Do not buy an AI workshop for new hires. Buy about five hours of structured practice across their first two weeks, built from material you already paid someone to make. Give one internal person the job of running it. A cohort session is the right shape only when you hire in waves.
The reason to put it in the first fortnight is arithmetic rather than politeness. The Bureau of Labor Statistics puts 20.6% of US wage and salary workers at a year or less with their current employer as of January 2026. A year after your company-wide session, roughly one person in five on the floor was never in the room, and nobody noticed, because the usage dashboard still looks the same.
The team did not forget what it learned. It was replaced, a few people at a time.
The second reason is what makes onboarding the cheapest AI enablement you will ever buy. Everyone else you train has a way of doing the task already and has to give it up. A new hire does not. For two weeks, the approved assistant can be the first way they ever write the weekly report, and then there is nothing to unlearn.
What goes in the first two weeks
Two touches, about five hours, none of it on the critical path of the role’s own onboarding.
| When | Length | What happens | Who runs it |
|---|---|---|---|
| Day 1 | 15 min | Account on the approved assistant, provisioned by IT with the laptop. One page of data rules, signed. | IT |
| Day 2 or 3 | 45 min | The team’s saved prompts for the three tasks this role does most, run live on a low-stakes document of theirs. | Internal owner |
| Day 4 or 5 | 30 min, plus 15 | One real task done with the assistant, reviewed by a human the same week. | Manager |
| Week 2, early | 60 to 90 min | A second sitting on the hardest recurring task in the role, with whoever does it best. | Peer |
| Week 2, late | 30 min | The review step practiced once: what gets checked before assistant-touched work leaves the building. | Internal owner |
| From week 2 | Standing | They join the existing slot where people bring the task that is fighting them. | Internal owner |
Nothing here is a workshop. It is an account, a page of rules, a prompt library, two sittings, a review habit and a named person to ask. The session you may already have bought is where the material comes from.
Your one-off session stops covering the team you have
Run the numbers on your own roster and the case makes itself. In August 2026 the BLS counted 5.2 million hires and 5.1 million separations in a single month, a separations rate of 3.2%. At that pace, a flat model of the share of your original roster still present looks like this.
| Months after the session | Share of the people trained still there | Read it as |
|---|---|---|
| 6 | about 82% | One in six seats has changed hands |
| 12 | about 68% | A third of the room is new |
| 24 | about 46% | The trained majority is gone |
A flat rate overstates the loss, since separations concentrate among people in their first year. The tenure survey is the better anchor and points the same way: one in five workers has been there under twelve months. Either way the fix is the same. This is the quiet version of the problem we argue in the case against buying licenses and calling it enablement, where license counts hold steady while capability drains.
Decay is the other half and the two compound. Pooled across 53 studies, skill loss reaches d = -1.4 after more than 365 days of nonuse, worst for cognitive, accuracy-based tasks. Writing a usable prompt is one.
So a new hire who waits ten months for the next all-hands session arrives with nothing to refresh, next to colleagues who have been practicing since spring. That mix bores one group and loses the other. It is part of why the annual date is the wrong unit, as we set out in the piece on whether you need a yearly refresher at all.
Week one sets a default whether you choose it or not
A new hire who is not handed an account in week one does not stop using AI. They use the one from their last job. Netskope’s 2026 Cloud and Threat Report still puts 47% of workplace generative-AI users on personal apps rather than company-provisioned accounts, down from a year earlier as employers issued approved tools. That is the point: the share falls when someone hands out an account, not on its own.
So the fifteen minutes on day one is the cheapest control you have, and it belongs with the laptop and the badge rather than in a training calendar. If the account arrives in week six, you have had five weeks of your documents in somebody’s personal history, and you have taught the new hire that the approved tool is optional.
The other half of that fifteen minutes is the data rules. One page, read once, signed: what never goes into a general assistant, what needs a human check before it leaves, who to ask when the answer is unclear. Teams running more than one assistant should name which tool is approved for which kind of work on that same page. A new hire cannot infer it and will otherwise pick by habit, a problem we unpack in the piece on training a team that already runs three different assistants.
