AI training that builds capability, not awareness: what a program should leave behind

Why do AI tools on their own change so little?

Giving people AI tools is not the same as changing their work. A study of 25,000 workers found that, when staff were simply given AI chatbots, the average time saved was 2.8% of work hours (Humlum and Vestergaard, 2025). The tools were there; the redesign of the work was not.

That matters because the change is arriving task by task, not job by job. The International Labour Organization estimates that one in four workers worldwide is in an occupation with some exposure to generative AI, and that most of those jobs will be transformed rather than replaced (ILO, 2025). Transformation at the level of tasks needs training at the level of tasks: people working on what they actually do.

What does the research say works?

Three findings recur across the strongest studies.

Applying AI to real work produces large gains. In a randomised experiment with 453 professionals, those using AI on writing tasks took 40% less time and produced work rated 18% higher in quality (Noy and Zhang, Science, 2023).

Knowing where AI fails is part of the skill. In a randomised study of 758 consultants, AI users produced 40% higher-quality work, 25% faster, on tasks AI handles well, but were 19 percentage points less likely to reach the right answer on a task it handles badly (Dell'Acqua et al., Harvard Business School working paper, 2023). A program that only shows what AI can do leaves people exposed to the second result.

Organisations that redesign work get the value. Among organisations that get the most from AI, three in four have fundamentally redesigned workflows, against one in four of the rest (McKinsey, The state of AI in 2026; Indicative). And direction from the top changes results: where employees see a clear AI strategy, 80% report measurable impact from AI, against 55–60% where it is limited (BCG, AI at Work 2026; Indicative).

Together, these describe what a program has to do: put people on their own tasks, teach them where AI breaks, change a workflow, and bring leaders in early.

What should each audience leave with?

Capability looks different at each level of the organisation, so the output should too.

What does a capability-building day look like?

The structure is simple, and it is the opposite of a lecture:

1. Start from a real task. Each person or table brings work they do every week, and times or scores it before AI is involved.
2. Build on approved tools. The work happens on the organisation's own tools or on models it controls, with realistic sample data.
3. Test on real inputs. Outputs are checked against the cases where AI is likely to fail, not only the easy ones.
4. Agree the approval point. Before anything leaves the room, the team names who signs off and when.
5. Measure again. The same task is timed and scored after, and the result is recorded as the baseline for follow-up.

How can you tell capability from attendance?

Ninety days after a program, four signals separate the two:

• The outputs are in use. The workflow runs, the agent is in pilot, the decision brief was acted on.
• The measure moved. Time and quality on the baseline tasks have changed, and someone recorded it.
• There are owners. Each output has a named person responsible for it.
• Decisions were made. Leaders funded, paused or stopped ideas, with reasons.

If none of these is visible, the organisation bought awareness.

Measure where your own team stands: take the free Readiness Snapshot →