Training people on their own work

Redesign changes the work. People then have to be able to do the new work, with the approved tools, on their own tasks. That's training, and the Deployment Framework is specific about what kind: managers first, in protected hours, on their own real work, with approved tools. Each part of that sentence is there for a reason.

Most AI training doesn't stick

BCG's 2025 survey of more than 10,600 leaders, managers and frontline staff found that only about a third of employees felt their AI training had been enough. The same survey shows what the better training looked like. People who had more hours of it, in person, with a coach, were far more likely to be using AI regularly.

Read that as a pattern, not proof. People who already use AI may go looking for more training. But the same survey points the same way on format: 79% of those trained in person used AI regularly, against 67% of those who weren't, and 84% of those with access to a coach, against 70% without one.

Train on the work people actually do

A course built around someone else's examples asks people to translate it back into their own job, alone, later. Most don't. Practice on their own recurring task skips that step: the hour of training is also an hour of improving the work, and the improvement is still there the next week.

You already have the curriculum. Every task you marked “AI-assist” on the redesign canvas is something someone has to learn to do with the approved tool, checking the output as they go. Start there. Everyone else practises on one recurring task of their own.

Managers first

Managers go first for a practical reason: they coach everyone who comes after. In the framework's role grid, managers coach and run the weekly meeting during Improve. A manager who has redesigned their own work can help someone else redesign theirs. A manager who hasn't tends to fall back on asking how much AI people are using, which is exactly the measure the framework rules out.

In protected hours

Your aim page set protected hours: time each month, on the calendar, for every person to learn and improve their own work. Training happens inside them. Squeezed into spare minutes, it loses to whatever is urgent that day. The sponsor's job in Improve, according to the role grid, is to protect those hours when the quarter gets busy.

Protected hours also set the pace. At four hours a month, five hours of practice takes about six weeks. That's why the cohorts overlap and why managers start on day 15, not day 45.

Plan your cohorts below. The planner reads your protected hours and day 0 from module 1, and suggests what each cohort practises on from your redesign.

Keep the record. It counts

Since 2 February 2025, Article 4 of the EU AI Act has required organisations that use AI to take measures that support AI literacy. That covers their staff and others who use AI on their behalf, such as contractors. Since an amendment in July 2026, the duty doesn't require any particular level of literacy for any individual. No certificate is needed: an internal record of the training is the evidence. That's how the framework treats training records, starting with the baseline session everyone attends at day 0.

The record is simple. For each cohort: the dates, who took part (as a group and a total, not a ranking), the workflow they practised, the approved tool, the hours and the coach. IT, security, compliance or HR keep it, as the role grid says.