An AI workshop for large teams does not exist as one event. Providers advertise hands-on training “for teams of 10 to 200”, then cap the hands-on room at 20 or 30 in their own format tables, so 200 staff is eight to ten sessions, run in an order you choose, across most of a quarter. Budget $45,000 to $120,000 at published rates, or $225 to $600 a head.
That reframing is the whole job. Three decisions carry the outcome: how many people sit in one room, which rooms go first, and how long the gap between first cohort and last runs. Price follows from those three, and most rollouts get quoted before any is settled. That is how a training line doubles between proposal and invoice.
The number the vendor page will not print
Read a provider’s marketing copy and you get one range: training for teams of 10 to 200. Read the same provider’s format table and it splits in two. These capacities were on their own pages on October 22, 2026.
| Provider | Format | Published capacity |
|---|---|---|
| Institute of Business AI | Full day / half day | 10 to 20 / 10 to 30 |
| Institute of Business AI | Multi-day program | 20 to 200, run as cohorts |
| Agent Accelerator | Half day; second facilitator above 30 | up to 50 |
| Correlation One | 201/301 cohorts | up to 40 per cohort |
| Correlation One | 101/102 single sessions | 25 to 300 |
| Teamland | Standard session | 10 to 50; multi-cohort above 200 |
| NEU Media Group | Capped day | up to 20 |
The ceiling tracks how hands-on the format is. Correlation One will put 300 people through a single introductory session; its seven-week cohorts, where people build things, stop at 40. One provider, both numbers, and the gap is the part that matters.
A briefing scales. Practice does not, because practice means a facilitator reading over a shoulder the moment a prompt goes wrong. Published guidance converges on 8 to 20 for a session with real exercises, 12 to 16 if everyone’s work gets looked at.
So the cohort math for 200 people is fixed before you talk to anyone:
| Room cap | Sessions for 200 staff | What that format is |
|---|---|---|
| 20 | 10 | Hands-on full day |
| 30 | 7 | Half day, two facilitators |
| 50 | 4 | Half day, breakouts, thinner feedback |
| 300 | 1 | A briefing, not training |
Pick the cap and you have picked the session count, the calendar, and most of the cost. A provider who answers “how many can you take” with “as many as you like” is quoting the briefing and letting you hear a workshop. Half day or full day is its own call, and it moves the session count by three either way.
Ten sessions, not one, and what that does to the price
Take the ten-session version. A capped full day for up to 20 people is published at $4,500, and a bespoke day built on your own documents reaches about $12,000 once a second facilitator is in the room. Ten of those at list is $45,000 to $120,000, or $225 to $600 a head. Published bands for an AI workshop for teams of 15 to 30 run wider, to $50,000 for a heavily customized day, but a rollout priced at that ceiling is a different purchase and should be scoped as one.
Now the part worth negotiating. A corporate AI workshop has two cost blocks that behave differently across a rollout. Design is fixed: somebody reads your intake forms, client emails, and board decks and builds drills around them, once, for ten cohorts. Delivery is variable and does not compress, because the tenth room needs a facilitator for a full day like the first.
You can see the shape in how vendors price duration. Two providers charge 1.5x and 1.8x their half-day rate for double the clock, which only makes sense if a fixed design block is spread over more delivery hours. The same logic runs sideways across cohorts, so sessions two through ten should not cost what session one costs.
Ask for the rollout as one engagement with design priced once, and expect the rate after the first session near two-thirds of the opener. Ten sessions quoted as ten separate days is a provider billing you ten times for reading your documents once, so understand how a single day is priced before you take a quote.
Then the number your CFO reaches for. ATD’s 2026 State of the Industry report, drawn from 340 organizations, puts direct learning expenditure at $846 per employee for 2025, down from $1,254, against 16.7 formal learning hours each. For 200 staff that is roughly $169,000 of annual learning budget at the category average. A $45,000 to $120,000 rollout is a quarter to three-quarters of your whole year of learning spend, and a six-hour session is more than a third of the year’s formal hours in one sitting.
