Should your team learn AI together? A manager's practical rollout playbook

A department head I spoke to had done everything the playbooks tell you to. He'd bought his team licences to a well-regarded AI course, sent a warm all-hands email about "investing in our people," and put a line about it in the quarterly goals. Three months later he checked the dashboard. Average completion across the team: under fifteen percent. The keenest person had finished. Everyone else had watched one or two videos and quietly drifted back to doing their jobs the old way.

He hadn't bought the wrong course. He'd bought the wrong shape of learning. And it's worth understanding exactly why that shape fails before spending another budget on it.

Why self-serve courses stall

Hand someone a course and a login, and you've created a task with three quiet killers built in.

No accountability. "Finish it sometime this quarter" competes with every actual deadline they have — and loses, every time, because the course has no due date that matters and no one is waiting on it. Optional learning is the first thing dropped when the week gets busy. It is always a busy week.

No shared context. Each person learns alone, from generic examples that have nothing to do with the team's actual work. There's no one to ask when they get stuck, no shared vocabulary forming, no "oh, you could use that for the Tuesday report" moment because nobody's in the room to make it.

No transfer to the real job. This is the deepest one. The course teaches AI on the course's tidy examples. The employee's real work is messier — different tools, awkward data, edge cases the tutorial never mentioned. The gap between "I watched a video about this" and "I can do this with our actual stuff" rarely gets crossed on your own. So even the few who finish often don't change anything.

Self-serve courses fail not because the content is bad, but because individual, optional, context-free learning is the wrong shape for a team. Fix the shape and the same people suddenly succeed.

Start with ONE real workflow in ONE team

The instinct is to roll out AI training to everyone at once. Resist it. Breadth at the start guarantees shallowness everywhere.

Instead, pick one team and one workflow — a single recurring task that's genuinely painful and genuinely common. The weekly report that eats a morning. The first-draft responses that pile up. The data clean-up everyone dreads. Something concrete, frequent, and annoying enough that improving it is obviously worth the effort.

Why so narrow? Because a focused win does three things a broad rollout can't:

• It produces a real, visible result — a task that's measurably faster or better — instead of a vague sense that people "know more about AI."
• It creates internal proof and internal champions. When the rest of the organisation sees one team actually working differently, you get pull instead of push. People ask to be next.
• It's learnable. A team can genuinely master one workflow in a few weeks. It cannot master "AI" in any amount of time, because that's not a thing you can finish.

Win one specific workflow completely before you widen. Depth first, breadth later.

The weekly 45-minute cohort

Here's the format that consistently works where solo courses fail: a small group, the same people, meeting for about 45 minutes once a week, learning the same thing together and applying it to their actual work between sessions.

Why together beats alone, concretely:

• The calendar invite is the accountability. A standing weekly slot with colleagues is a commitment people keep, in a way "finish the course sometime" never is. You show up because others are showing up.
• Shared context compounds. When one person discovers something useful, the whole group hears it. Questions get answered in the room. A common vocabulary forms, so people can actually help each other the other six days of the week.
• It's social, and social sticks. Learning alongside people you know is simply more motivating than grinding through videos in isolation at 9pm. Momentum becomes a group property.

Forty-five minutes is deliberate. It's short enough to protect against "we can't spare the time" and long enough to actually do something hands-on. Keep it small — a handful of people who share the workflow you chose — so everyone stays involved rather than spectating.

Hands-on beats theory, every time

The single biggest predictor of whether team training translates into changed behaviour: did they practise on their own real tools and tasks, or on generic examples?

Generic AI training teaches concepts on toy problems. People nod, understand it in the abstract, and then can't bridge to their actual messy work — the same transfer gap that kills solo courses, just in a room. So flip it. Spend each session working through the team's actual workflow with whatever AI tools they'll really use. Real data, real edge cases, real "wait, ours doesn't look like that" moments.

This is the heart of it: when the practice is the work, there's no transfer gap to cross. They're not learning AI and then trying to apply it later — they're improving Thursday's report on Tuesday, with help in the room. Building those practical, applied skills around real tasks is exactly what the AI & automation track is built around, and it's why it lands differently than a video library.

Measuring without vanity metrics

Course completion is a vanity metric. So is "hours of training delivered" and "number of people who attended." They measure activity, not change, and they're how that department head ended up reporting a successful initiative that changed nothing.

Tie your measurement to the workflow you chose. Honest questions to ask:

That last row is the real test. If the team would happily keep doing it after you stopped mandating it, it worked. If it quietly dies the moment you look away, it didn't — no matter what the completion dashboard says. You're aiming for a habit, not an attendance record.

Build it in-house, or bring someone in?

You can absolutely run this with internal people, and sometimes you should — especially if you already have someone who knows both the AI tools and your specific work, and has the time to lead a cohort properly. Don't bring in an outsider to do something you can do better in-house.

A focused outside expert tends to earn their keep in a few specific situations:

• You don't have the internal expertise yet. No one on the team is far enough ahead to teach the rest, and waiting for someone to figure it out solo is slower and more expensive than it looks.
• You want adoption to actually happen, soon. An experienced guide compresses the timeline and steers the group around the dead ends — the difference between "we tried AI" and "we changed how we work."
• Your work is domain-specific. Generic corporate training won't touch your actual tools and tasks. Someone who'll learn your workflow and teach against it is a different proposition entirely.

The honest principle: bring in outside help to accelerate and de-risk adoption, not to replace your team's ownership of it. The goal is always that your people end up able to run without the help. If a provider's incentive is to keep you dependent, that's the wrong provider.

If you want to see how I think about this for a working group specifically, the teams page lays it out, and the professionals page covers the individual version for the people on that team.

The recap

• Self-serve courses stall because individual, optional, context-free learning is the wrong shape — no accountability, no shared context, no transfer to real work.
• Start narrow: one real workflow, one team. A focused win creates proof and champions; "roll out AI to everyone" creates nothing.
• Use a weekly ~45-minute cohort. The calendar invite is the accountability; learning together beats learning alone.
• Practise on the team's actual tools and tasks, not generic examples — that's what closes the gap between knowing and doing.
• Measure change, not activity. The real test: would they keep doing it if you stopped pushing?
• Bring in outside help to accelerate adoption, not to own it. You want your team running on its own afterward.

That department head eventually re-ran it the right way: one team, the weekly report, forty-five minutes a Thursday. Within a month the report took a fraction of the time, and — more tellingly — the team kept doing it after the sessions ended, because it had become simply how they worked. That's what success actually looks like. It doesn't show up on a completion dashboard.

If you're weighing whether to run something like this with your group, that's exactly the kind of thing a focused 1-on-1 or small-cohort engagement is built for — happy to talk through whether it's even the right fit before anyone commits.