Meeting overload: 23 hours a week and how to stop

The number comes from Stop the Meeting Madness, by Leslie A. Perlow, Constance Noonan Hadley and Eunice Eun, published in Harvard Business Review in July 2017. Their figure is that executives spend an average of nearly 23 hours a week in meetings, compared with less than 10 hours a week in the 1960s.

Two qualifications matter, because this statistic is repeated more often than it is read. It describes executives, not every office worker: an individual contributor with six meetings a week is not failing to hit some norm of 23. And it is drawn from executive calendar data inside an argument about meeting culture, not from a longitudinal national survey. It indicates a real direction of travel; it is not a precision instrument. Quoting it as "the average professional spends 23 hours a week in meetings" inflates it, and the inflated version is the one that circulates.

The article's more useful contribution is a survey of 182 senior managers across a range of industries. Sixty-five percent said meetings kept them from completing their own work, 71% called meetings unproductive and inefficient, and 64% said meetings came at the expense of deep thinking. Almost nobody in that spread is defending the meetings, and the meetings continue anyway. It is not a disagreement, it is an accumulation.

Accumulation happens because a meeting is cheap to add and socially expensive to remove. Creating one requires a reason on a single day; deleting one requires telling a colleague their meeting is not worth your time. So recurring meetings acquire tenure, and the calendar treats a standup created for a project that shipped last year exactly like the decision forum you need on Thursday.

The arithmetic makes this worse than it looks. A weekly one-hour meeting costs roughly 45 hours a year, every year, until somebody deletes it, and it costs that for every person on the invite. A nine-person weekly hour is most of a full-time month. Recurring invites deserve the most scrutiny on the calendar and reliably get the least.

Parkinson's Law, from C. Northcote Parkinson's 1955 essay in The Economist, explains why the survivors expand: work expands to fill the time available for its completion. A 60-minute slot generates 60 minutes of discussion whether or not it had 60 minutes of substance. The reverse also holds: cut the same meeting to 30 minutes and the same decisions usually get made, because the meeting was never sized to the agenda. The agenda was sized to the meeting.

A meeting audit is a scheduled review of every recurring meeting on your calendar, judged on what it actually produced rather than on why it was created. It is the same instrument as a time audit, narrowed to the one category of time you share with other people.

The test is a single question, asked per meeting: if this meeting did not exist, would we create it today, in this form, with these people? Most recurring meetings fail it and survive anyway, because nobody has been asked to answer it out loud. Sorting the answers needs a fixed vocabulary, so every meeting gets exactly one of four verdicts, each with a concrete next action. An audit that ends in analysis is worse than no audit: it costs a month and changes nothing.

Three patterns turn up in almost every audit. The read-out produces information and no decisions: a written update wearing a calendar invite, and the bulk of what async replaces. The decision forum with spectators produces real decisions, but three of nine attendees make them: the meeting earns its slot, the invite list does not. The meeting that is really a conversation traces every useful output back to one attendee, and wants to be a 1:1 or a thread.

Framing matters for the drop verdict. A cancellation proposed as permanent invites a defence of the meeting; one proposed as a six-week trial with a scheduled review invites an experiment. The review is rarely needed, because after six weeks nobody remembers what the meeting was for.

The meetings that survive the audit can still be bad, and the usual reason is that the agenda was written in the first two minutes, out loud, by whoever spoke first. An agenda is not a list of topics. It is a decision, made in advance, about what the meeting is for and how long each part of it gets. Asana's agenda guidance and Harvard Business Review's 2015 piece on designing a meeting agenda converge on roughly the same components, all of which must exist before the meeting starts:

• A stated purpose, in one sentence. Not a subject, an outcome: "decide the launch date" rather than "launch." If it cannot be written in one sentence, the meeting is two meetings or none.
• Topics as questions. HBR's advice is to phrase each item as the question the group needs to answer, which makes it obvious when the item is finished. "Pricing" has no end condition; "do we hold the current price through Q4?" does.
• A type per item: inform, input, or decide. Most meeting friction is a mismatch here, where the chair wanted input and the room thought it was being informed. Marking the type settles it before anyone speaks.
• Time allocations per item, and an owner. Ten minutes on the launch date, five on the support rota. Allocations make the agenda a budget instead of a wish list, and they surface an over-stuffed meeting while there is still time to cut something.
• What people should read or prepare beforehand. Attach it to the invite. A meeting that opens with fifteen minutes of context wasted its first quarter.
• The last five minutes reserved for close-out. No new topics: decisions, owners, next step. This is the single habit that most reliably keeps a meeting inside its own edges, and it is the subject of the decision log.