That is not an argument against doing it. It is the reason to be exact about what the sitting produces, because nothing else in the learning budget gets checked at that ratio. The case for the spend is the one that justifies the license bill underneath: paying for seats is not the same as building capability.
Cohort order is the cheapest lever you have
Sequencing costs nothing and moves the result more than anything else in the plan. Most companies default to volunteer order, alphabetical by department, or whoever complained loudest. All three waste cohort one.
Put the management layer one cycle ahead of their teams. Microsoft’s 2026 Work Trend Index gives the clearest evidence for why: organizational factors such as culture, manager support, and talent practices account for more than twice the reported AI impact of individual factors, 67% against 32%. When managers actively modeled AI use, employees reported a 17-point lift in reported AI value, a 22-point lift in critical thinking about their own use, and a 30-point lift in trust in agentic systems. Where managers made room for experiments, people reported up to 20 points higher readiness and were 1.4x more likely to be frequent users.
Read that as a scheduling instruction. A manager who sat through the session two weeks before their team can sanction the new behavior when it surfaces in a one-to-one. One who did not will treat it as a thing that happened to their people on a Thursday. Same content, same price, and the only difference is which week the invite went out.
After the managers, order the rest by how much of their work is text. Teams that draft, summarize, reply, and reconcile all day convert in week one because the material is already in front of them. Put one openly skeptical team in the early group too. A finance or operations lead who came in unconvinced and left with two things they will keep doing is more persuasive than any readout you can write, and we have watched one skeptic’s line in a company channel fill three cohorts.
Leave for last any group whose answer to “what task would this replace” is vague. A team that cannot name the task is not ready, and the seat it wastes could have gone to one that could.
The quarter between the first cohort and the last
Ten sessions at one a week is ten weeks. Add a holiday, a quarter close, and the week your ops floor cannot spare twenty people, and the honest figure is three to four months. That elapsed time is the risk nobody prices.
Two things happen in it. Cohort one’s habits either hardened or drifted, and whichever happened is settled by the time cohort ten sits down. Meanwhile the tools move: the interface learned in week one has shipped a feature by week twelve.
Three ways to spend the gap, in order of how often they get skipped:
Give each cohort one named task to bring back. Not a certificate, not a survey. One recurring piece of work they will do differently next week, picked by them in the room. It is the only artifact that survives Monday, and the design choices that produce it are what catalog training leaves out.
Check in with earlier cohorts while later ones are still training. Thirty minutes at week four, their own examples on screen. It is also where you learn which parts of the session missed, in time to fix them for the cohorts ahead.
Compress the schedule where you can. Two sessions a week with two facilitators halves the gap. It costs more inside one quarter and protects what you already bought. Format is a lever too, since remote delivery takes travel out of the calendar and each format gives up something different.
Somebody internal owns this, whether or not they were asked
At 50 to 500 people there is no Chief AI Officer. The rollout lands on a VP of Operations, a Chief of Staff, or a Head of L&D as a second job, and scheduling alone eats a day a week. Ten cohorts is ten calendar fights with ten managers, each with a reason their week is wrong.
Microsoft’s own guidance for rolling its assistant across a company is worth copying because it is deliberately unglamorous: start with a focused group and high-value, low-risk scenarios, build a network of power users who can support and inspire others, then broaden access as governance matures. No ratios, no targets. One named person per department who is good at this and will field questions.
Two notes from running these. Name that owner in the contract, not after it, and have them sit in cohort one as a participant, not a host. The questions asked in that room are the ones they spend three months answering.
When you stop buying sessions and start buying facilitators
There is a headcount where facilitated delivery stops being the right purchase, and 200 is close to it. Ten sessions runs $45,000 to $120,000 and ends when the last does. Train-the-trainer, which sits in the same format tables at 5 to 15 champions over two to three days, buys people who can run cohort eleven, the new-hire group in March, and the refresher after the next interface change.
