Weekly AI office hours are worth buying, and the version most teams run is worth almost nothing. The gap between the two is a design problem, not a scheduling one.
The case for the format is simple. A workshop ends on a Thursday and the work it was meant to change starts on a Monday, with a real deadline and nobody to ask. A standing half hour where anyone can bring the task that is fighting them closes that gap more cheaply than any other follow-up. Buy six weeks of it, weekly, starting the week after the session.
The case against the default version has numbers behind it. Office hours are a borrowed format, and the universities that invented them have measured their own take-up for years: two thirds of students never attend. The reasons have little to do with the calendar, and they bite harder on adults asked to admit in public that they cannot work a tool their employer just paid for.
The week after the workshop is where the spend goes missing
Dose is the part of AI enablement that gets cut first and matters most. BCG surveyed more than 10,600 leaders, managers and frontline employees across 11 countries and found that only one third said they had been properly trained, with regular use rising sharply at five or more hours of training plus access to in-person training and coaching. A one-day session gets you most of the hours. The coaching half is what a single workshop cannot deliver, because coaching happens on live work and live work happens later.
The backdrop is worse than most buyers assume. Gallup’s May 2026 reading has 15% of US employees using AI daily and 30% using it a few times a week or more, with only 25% saying their organization has communicated a clear plan for integrating it. A team that has had one good session is already ahead of that field. What it lacks is somewhere to go on week two.
That is the whole argument for AI office hours, and it is narrow. They are not training: they teach nothing in sequence and are no use as a first exposure. They exist to catch the first three attempts someone makes on their own work, the window where a habit either holds or quietly stops.
The empty room is the default outcome
Almost every provider recommending office hours says weekly. Few say what happens in week three. Here is the evidence they skip.
Smith, Chen, Berndtson, Burson and Griffin surveyed 625 undergraduates at a large public research university about a format that is free, standing, staffed and advertised in every syllabus. Only one third described using office hours even once a semester, and two thirds reported never using them. The qualitative half is the useful part. Asked what would make them more likely to attend, 57% of 724 comments were a question back at the researchers: what are office hours for, how would I use them, what would I bring. A further 37% framed the format as reserved for emergencies. One answer gave the paper its title: office hours are kind of weird.
Read that as a buyer: the barrier is not time or goodwill. People stay away because they cannot picture a legitimate reason to come, and because showing up files a problem they would rather not file. The people running the room read it the other way. In a Chapman University survey, 40% of students named scheduling while only 16% of instructors agreed, and over a quarter blamed a lack of student effort.
Now add the barrier a campus does not have. Microsoft and LinkedIn surveyed 31,000 people across 31 countries and found 52% of those using AI at work reluctant to admit using it on their most important tasks, and 53% worried it made them look replaceable. Half your room is already managing an impression. Then someone books a public session whose entry ticket is a confession that the tool is winning.
An empty AI office hour is not a sign your team does not need help. It is the format behaving as the evidence predicts when nobody redesigns it.
Six weeks weekly, then a decision
Weekly is right, open-ended is not. Across 150 organizations, the share of employees still applying what they were trained on ran at 62% right away and 44% at six months, so most of the loss lands early and so should the support. Six half-hours is three hours of someone’s time, and the only commitment a budget holder approves without a fight.
| Weeks after the session | Shape | What it is for |
|---|---|---|
| 1 to 2 | Pre-seeded with questions collected at the workshop | Catching first attempts before anyone decides it does not work for their job |
| 3 to 4 | One person screen-shares a live task, the room watches and edits | Moving from asking to practicing |
| 5 to 6 | Attendees bring wins as well as problems | Building the stock of examples the team reuses for a year |
| 7 onward | Continue fortnightly, fold into an existing meeting, or stop | Evidence, not habit |
The test at week six is specific. If more than a third of trained staff attended at least twice, continue every other week with an internal owner. If the same four people filled every session, you have four enthusiasts rather than a team habit, so switch to a short clinic per function. If almost nobody came, stop: a standing empty call costs more in signal than it saves in budget.
One provider puts a public boundary on that window. McCandless Consulting bundles three weeks of group office hours at two hours a week plus two hours of individual coaching into a per-person price, and expires all of it 30 days after the session. The expiry is the smart part: it pushes use into the window where use helps.
Seven design choices that decide whether anyone comes
Each of these answers something in the evidence above. None is a matter of taste.
Seed the first two sessions before the workshop ends. Collect three or four real questions in the last ten minutes of the session, with names attached. An open floor on day one guarantees the silence that kills the series.
Make the ask concrete. Name what people should bring: a document, a draft, the report they rewrite every month. The 57% finding says the barrier is not knowing what counts as a question, so answer that in the invitation, in one line, with examples.
Call it something else. Office hours carries a remedial smell from a context most adults remember. A working session, a clinic, a build hour. The name sets who feels entitled to attend.
Put it inside the working week, not on top of it. Thirty minutes attached to a meeting that already exists beats a separate invitation at an awkward hour. A session competing with real work loses to real work by week three.
Fix the stakes out loud. Say on the invitation and again at the start that nothing from the room reaches managers and nothing is logged against anyone. If the people worried about looking replaceable are the ones you need, the venue has to be visibly safe first.
Rotate who asks. Pick a volunteer each week and work their task in front of everyone. Watching a colleague struggle for five minutes does more for permission than any policy.
Keep something from each session. A prompt that worked, written where the team can find it. Without this you are running support calls; with it you are building material that survives a staff change, which is the ownership question to settle before the first session, as we set out in the piece on who keeps the curriculum once the trainer leaves.
All seven cost less than a second workshop, and renaming the slot costs nothing. The logic under them is the logic that makes a session land at all, which we work through in our guide to designing a session that changes Monday morning.
