Book two sittings of two and a half to three hours, two to three weeks apart, placed in hours your team already does not bill, with a coverage rota so client calls still get answered and a time code agreed before the invite goes out. That is the schedule that survives contact with a billable team.
Almost every guide to buying a corporate AI workshop tells you to pick a half day, avoid month-end, and split the group into cohorts. Fine advice. None of it answers the question a partner or a COO will ask when the date lands in the calendar: what did we stop billing for, and what did it buy.
Until you can answer that in a number, the date keeps moving.
So price the hours first. The fee for a private session is the small line in this decision.
Price the hours before you pick a date
Take the head count going into the room, multiply by the hours, multiply by the effective hourly rate the firm realizes after write-downs. That number, not the vendor quote, is what the session costs.
| Room | Hours in the session | Effective rate | Capacity consumed |
|---|---|---|---|
| 8 people, one function | 3 | $165 | $3,960 |
| 12 people, one function | 3 | $200 | $7,200 |
| 20 people, two cohorts | 3 each | $225 | $13,500 |
Use realized rate rather than list rate. A firm with a $300 standard rate and a 15% realization gap is spending $255 an hour of real capacity, and the people who sign off on training budgets already know the difference.
Two things change once the number is on paper. The fee stops being the headline: a private session at market rates runs about $2,000 to $15,000, which in the middle row above is roughly the same size as the capacity bill sitting next to it. And the choice between a half day and a full day stops being a preference. Doubling the hours doubles the larger of the two numbers, which is why we almost never recommend a seven-hour day to a billable team, and why the half-day and full-day units are worth comparing on their line items before anyone picks a date.
A third of the year is already not billed
Here is the part the scheduling advice leaves out. SPI Research’s 2026 Professional Services Maturity Benchmark, drawn from 509 firms, put billable utilization at 66.4% in 2025, the lowest reading in the study’s history. Attrition sat at 11.4%, and 27.1% of projects used generative AI, up about 40% on the prior year.
Read that first number carefully. A third of the paid year at a typical firm is already going somewhere other than a client invoice: internal meetings, admin, pitch work, bench time between engagements, the training the firm already runs. The question is not whether you can afford to stop billing for three hours. It is which of the hours you already do not bill this should be.
That reframing is what gets a date agreed, and it is honest accounting. Displace client work and you have bought capability with revenue; displace the Thursday internal update that has run for three years out of habit and you have bought it with waste. Most firms hold more of the second kind of hour than they think, and a week-by-week utilization report shows where it sits.
One caution. “We’ll fit it in around the work” is not a schedule. LinkedIn’s 2025 Workplace Learning Report, built on 937 L&D and HR professionals plus 679 learners, puts lack of time and resources at the top of the barrier list for managers and employees alike.
Hours nobody allocated are hours nobody takes. Name them, block them, code them.
Two sittings beat one long day, and the calendar agrees
The split schedule is usually sold as a concession to busy teams. It is also the stronger teaching format, one of the rare cases where the cheap answer and the right answer are the same.
Cepeda, Pashler, Vul, Wixted and Rohrer’s review in Psychological Bulletin pooled 839 assessments across 317 experiments in 184 articles. Of 271 comparisons between massed and spaced study, only 12 failed to favor spacing. The gap that produces best retention also grows with how long you need the material to last, so a two-to-three-week pause before the second sitting is not a compromise on a one-day agenda. It is closer to the shape the evidence points to.
There is an hour floor underneath that, though. BCG’s AI at Work 2025 survey of more than 10,600 people found regular use much higher among those with at least five hours of training and access to in-person sessions and coaching. The 2026 edition, covering close to 12,000 workers, found 36% feeling adequately upskilled and 66% getting limited or no guidance on what to do with the time the tools gave back.
Put the two together and the shape is fixed: five to six contact hours, split across two sittings a few weeks apart, mostly hands-on, with a named use for the recovered time. A single three-hour session with no second date is half a program, and the half you dropped is the one where people bring back the prompt that worked twice and failed the third time. The session design that makes those hours land is a separate craft, but the calendar decision comes first, because an unbooked second sitting never gets booked.
Protect the block or do not book it
Scheduling guides say to minimize disruption. For a billable team the opposite is closer to right: make the hours genuinely unavailable, or cancel.