Most 200-person companies should want the hybrid. Buy facilitated sessions for the managers and the first two or three business cohorts, where the material is still being tuned and a professional in the room earns the fee. Train six to eight champions in the same window and hand them the back half with the provider on call. You pay for about half the facilitated delivery and keep the capability, which is what matters when headcount is 230 next year.
The version to avoid: buy all ten days, finish in March, have nothing in the building when twenty people start in June.
A 200-person sequence that survives contact
For 200 staff at a 20-person cap, half days for most cohorts, full days for the heaviest text work:
- Weeks 1 to 2. Pick the cap, count the sessions, name the owner, quote it as one engagement with design priced once.
- Week 3. Managers and the leadership layer. Same hands-on content as everyone else, not a separate executive briefing.
- Weeks 4 to 6. Two cohorts: your most text-heavy team, and one openly skeptical team. Collect examples, fix what missed.
- Week 5 onward. Train six to eight champions from those first cohorts while the rollout runs.
- Weeks 7 to 14. The rest, ordered by text volume, two a week where the floor can spare them. Champions co-facilitate from the midpoint.
- Week 4 of each cohort’s clock. Thirty minutes on their named task. The cheapest line in the plan.
- Week 16. Read out what changed, by cohort, against the tasks they named. That readout funds whatever comes next.
Seven steps, one a purchase. The other six are scheduling and attention, about the right ratio for AI enablement at this size.
Frequently asked questions
How many people can you put in one AI workshop?
Between 8 and 20 for a session with real hands-on exercises, and 12 to 16 if everyone’s work gets looked at. Published caps for full-day hands-on AI training sit at 10 to 20, rising to 30 or 50 for half days with breakouts. Anything advertised for 100-plus people is either a briefing or a multi-day program that splits into cohorts anyway.
How long does it take to train 200 people?
Three to four months at one cohort a week, counting the holidays and quarter-end weeks you lose. Two a week with two facilitators brings it to six or seven weeks and lands the cost in one budget period.
What does it cost to run AI workshops for 200 staff?
$45,000 to $120,000 facilitated at published market rates, or $225 to $600 a head. The spread comes from how much of the content is built on your own documents and how many facilitators each room needs.
Should we train everyone or start with a pilot?
Start with the managers plus two business cohorts, then decide. Those sessions tell you what the material needs to become, and you will rewrite part of it. Commit all 200 seats first and you buy ten copies of a first draft.
Is it cheaper per person to train a large team?
Up to the room cap, yes. Past it, no. Per-head cost falls as a session fills because design spreads across more people, then flattens once you split into cohorts, since each needs its own facilitator and day. The floor is the session rate divided by the cap.
Which department should go first?
The management layer, one cycle ahead of their teams. Then your most text-heavy team, paired with one that thinks this is a waste of time. Last: any group that cannot name the task it would do differently.
Can we buy licenses and self-serve courses instead?
For 200 people it costs far less and tends to show up as dashboard activity rather than changed work. Self-serve content complements facilitated sessions well and substitutes for them badly, which is why the question keeps returning a year after the licenses were bought.
When does train-the-trainer beat buying every session?
Past four or five cohorts, which for hands-on formats is about 100 people. Above that, buy facilitated delivery for the early cohorts and internal capability for the rest.
Key takeaways
- 200 staff is 8 to 10 hands-on sessions, not one workshop. Pick the room cap first; session count, calendar, and budget follow.
- Expect $45,000 to $120,000 facilitated, $225 to $600 a head. Quote it as one engagement so design is paid for once, not ten times.
- Sequence managers one cycle ahead of their teams. Microsoft’s 2026 data puts organizational factors at 67% of reported AI impact against 32% for individual ones, and the lever is free.
- Budget three to four months end to end, and spend the gaps on one named task per cohort plus a week-four check-in.
- Train six to eight champions while the rollout is still running, so cohort eleven and next year’s new hires need no new purchase order.
Sizing a rollout at this scale and want the cohort count, sequence, and cost checked against what your team does all day? Book a free AI-readiness call. Half an hour, no deck, and the session count comes back confirmed or corrected.
Arthur Wandzel