Send the agenda far enough ahead that someone can decline, and put it somewhere the room can edit, so questions get added in advance and answered in writing. Where the meeting is a standing ritual rather than a working session, the format does the agenda's job instead: the fifteen-minute cap on a standup and the fixed structure of a retrospective exist for exactly this reason, and both are covered in standups and retrospectives.

Async is often sold as a replacement for meetings in general, which is wrong and makes the argument easy to dismiss. It replaces one thing well: the transfer of information from people who have it to people who need it. Writing in Forbes, Laurel Farrer put the share of meetings that could be handled asynchronously at around 40%, and status read-outs are the bulk of that share. Remote.com's case for async adds that a synchronous default slows projects down, because every question waits for the next scheduled window.

What async replaces badly is anything with a live negotiation in it: a decision where two people disagree, difficult feedback, a design discussion where the shape of the answer is not yet known, the first conversation with someone new. Forcing those into a document produces a thread that runs four days and ends in a meeting anyway.

The most useful public worked example is the GitLab handbook, readable by anyone, describing how an all-remote company of thousands across dozens of countries actually operates. Four of its habits transfer to a co-located team without adopting anything else:

• Documentation is the default, and the meeting is the exception. GitLab's handbook-first principle means a decision, process or answer goes into the written record, and pointing at the record replaces repeating it.
• Every meeting has an agenda document created before it, and it is where the meeting happens. Attendees add questions in advance and notes are typed in live, so the record exists the moment the meeting ends rather than being reconstructed afterwards.
• Meetings are recorded, and attendance is optional where it can be. Once the recording and the document exist, attending is a choice about whether you need to speak, not a duty to stay informed.
• Reply time is a stated expectation, not a hope. Async fails when nobody knows whether a comment will be answered in an hour or a week, because the safe response to that uncertainty is to book a meeting.

Async is not the absence of a norm, it is a different norm, and it needs the explicitness that meeting culture gets for free from the calendar. The cultural side of the switch is covered in async communication culture.

Cutting meetings is only half of it: the same number costs very different amounts depending on where they sit. Three 30-minute meetings scattered across a day do not cost 90 minutes. They cost 90 minutes plus two unusable 40-minute gaps.

Sophie Leroy's work on attention residue is the mechanism: switching to a new task before the previous one is closed leaves part of your attention on the unfinished one, degrading performance on what comes next. A scattered day pays that cost repeatedly. Microsoft's Human Factors Lab found the physiological version in a 2021 EEG study, where back-to-back video meetings produced a buildup of stress-associated brain activity that short breaks let reset instead of accumulate.

Batching is the response, at two scales. At the week scale, concentrate recurring meetings into one or two days and defend the rest; many teams do this as a named no-meeting day, which works better than an individual policy because it removes the negotiation. At the day scale, stack meetings back to back so that what remains is contiguous rather than perforated.

Two cautions. A fully stacked day still needs breaks inside it: writing in Forbes, Maura Thomas argues for batching meeting time while leaving around ten minutes between meetings, which is also where the Microsoft study points. Setting the calendar default to 25 and 50 minutes builds that in without anyone remembering. And batching without cutting just produces two brutal days, so run the audit first. The mechanics of using the protected side of the week are in time blocking and timeboxing.

The weak point of the procedure is step two. Reconstructing on Friday what four meetings produced this week is exactly the recall that favours the meetings you enjoyed, and an audit built on it audits impressions. Two habits fix most of that: writing decisions and owners down in the last five minutes while everyone is still there, and keeping the agenda document as the notes document, so there is one artefact per meeting rather than none.

A meeting audit needs a record of what was actually said, and that record has to exist before you know you want it. Earkeep is a local-first, always-on transcriber that produces one: it records mic and system audio continuously, transcribes locally on the device, never writes audio to disk, and stores transcripts as plain files you can search. Meetings are defined afterwards by selecting a span on a timeline, so nothing depends on remembering to press record. The workflows built on that are described in meeting workflows.