An AI training rollout is the sequence of decisions that gets a whole workforce using the same AI tools properly: who owns it, which teams go first, what the training replaces, and what happens when somebody gets it wrong.
To roll out AI training across a 100-person team: name one accountable owner, agree what the training takes off the workload before you add it, pilot with two departments for six weeks, then expand department by department rather than in one company-wide cohort. Expect about a quarter to reach measurable behaviour change, not an afternoon.
The hard part is not the teaching. Most rollouts add AI to a job nobody removed anything from, then read the resulting non-adoption as a skills gap. It is a capacity gap, and more workshops do not close it.
What needs to be in place before you start?
Three things: an approved tool list, a written policy saying what data may go into those tools, and a named owner with budget. Without them, training teaches people to use tools they are not sure they are allowed to use, on data they are not sure they can paste.
Keep the approved list short. Two or three tools, chosen because they fit the work: Microsoft 365 Copilot if the organisation already runs on Microsoft, Claude or ChatGPT Enterprise for drafting and analysis, one automation tool. Curation is the point. A list of 40 approved tools is not governance, it is a menu.
There is also a legal floor now. Article 4 of the EU AI Act has required providers and deployers to ensure "a sufficient level of AI literacy" among their staff since 2 February 2025, and it applies to every deployer, not only those running high-risk systems. If you operate in the EU or sell into it, the rollout stopped being discretionary eighteen months ago. If you do not, it is still the clearest available benchmark for what "trained" is supposed to mean.
Get the written AI policy done first. It takes two to three weeks and it makes every later conversation shorter. If you have not yet worked out where the team actually is, the readiness model behind training a team in AI covers the assessment layer that sits underneath everything below.
Who should own the rollout internally?
One person, named, with a direct line to the executive team. In a 100 to 500 person organisation that is usually the head of L&D or the people director, with the COO as sponsor and one champion per department doing the day-to-day work.
It should not sit with IT alone. IT can approve tools and configure them; it cannot decide what a finance analyst should stop doing. Nor should it sit with a committee, because a committee cannot be chased.
The sponsor matters more than the org chart suggests. Microsoft's 2026 Work Trend Index, which surveyed 20,000 full-time knowledge workers across 10 countries between 18 February and 7 April 2026, found employees reported a 17-point lift in the value they got from AI where their managers actively modelled using it. A rollout in which the leadership team exempts itself is already behind.
What should AI training replace, rather than add?
Every department should name one task the training makes redundant, before the training happens. Not "we will be more efficient". One specific, recurring, nameable task with an hours-per-week figure beside it, which stops being done the old way.
This is the step nearly every rollout guide skips, and it is why so many programmes stall around week six. The evidence points the same way. Microsoft's analysis tested 29 variables and found organisational factors, meaning culture, manager support and talent practices, accounted for 67% of the reported impact of AI against 32% for individual factors. Training the individual is the smaller half of the problem.
In practice: the customer support team agrees that first drafts of tier-one ticket responses are now AI-written and human-approved. Finance agrees the monthly commentary starts from a generated draft. Marketing agrees the weekly report is no longer written from scratch. Each is a small, checkable commitment, and each frees the time the training needs to land.
If a department cannot name the task, that department is not ready. Send it to the back of the queue rather than through the programme. A team trained with no time to practise produces a certificate and no change.
Which departments should go first?
Two of them: one high-volume, text-heavy team, and one sceptical team with a respected manager. The first gives you a fast, visible result. The second gives you credibility, because the people most likely to dismiss AI will not be persuaded by the enthusiasm of the team that was always going to enjoy it.
In most scale-ups the first is customer support, sales development or marketing. The second is often finance, legal or operations. Avoid opening with the engineering team, who have usually adopted something already and will tell you the training is beneath them. They will be about half right.
Sequence everyone else by how much of their work is text and how repeatable it is, not by seniority. The leadership team goes early regardless, but as participants, not as a separate cohort with a nicer lunch.
What does a six-week two-department pilot look like?
Six weeks, two departments, one measurable outcome each. That shape works because it is long enough for a habit to form and short enough to get approved without a board paper.
Week 1. Baseline. Measure the named task as it is done today: time taken, volume, quality bar. Without that number the pilot cannot be judged, so it will be judged on impressions instead.
Weeks 2 to 3. Two live sessions per department, built on that department's real work rather than generic prompting exercises. Between them, people apply it to live tasks with somewhere to ask questions.
Week 4. The awkward week. The novelty has gone and the old method is still faster for some people. This is when the champion earns the title.
Weeks 5 to 6. Re-measure the baseline task. Write down what worked, what did not, and the two things you will change before the next department starts.
Budget three to four hours of contact time per person across the six weeks, plus roughly a day a week of the champion's time. If you cannot protect that, do not start. A pilot everyone joins late and leaves early tells you nothing except that people are busy, which you knew.
How long does a rollout across 100 people take?
Around 16 to 20 weeks from a standing start to the final department, with the first real result at week eight. That assumes the policy work is done and two departments run at a time.
Stage | Elapsed time | What "done" means |
|---|---|---|
Governance and approved tool list | 2 to 3 weeks | Policy signed off, two or three tools approved, owner named |
Pilot, two departments | 6 weeks | One baseline task re-measured in each |
Wave two, two departments | 3 to 4 weeks | Same again, using the pilot's revised material |
Waves three and four | 3 to 4 weeks each | Remaining teams, including leadership |
Embedding and review | First review at 90 days | Named tasks still being done the new way |
Anyone promising a trained 100-person workforce in a fortnight is selling attendance. The training itself can be compressed. The habit cannot, and the habit is the thing you are buying.
The number worth tracking is not course completion. It is how many of the named tasks are still being done the new way three months later. Track that one and most of the other metrics stop mattering.
What happens when someone uses AI wrong?
Decide the answer before it happens, and make it boring. Somebody will paste client data into a consumer chatbot, or send a customer a confidently wrong AI-drafted figure, usually inside the first month.
Write a three-line response path into the policy: report it to a named person within one working day, that person assesses whether data left the organisation and whether the customer needs telling, and the incident becomes a worked example in the next session. No disciplinary framing for a first honest mistake.
Make it boring because punished mistakes go underground, and unreported AI use is the real exposure. A team hiding its ChatGPT tab is far harder to govern than a team that tells you what it tried. Worth remembering that when the Office for National Statistics found 49% of UK businesses with 250 or more employees using at least one AI technology in its June 2026 survey, it was counting organisations, not the informal use going on inside them.
Why don't one-day workshops stick?
Because one day changes what somebody knows and nothing about what they are measured on. On Monday the same targets, the same deadlines and the same manager are waiting, and the fastest route to hitting them is still the old one.
The ONS numbers show the shape of it. Among UK businesses citing a lack of expertise as a barrier to AI, around 62% said they were training or retraining existing staff, while only 11% reported that more than half their workforce had received AI-related training. Plenty of sessions are happening. Not much is reaching most of the people.
A single day works as an opener, and there is a real argument for a live session over a self-paced course, because people ask better questions of a person than of a video. Treat it as week one of six rather than as the programme. The other five weeks are where the habit forms, and they cost nothing except protected time, which is the one thing nobody wants to give up.
The rollout that works is unglamorous. A short tool list, one owner, one named task removed per department, two departments at a time. The training is the easy half.
If you want a second opinion on the sequence before committing a quarter to it, book a call with ivee and bring your department list. The first useful thing we usually do is tell you which two teams to start with, and which one to leave until last.
