AI workshops: which format suits your team, and what a good day contains

What an AI workshop is, and what it is not

An AI workshop is a facilitated session where a group uses AI on its own real work, with the facilitator in the room while they do it. That is the whole definition, and it is what separates a workshop from the three things it gets confused with.

A webinar is one person talking to many. A course is structured content delivered over time, usually self paced. A demonstration is somebody showing you what the tool can do. All three have their place. None of them is a workshop, because in none of them does the learner produce anything from their own work while somebody experienced watches and corrects.

The test to apply is simple. At the end of a workshop, every person should have made something with AI that they will use again the next morning. If the output of the day is notes, it was a webinar.

The five formats, and who each suits

Almost every request that comes in is one of five shapes. Matching the shape to the audience is the single biggest decision, and it is usually made badly, with a whole department sent to a session designed for one team.

Our own one day AI team workshop is the third row, deliberately. If you are booking for a board instead, the considerations are different and AI training for executives covers them. For groups under about eight, AI training for small teams is the same idea scaled down.

What a good one day workshop actually contains

This is the part providers are vague about, so here is the shape that works, in order.

Before the day there is a scoping conversation and a look at real artefacts: the actual report, the actual inbox, the actual spreadsheet. Without that step the day defaults to generic content, which is the most commonly reported failure in the SKAI evidence.

The first block sets a shared baseline. Not a history of AI, but how the model behaves, where it is confidently wrong, and what it cannot see. Thirty to forty minutes, because the room already contains people who use AI daily and people who have never opened it, and both need the same vocabulary.

The first exercise is on their own work. Not a sample dataset, not a puzzle. Somebody's real task, done badly with AI first so the failure is visible, then done properly.

The middle of the day is a task teardown. Pick three recurring tasks the team actually owns, rebuild each one with AI in the room, and time both versions. This is where the value is, and it is the block that gets cut when a provider has a fixed curriculum to get through.

The afternoon turns to guardrails and judgement: what never goes into a prompt, how to check an output that looks plausible, and who owns the decision at the end. The SKAI framework treats responsible use as core content rather than an advanced module, and it is right.

The last hour is build and save. Everyone leaves with a working prompt, template or small workflow, stored somewhere shared rather than in their own history. A workshop whose outputs live in twelve separate chat logs has not changed how the team works.

One thing to insist on: people should work in their own accounts, on their own files, inside their own policy. A workshop run on a demo login teaches people how to use a demo login.

Leadership workshops are a different session

Senior teams are routinely sent to hands on sessions they do not need. A director does not have to get quick at prompting. They have to decide what the organisation buys, what it forbids, what it measures, and what risk it is accepting.

So a leadership workshop covers a different agenda: what the technology genuinely does and does not do, what it costs to run properly, where the legal and reputational exposure sits, and which decisions land on them in the next two quarters. It is shorter, more argumentative, and it should end with decisions written down rather than skills practised.

It also matters more than it looks. The SKAI survey named weak leadership buy in as a gap at 22%, and weak or inconsistent leadership support as one of its thirteen pitfalls. Train a team whose leaders have not engaged and the habit decays within a quarter.

We run these UK wide and in person around Birmingham, alongside our Birmingham AI training and the wider UK programme.

How long, and how many people

On length, the evidence pushes shorter than instinct suggests. The SKAI guide found very short stackable modules of 30 to 90 minutes are often more practical than longer courses, and that modular design lets people enter at different levels and come back as their needs change. A single full day still works, but only when it is scoped to one team and a small number of named tasks.

On group size, hands on breaks down above roughly twenty people in one room without a second facilitator, because the whole value is somebody looking over shoulders. Above that number you are running a lecture and should price it as one.

On time, be honest with yourself before you book. The framework treats paid, protected time as a criterion for the training being reachable at all, and employers in the survey named cost at 42% and limited availability at 37% as the main barriers to participation. A workshop people attend while answering email is a workshop you paid for twice.

What makes it stick

Most workshops produce a good day and no change. The difference is almost entirely in what happens afterwards, and the SKAI guide is specific about it: revisit the training at three and six months so people can reflect on how they are actually using AI and adjust, and review whether confidence, quality and consistency of use and decision making have moved.

Two design details matter as much as the follow up. First, recognise what people already do. The framework says training should build on existing informal or self taught AI use rather than assuming a zero start, and rooms where the confident users are treated as beginners lose them in the first twenty minutes. Second, leave the team something reusable: a use case library and worked examples, so the next person to join inherits the day rather than missing it.

On site or online

Both work. The SKAI guide lists in person sessions, online learning, hybrid models, workplace based training and self paced study as legitimate options to be matched to the audience rather than ranked.

In practice, on site suits teams who need to work shoulder to shoulder on shared documents and benefits from the room being away from the desk. Live online suits distributed teams, costs nothing in travel, and is easier to run in shorter blocks across several weeks, which the evidence mildly favours anyway. We have written up the trade off in more detail in on site versus live online AI training, and if budget is the deciding factor, what corporate AI training costs sets out how the formats are priced.