The founder I was talking to last week had eighteen recurring meetings on her calendar. She suspected eight should not exist. She had killed precisely zero. Each had a small, defensible reason behind it. The board wants the operations review. The team likes the Monday stand-up. The supplier expects the Friday catch-up. None of these reasons was wrong on its own. Together they ate her week.
That is the meeting trap. The kill itself feels rude. The relationship cost feels asymmetric. The mini-justification for each meeting is real. So nothing changes, the calendar fills, and the founder works late on the actual job.
This post is about cancelling, not optimising. The killed meeting that earns its place is the one you had to think hardest about killing. AI can help you do that thinking in five minutes per meeting.
Why founders keep meetings they suspect should not exist
The reason is rarely poor calendar discipline. Every recurring meeting carries three separate justifications, and only one of them is operational. The first is the work the meeting nominally does. The second is the relationship maintenance bundled into it. The third is the reassurance that a thing is being attended to. Kill the meeting and you might lose any of the three, and you cannot tell in advance which.
Steven Rogelberg’s research in The Surprising Science of Meetings is consistent on this. Attendees rate around half of their meetings as low value, but the same attendees resist removing specific ones because the cost of being wrong is concentrated and visible while the benefit of removal is diffuse and invisible. The HBR Stop the Meeting Madness piece by Perlow, Hadley and Eun documented the same pattern in 2017 and not much has changed. Shopify’s January 2023 purge, when Tobi Lütke cancelled every standing meeting under three people, only worked because it was indiscriminate. Individual cancellations one at a time would have stalled.
You are not running Shopify, and an indiscriminate purge will damage the relationships your business depends on. What you need is a faster, calmer kill-test.
The five-minute kill-test, prompt and all
The kill-test has four inputs and one output, and it runs inside any decent chat AI in five minutes per meeting. Feed it the meeting name and cadence, the attendee list with rough seniority, the typical agenda or last three real outcomes, and the decisions that currently depend on the meeting. Out comes a structured triage: keep, downgrade, or kill, with the reasoning attached. Run it on one meeting at a time.
The prompt I use, with the founder’s specifics filled in:
Act as a calm, experienced chief of staff. I have a recurring meeting called X, weekly for sixty minutes. Attendees are A, B, C and D, with A and B senior. The typical agenda is the last three outcomes I can name. Decisions depending on it are Y and Z. Tell me, in this order: the real cost per year at conservative billable rates, whether the meeting has a clear decision-owner, whether any dependent decisions could move to async or another forum, and whether you would keep, downgrade, or kill. If downgrade, suggest cadence and length. If kill, suggest a one-line replacement that preserves the relationship.
Shopify’s meeting cost calculator does the cost layer in two clicks. AI does the structural layer that the calculator cannot, the bit that asks whether a decision-owner exists and whether the meeting is doing relational work that should sit elsewhere. Together they give you the kind of structured second opinion you would normally pay an executive coach for. The outputs are usually less binary than founders expect, and a meaningful share of meetings come back as downgrade candidates rather than outright kills.
The downgrade pattern, and why it is the underrated answer
Downgrading is the right answer for around half the meetings you run the kill-test on, and treating it as a quiet win rather than a failed kill matters. The pattern is one of three moves: cadence (weekly to fortnightly to monthly), length (sixty minutes to thirty), or mode (synchronous to async written summary, ideally with a short video walk-through when context is needed).
The Bain real-time reform note made the structural case for this in 2014 and it still holds. Cal Newport’s argument in A World Without Email goes further, that synchronous coordination has become the default because nobody pays the cost of starting a meeting, but everybody pays the attention residue when one ends. Reclaim and Granola have built businesses on automating the async-summary version of the downgrade.
A worked example. The Friday supplier catch-up the founder ran weekly for forty-five minutes was the meeting she felt worst about killing. The kill-test put it as a downgrade: monthly thirty-minute video, plus a shared async note when something material changes. Annual saving in her time alone, around thirty-five hours, the same again for her supplier contact. Both said yes within an hour of the message landing.
The message that makes the cancellation land warmly
The kill is the easy part, the relationship is the hard part, and the message you send is what decides whether the kill costs you nothing or costs you the supplier. The shape that works has four beats: name the meeting and what it has been doing, name what you have noticed about its current value, name the replacement, and offer a low-friction route back if the replacement turns out to be wrong.
AI is genuinely useful here, because the message has to be tuned per recipient. The board chair gets a different note from the operations manager. Drop the four beats and the recipient’s name, role and history into a chat AI and ask it for a warm, direct, British-English message in your voice. Run the output past your own ear before sending. Founders commonly find the AI version is 80% there and they tighten the closing line. That is the right division of labour.
If you are sending five cancellations in a sweep, do them on the same day with the same shape. Drip them over a fortnight and it reads as wavering. Send them in a batch and it reads as considered calendar reform.
The week-four check, and what it tells you
Run the kill-test on five meetings, send the messages, and put a check in your diary for week four. The check has one question. Did the missing meeting cause a real problem, or did you not notice it was gone?
For a meaningful share of kills, the honest answer is the second one. Shopify reported a 25% productivity gain in the months following the purge, and the leaders said in interviews that the most striking thing was how little anyone missed the cancelled meetings. If you find the same in your week-four check, that is your signal to run the kill-test on the next five recurring slots.
For the kills that did cause a real problem, the fix is rarely to put the original meeting back. The fix is to put the specific lost function back. If the cancelled supplier catch-up cost you visibility on inventory, the answer is a weekly five-minute async note, not the forty-five-minute slot. The kill-test has done its job either way, because you now know what the meeting was actually doing rather than what it nominally was for.
If you are starting from a typical owner-operator calendar of fifteen to twenty-five recurring weekly meetings, two passes of the kill-test usually free six to ten hours of focused time per week. That is the reason this sits in the eliminate quadrant of AI for your own work rather than the optimise quadrant. The first move is having fewer meetings, not better ones.
A natural starting point is to run an AI audit of your week first, surface the recurring slots, and then bring the kill-test to the candidates that come out. If you would rather have a second pair of eyes on your calendar before you start cutting, book a conversation.