Microsoft’s 2025 Work Trend Index, Breaking down the infinite workday, measured what a knowledge worker’s day now looks like inside Microsoft 365: an interruption roughly every two minutes, about 275 a day, 117 emails and 153 Teams messages, with meetings starting after 8pm up 16% year over year. Nearly half of employees and just over half of leaders described their work as chaotic and fragmented.
A session dropped into that environment without protection does not teach anybody anything. Three hours with laptops open to client inboxes produces a room of people who heard a demo. The mechanics that work are dull, and they are the whole difference: status set, phones face down, inbox closed, a named person outside the room holding the pager, and a facilitator who stops rather than talking over a half-attending room.
Remote raises the difficulty here rather than lowering it, which surprises people who assume a video session is the easy schedule. It is easier to place and harder to protect, and the on-site and remote comparison takes apart what each format delivers once you account for that.
Build the blackout calendar first
Mark the dates you cannot touch, then offer the vendor what is left. Running it the other way around is how L&D ends up known as the team that booked a workshop during close.
| Window | Why it is closed | Where it falls |
|---|---|---|
| Month-end and quarter-end close | Finance and anyone feeding the close are unavailable, and so is the system of record | Last three and first two working days |
| Busy season | Tax, audit, renewal, open enrollment, peak campaign: utilization targets are at their highest | Sector-specific, known a year out |
| Client deadline weeks | Named delivery dates on live engagements | Pulled from the project plan |
| Year-end and the two weeks around it | Coverage is thin and attention is gone | Mid-December to early January |
| Board and partner meeting weeks | The people who have to model the behavior are not in the building | Fixed in advance |
What is left is usually better than it looks. The windows that tend to work for a billable team are the first full week of a month, the lull after a major delivery lands, and the week a cohort comes off one engagement before the next kicks off. Mid-morning starts beat afternoons, because the energy is better and because an 8am start punishes the people with school runs and clients an hour or three ahead of them.
One more note on timing. Book the first sitting four to six weeks out, not two. You need the coverage rota agreed, the documents gathered, and the date to survive one round of client reschedules without collapsing.
Coverage, not attendance theater
Splitting a team into cohorts gets recommended for group-size reasons, and for a billable firm it has a second use: half the team can cover the desk while the other half trains, so nothing goes unanswered and nobody has a reason to half-attend.
Run the same session twice, two weeks apart, same material. The first cohort becomes the informal help for the second, which costs you nothing and speeds the second group up. Keep each room at 15 per facilitator at most, and sort by function rather than by seniority, so every exercise uses documents the whole room recognizes. Above roughly 50 people the sequencing itself becomes the project, and rolling a program across a few hundred staff has its own cost and ordering logic.
Managers attend and do the exercises. A manager watching from the back of the room is the clearest signal available that these hours are optional, and the team reads it in about four minutes.
Five things to settle before you accept a date
- The time code. Decide now whether the hours land as internal training, business development, or general non-billable, and tell people before the invite. Teams on utilization-linked pay will otherwise assume the hours come out of their own number, and the quiet ones will simply not show up.
- The attendance rule, in writing. Who may decline, and who cannot. “Everyone unless a client deadline says otherwise” is a rule. “Please come if you can” is a 50% room.
- The make-up path. Someone will get pulled. Name in advance where they go: the other cohort, or the follow-up clinic with a short catch-up beforehand. Without a path, the first pull-out becomes the precedent.
- Who gathers the documents. The session runs on the team’s own work, so three or four real files have to arrive a week ahead, redacted if needed. Give that to one named person, not to the group.
- The second date. Put both sittings in the calendar in the same action, while the budget is open. Almost nobody books a follow-up two weeks after a session that went well.
What the catalog will and will not schedule
Worth knowing before you spend a month trying to make published training fit a billable calendar. Checked against current public pages this month:
| Provider | Published unit | Published price | Scheduling terms |
|---|---|---|---|
| Certstaffix | Half-day business overview, four one-day live online titles | $350 per person per title | Public schedule plus private online or onsite; no dates, time zones, or group minimum published |
| Section | 1.5 hours live instruction, with recordings | $195 list, or membership at $62.50 per month billed annually | Fixed public sittings |
| Google AI Essentials on Coursera | Five self-paced courses, under 10 hours total | $49 per month after a 7-day trial | Self-directed |
Everything published is a fixed public sitting or a self-paced library. Both are per-seat products, and neither can be placed on a date you choose with a cohort you pick, the only unit that works when the room has to be covered. Private delivery exists at all three and is quoted on a call, so the comparison a billable buyer most needs is the one the websites cannot give. Which of these formats transfers a skill rather than a vocabulary is a separate question, covered in our look at what works for non-technical teams.
