (Zhou & Pazos, 2020). So the logic holds, but I wouldn’t oversell the size of it. A genuine lever, not a guarantee.
The proxy that’s real for me is simpler. If you can instantly recall every action from last week, the week runs smoother, and you get room to decide what to delegate and what to hand to AI instead of spending Monday working out what got decided the Monday before. And if the same question keeps surfacing inside a fortnight, that’s usually not a team problem. It’s a memory problem. Check the record before you assume it’s a people issue.
What the AI should never own
The AI captures. The AI chases actions. It does not run your judgment calls. Working out what a pattern means for the team, and having the coaching conversation off the back of it, stays entirely human. The tool’s job is to make sure nothing gets missed. Yours is to decide what matters once it’s found.
Steal these
Four prompts, in the order you’d use them.
1. Before the meeting.
Look at my notes from the last [meeting name]. List every action that was assigned, who owns it and what the deadline was. Then flag anything that hasn't been mentioned since, so I can chase it before we start.
2. After the meeting.
Here's today's meeting. Give me four lists, one line each: every commitment made and who made it, anything that needs updating in the CRM, any new meetings or appraisals to book, and anything raised that we said we'd come back to and didn't.
3. What you talked past.
Read this transcript back and tell me what I didn't respond to. Anything someone raised that got skipped over, talked past, or agreed to without a next step.
4. The pattern check.
Across all my meeting notes from the last 30 days, show me every action item raised more than once, grouped by person. I'm looking for what keeps getting re-committed to and never closed.
Consent first. Name who owns both ends of the loop. Let the meeting itself stay exactly as unremarkable as it’s always been.
Not a smarter transcript. A memory that’s still working while you’re in the next meeting.
With the vendor’s explicit consent, try this on the last appraisal you recorded:
Here's the transcript of this morning's appraisal. In the vendor's own words, what did they say about price, about timing, and about why they're selling? What did they say they were worried about? And what did I not ask them?
That last question is the one that makes you better. And follow up accordingly.