Who runs the room, and what that costs
Your trainer, for the first few weeks. Worth paying for while questions are still about the tools rather than about your work. Private hands-on AI training sits in a market range of roughly $2,000 to $15,000 by length, headcount and customization, and a short run of follow-ups is normally a small add-on, not a second engagement. Ask for it as a line item before you sign.
An internal owner, from week three. Cheaper, better informed about your actual work, and the only version that lasts. It needs one named person with a few hours a week and the standing to rule on what is allowed. Pick them before the workshop so they sit in it with that job in mind.
Both, with a handover. The provider runs the first two, your owner co-hosts weeks three and four, then takes it. The hard part is not skill in the chair, it is knowing which of your files are fair game.
Know this market’s pricing habits before you negotiate. Office hours are nearly always bundled and almost never priced. IntuitionLabs lists office hours and advisory as a core service with no published rate. Enablify sells ongoing access as an annual retainer at $4,000, $10,000 and $25,000 a year, which is a different product: advice to a leader, not weekly practice with a team. The follow-up has no market rate, so it is negotiable, and the best moment to ask is while the quote is still open.
Three numbers, not nine
Most guidance on AI office hours offers nine or ten measures, from questions by category to confidence lift. Nobody tracks nine things about a 30-minute call. Track three.
- Attendance as a share of the trained group. You want the share who ever came, not the average headcount.
- Repeat attendance. How many came twice or more. One visit is curiosity, two is a habit forming, and this is the continuation signal.
- One timed task. Pick a recurring job, time it before the workshop and again at week six, and write both numbers down. License dashboards will not tell you whether capability moved.
The third number is the one that survives a budget review. The same discipline settles the rest of the schedule, including whether you need an annual refresher at all.
When weekly office hours are the wrong purchase
When the team has not had a session. Office hours are reinforcement. With no shared starting point, the room fills with beginners asking different questions and collapses into slow one-to-one support.
When the group is under about six people. Paired desk time is better: 20 minutes with each person on their own work, no room needed.
When the work spans time zones or shifts. A standing live slot excludes whoever is asleep. An async channel with a named owner who answers inside a day does the same job and leaves a searchable record. The same trade-off makes a video library a weak follow-up, as we unpack in the piece on recording a session and what it costs you.
When no one will own it. An unowned recurring call dies in four weeks and teaches the team that the AI effort is theater. Better to run nothing, which is the broader point behind our argument that buying licenses is not the same as building capability.
Frequently asked questions
Do weekly AI office hours actually work?
They work when the gap is real and the design accounts for low attendance. BCG’s survey of more than 10,600 workers found regular AI use rising at five or more hours of training combined with in-person coaching, and office hours are the cheapest way to supply that coaching. The format itself has a measured take-up problem, so the design choices decide the outcome more than the cadence does.
How long should an AI office hour be?
Thirty minutes. Long enough for two live tasks, short enough that people come while curious rather than while free. An hour looks more generous and draws fewer people, because an hour must be defended against other work.
Weekly or every two weeks?
Weekly for the first six weeks, then fortnightly if attendance justifies it. Skill loss is front-loaded, so the early sessions do most of the work. Dropping to fortnightly too early means someone who hits a wall on Tuesday waits nine days, by which point they have gone back to the old way.
What do we do if nobody shows up?
Treat it as a design fault, not a verdict on your team. Seed the next session with real questions from two or three people by name, rename the slot, attach it to an existing meeting, and have one volunteer work a live task while the room watches. If attendance stays near zero after two reworked sessions, switch to an async channel or paired desk time.
Should office hours be included in the workshop price?
Ask for it in the proposal. One provider publishes three weeks of group office hours at two hours a week inside its per-person training price, with everything expiring 30 days after the session. Others list office hours as a service with no published rate, which means the follow-up is negotiable. Raise it while the quote is open.
Who should run AI office hours internally?
One named person with a few hours a week, enough standing to rule on what data is allowed, and credibility with the group. Function matters less than those three traits. Name them before the workshop so they attend as the future owner.
Can an async channel replace office hours?
For spread-out teams and shift work, often yes, and it leaves a searchable record a live call does not. What it loses is the watching. Seeing a colleague’s rough first attempt get better in four minutes is the part that transfers, and a thread rarely does that. Teams able to do both should run the channel year-round and the live slot for six weeks.
How do we measure whether office hours are worth continuing?
Three numbers: the share of trained staff who ever attended, the number who attended twice or more, and one recurring task timed before the workshop and again at week six. Repeat attendance is the continuation signal; the timed task is what a finance review accepts.
Are AI office hours the same as ongoing AI coaching?
No. Office hours are a group slot in a fixed window after a session, built to catch early attempts. Coaching is one-to-one, runs longer and is aimed at one person’s job. Office hours cost a fraction as much and reach everyone, which makes them the right first buy for most teams of 50 to 500 people.
Key takeaways
- Buy six weeks of weekly 30-minute sessions starting the week after the workshop, then continue, narrow or stop on what attendance and one timed task tell you.
- The format has a measured attendance problem. Two thirds of surveyed students never used office hours, and 57% of their comments amounted to not knowing what to bring, so the invitation has to answer that.
- Your team has a barrier a campus never had: with 52% of AI users reluctant to admit using it on important work, a public help session needs the stakes removed out loud.
- Seed the first sessions, rename the slot, attach it to an existing meeting, rotate who brings the live task, and write down one prompt that worked each week.
- Office hours are almost never priced on their own, so negotiate them into the workshop quote while it is open, and name the internal owner before the session.
Decide the follow-up before you book the session, because adding it later is harder and costs more. A free AI-readiness call will cover which format suits your team’s size and schedule, what to ask a provider to put in writing, and which owner setup tends to hold.
Arthur Wandzel