The gap is the point. Buy the hours you can protect, from someone who will place them where your utilization report says they belong, and judge the result against the task you trained rather than against a completion rate. That standard is what separates AI enablement from attendance, and it is the whole argument for treating this as capability you build rather than licenses you issue.
Frequently asked questions
When should we schedule an AI workshop for a team that bills by the hour?
Four to six weeks out, in the first full week of a month, mid-morning, outside close and busy season. The best windows in practice are the lull right after a delivery lands and the week a cohort moves between engagements. Mark the blackout dates before you offer the vendor anything.
How long should the session be if everyone is on client work?
Two and a half to three hours per sitting, two sittings, two to three weeks apart. That reaches the five to six contact hours BCG ties to regular use when they come with in-person coaching, and it never takes a full day of capacity off the board at once.
What does the session cost us in lost billable time?
Head count times hours times realized hourly rate. Twelve people for three hours at a $200 effective rate is $7,200 of capacity, which is often larger than the workshop fee itself. Do that arithmetic before the date conversation, because it is the first question a partner group asks.
Should we run one session for everybody or split into cohorts?
Split, for two reasons. Rooms over about 15 per facilitator stop getting individual correction, and a split lets half the team cover the desk while the other half trains. Run the same material twice, two weeks apart, and let the first group be the help desk for the second.
How do we code the hours in the time system?
Pick one code, decide before the invite goes out, and say so plainly. Internal training is the cleanest. What matters is that nobody on a utilization-linked bonus is left guessing whether attending costs them money, because the ones who assume it does will quietly decline.
What happens when someone gets pulled into a client call mid-session?
They lose the thread, and in a three-hour hands-on session they do not get it back by rejoining at hour two. Name the make-up path up front: the other cohort, or the follow-up clinic with a short catch-up first. Also name the one person outside the room who takes the escalations so this happens once rather than six times.
Can we just run it after hours or on a Saturday instead?
Running it after hours protects the utilization number and undermines the session. Microsoft’s 2025 index already has meetings after 8pm up 16% year over year, and evening attendance is poor, resentful attendance. Training outside working hours also tells the team the firm does not count this as work, which is the opposite of the message a program needs.
Does remote scheduling make this easier?
Easier to place, harder to protect. A video session removes travel and lets you run two shorter sittings without anyone leaving their desk, and it also puts every attendee two clicks from a client inbox. If you go remote, tighten the room size and insist on cameras and closed inboxes rather than assuming the format will carry itself.
Is there a point in the year we should avoid entirely?
Mid-December through early January, and whatever your sector calls busy season. Coverage is thin, attention is gone, and a session nobody remembers in February is capacity spent twice.
Key takeaways
- Price the capacity before you pick a date: head count times hours times realized rate. At twelve people and a $200 effective rate, a three-hour session consumes about $7,200, which usually dwarfs the fee.
- SPI’s 2026 benchmark puts billable utilization at 66.4% across 509 firms. A third of the year is already not billed, so the real decision is which non-billable hours this replaces, not whether you can stop billing.
- Two sittings of two and a half to three hours, two to three weeks apart, is both the schedule a billable calendar can absorb and the one the spacing evidence supports. Book both dates in one action.
- Protect the block or cancel it. With an interruption about every two minutes in a normal workday, hours that are not genuinely unavailable buy you a room of people who watched a demo.
- Settle five things before accepting a date: the time code, the attendance rule, the make-up path, who gathers the documents, and the second sitting.
- Treat the calendar as part of the AI enablement decision, not the admin that follows it. A date nobody defended is the most common reason a well-designed session teaches nothing.
Pull your utilization report by week, mark the closed windows, and look at what is left in the first full week of each of the next two months. A free AI-readiness call will tell you which function should take the first pair of sittings and what the hours ought to produce by the second one.
Arthur Wandzel