If you have EU customers or an EU entity, that signed page plus a dated roster is also the cheapest standing evidence for the AI literacy duty in Article 4 of the EU AI Act, which has applied since 2 February 2025.
Why two touches and not one
One block on day two fails for two measurable reasons.
The first is dose. BCG surveyed more than 10,600 leaders, managers and frontline employees across 11 countries and found that 79% of those with more than five hours of training were regular AI users, against 67% below five hours, with in-person sessions and coaching strengthening the effect. Five hours is not a tool tour. It is a sequence.
The second is spacing. Cepeda, Vul, Rohrer, Wixted and Pashler ran over 1,350 people through a design that varied the gap between study and review and the delay before testing. Performance rises with the gap, peaks, then falls, and the best gap was about 20% of the retention interval at delays of a few weeks. A new hire’s first real review is due within weeks, not quarters, so the second sitting belongs about a week after the first. That is where week two already sits.
Practice on the new hire’s own documents, not a sample set. The design logic is the same as for any session that has to change behavior rather than impressions, and we work through it in the session design guide.
Onboarding does not need a workshop
A corporate AI workshop is a cohort event with a facilitator, a room and a date. Onboarding is continuous and serves one person at a time. The market keeps forcing those two shapes together.
What fits onboarding is material plus an owner: saved prompts by role, a page of data rules, a review step, a standing question slot, and one internal person who keeps all of it current. Once that exists, making the next hire fluent costs an hour or two of somebody’s week.
A cohort session is the right call in three cases:
- You hire in waves. An AI workshop for teams needs a team in the room, and an intake of eight or more at once fills one. Below that you pay setup costs to train three people.
- Nobody internal owns the material yet. The first purchase is then the session that produces the material and trains the owner, not a new-hire module.
- A whole function is new. A new sales team or support desk is not onboarding. It is a first session for a team that happens to be new.
Everything else is better served by the sequence above. Settle who keeps the material before you book anything, because the onboarding version quietly disappears when nobody owns it, the failure mode we trace in the piece on who keeps the curriculum once the trainer leaves.
What it costs
Three ways to pay for onboarding AI enablement, with sharply different unit economics.
Reuse what you already bought. If you ran a private session and asked for the artifacts in the scope, the material exists. Cost per new hire is the owner’s time, one to two hours, plus a quarterly pass to keep prompts and rules current. Most buyers who think they have nothing have this.
Add the artifacts to a session you were buying anyway. Private hands-on AI training sits in a market range of roughly $2,000 to $15,000 by length, headcount and how much is built around your own documents. Role-sorted prompts, an editable data-rules page and a trained internal owner cost little at proposal stage and nothing later.
Buy a packaged onboarding cohort. Capstone Workforce publishes an Onboarding Class at $18,000 per cohort for up to 50 new hires over a 12-week cycle, alongside seat plans from $290 a month for 25 seats. At a full 50 seats that is around $360 a head: sound arithmetic, poor fit. A 200-person company hiring 30 people a year fills that cohort in twenty months, and the tracks on offer are role-readiness practice rather than fluency with the assistants your team runs.
Refuse the fourth option, a vendor session per new-hire batch billed monthly. You pay facilitation setup every time, for three or four people, forever.
What to demand before the first session
Four lines in the scope of work decide whether an onboarding block is possible later. Ask for them before you sign, when they are free.
- The prompt library in editable form, sorted by role. Not a PDF of screenshots. Prompts your team wrote and tested during the session, in a document someone can revise in March.
- Permission to reuse and modify internally, with no per-seat fee for future hires. This is the line that turns a session into an asset. Providers who decline are selling access rather than capability.
- A named internal owner, trained during delivery. Someone who sat beside the facilitator, knows why each prompt is shaped that way, and can run the week-two sitting alone.
- A short facilitator note for the onboarding block. Half a page: the three tasks to practice per role, the order, the common failure. A recording is the weakest artifact here and the one most often offered in place of the others.
The one number worth recording
Pick a task every new hire in the role does, time their first attempt with the assistant in week two, and time it again at day 60. One task, two timings, written down.
Attendance tells you nothing. Neither does license activation, which only says someone logged in. The timed task tells you whether the sequence worked for this role, and the record makes the next budget conversation short. In the BCG data only 36% of employees think the training they got was enough, so the bar is lower than it feels.
When to skip this
Three cases. You hire fewer than about four people a year and already run a short pass each quarter, so the next one catches new arrivals within weeks. Nothing has been built yet, so the first purchase is the session and the owner. The role does not touch the work, true for some floor and field positions, where the signed data-rules page is the whole obligation.
Outside those three, the block pays for itself the first time a new hire’s second week produces a usable draft instead of a question.
Frequently asked questions
Should you run an AI workshop for new hires?
Only if you hire in waves of eight or more at once. Below that you pay facilitation setup to serve three people. For a steady trickle of hires, what works is about five hours of structured practice across the first two weeks, drawn from material the original session produced and run by one internal owner.
How much AI training should a new hire get in the first two weeks?
Around five hours, split across two sittings and a review. BCG found 79% of employees with more than five hours of training were regular users against 67% below that line, which makes five hours a floor rather than a target.
What goes in a new hire’s first week?
An account on the approved assistant provisioned with the laptop, a one-page set of data rules, 45 minutes on the team’s saved prompts for the role’s three commonest tasks, and one low-stakes real task done with the assistant and reviewed by a human the same week.
Why does the second sitting belong in week two?
Spacing research across more than 1,350 participants found the best gap between learning and review runs at roughly 20% of the retention interval at short delays. For a new hire whose first real review is weeks away, that puts the second pass about a week after the first.
Can we reuse our existing workshop material for onboarding?
Usually yes, and it is the cheapest route. It needs the prompt library in editable form, written permission to reuse and modify internally with no per-seat fee for new hires, and one internal person who owns keeping it current. Ask for all three before the first session.
Who should run the onboarding block?
An internal owner, with the manager handling the first task review and a strong peer taking the week-two sitting. Outside facilitation is for cohorts and for building the material. Onboarding runs too often to outsource one arrival at a time.
How much does new-hire AI onboarding cost?
Close to nothing per head once the material exists, since the cost is one to two hours of an internal owner’s time. Packaged cohorts are priced for volume: one published onboarding class runs $18,000 for up to 50 hires over 12 weeks. Private hands-on sessions that produce the material sit in a market range of about $2,000 to $15,000.
What happens if we leave AI out of onboarding?
Two things. Your trained share falls as the roster turns over, with one in five US workers at a year or less of tenure. And new hires default to the account from their last job. Netskope still finds 47% of workplace generative-AI users on personal apps rather than provisioned ones.
Does onboarding help with the EU AI Act literacy duty?
For companies with EU customers or an EU entity, yes. Article 4 has required providers and deployers to ensure proportionate AI literacy among staff since 2 February 2025, and prescribes no format. A dated onboarding block with a roster makes the duty continuous rather than something to prove once a year.
Key takeaways
- A trained team becomes an untrained one by arrival rather than by forgetting. One in five US workers has under a year of tenure, so a year on, roughly that share of the room was never in it.
- A new hire is the cheapest person you will ever make fluent, with no existing way of doing the task to give up.
- Two touches across the first fortnight, about five hours total, clears the dose line where regular use jumps from 67% to 79%.
- Material plus an owner fits onboarding. A cohort session fits hiring waves and building the material in the first place.
- Four scope-of-work lines decide whether any of it is possible: editable prompts by role, internal reuse rights with no per-seat fee, a trained internal owner, and a short facilitator note.
Work out what your onboarding sequence should carry before your next hire starts, not after. A free AI-readiness call will tell you what share of your team your last session still covers, which artifacts you are missing, and whether your hiring pattern calls for a cohort or a sequence.
Arthur Wandzel
