Enablement6 min read

Capability That Outlasts the Consultants: Building AI Skills That Stick

Training days are easy to run and easy to forget. Real capability gets built in real work, by the right people at the right depth, and it keeps going after the programme ends.

A group working through an exercise on laptops in a workshop
On this page 10 sections
  1. The Monday after the workshop
  2. Start with an honest baseline
  3. Different roles need very different depth
  4. Learn on real work, and ship something
  5. Champions keep the momentum going
  6. Leaders have to go first
  7. Measure behaviour, not attendance
  8. Common mistakes we see
  9. What the first 90 days can look like
  10. The real goal

The Monday after the workshop

Most of us have been to one. A full-day AI workshop, a good facilitator, some genuinely eye-opening demos. People leave buzzing. A few post about it. The feedback scores are excellent.

Then Monday arrives. The inbox is full, the quarter-end report is due, and the new tools sit untouched in a browser tab. Within six weeks, almost nothing has changed in how the work actually gets done.

This isn’t because people weren’t interested, or because the content was poor. It is because capability was treated as an event rather than a change in how people work. Real capability shows up somewhere else entirely: in work that gets shipped, in decisions made faster and better, and in teams that keep improving without anyone from outside telling them to.

Here is what we have found makes the difference, starting before the first session is even planned.

Start with an honest baseline

Before designing any programme, it helps to know where people actually are, not where the organisation chart assumes they are.

A short, anonymous survey and a handful of conversations usually reveal a lot. Some teams are already using AI tools quietly, sometimes in ways that worry the security team. Others haven’t touched them and are slightly anxious about what they mean for their jobs. A few managers are enthusiastic; others are sceptical and haven’t said so out loud.

All of that is useful. It tells you where to start, who your early champions might be, which concerns need addressing openly, and which policies need to catch up with reality.

Different roles need very different depth

One programme for everyone sounds fair and efficient. In practice it tends to serve nobody well. The executives find it too technical, the engineers find it too shallow, and everyone in the middle wonders what it has to do with their job.

Think about what different people actually need:

  • An executive doesn’t need to write prompts all day. They need to judge which use cases are worth funding, ask the right questions about risk, and know what good looks like when a team presents results.
  • An analyst or operations lead needs hands-on fluency with the tools in their daily work, and the confidence to redesign a process rather than just speed up the old one.
  • An engineer needs to build, evaluate, secure and run AI systems, which is a genuinely different discipline from traditional software.

That is why we structure capability in levels: awareness for everyone, then practitioner, innovator, strategist and, for a small group, transformation leadership. Each person learns what their role actually needs, and can see what the next level looks like if they want to grow.

Capability isn’t an event. It shows up in work that gets shipped, in better decisions, and in teams that keep improving without being told to.

Learn on real work, and ship something

The fastest way to build capability is to use it straight away on a real problem that matters to the person learning.

So instead of generic exercises, each cohort works on use cases from its own function. A finance team might automate part of its month-end commentary. A procurement team might build an assistant that checks supplier documents against policy. An HR team might rethink how it answers the hundred most common employee questions.

And the measure of success is simple: did you ship something? Not a slide, not a demo, but a change to how work is done, used by real people.

This has a useful side effect. Every cohort produces small, visible wins, with names attached. Those wins do more to build support for larger AI investments than any strategy deck. People believe their colleagues far more readily than they believe consultants.

What a good cohort project looks like
  • It comes from the team’s own backlog, not a list of suggestions.
  • It can show a result within a few weeks.
  • It has a manager who wants it to work.
  • It touches real data, with proper permissions.

Champions keep the momentum going

Every programme ends. What happens next depends almost entirely on people inside the business.

A network of AI champions, drawn from every function and level, becomes the connective tissue. They answer the everyday questions that would never justify a ticket. They share what is working in one team with another. They notice the next good opportunity because they understand both the work and the tools.

But champions don’t happen by accident, and they don’t survive on goodwill alone. They need three things:

  • Time. A few hours a month, formally recognised, not squeezed in around everything else.
  • Recognition. Visible acknowledgement from leadership, and a role that counts in their development.
  • A direct line to the people running the AI programme, so their insights actually change priorities.

Given those, champions become the most effective adoption channel an organisation has. Without them, the energy from even the best programme drains away within a few months.

Leaders have to go first

There is one thing no programme can do on its own: convince people that AI really matters to how they will be judged.

That comes from leaders. When a senior leader openly uses AI in their own work, talks about what they learned, and asks “Did you try AI for this?” in a normal team meeting, it sends a stronger signal than any internal campaign. When leaders delegate AI to “the digital team” and never mention it again, people draw their own conclusions.

So we always include leaders early, in sessions designed for them, not as an afterthought. Not to turn them into experts, but so they can lead the change with confidence rather than enthusiasm alone.

Measure behaviour, not attendance

Completion rates and satisfaction scores are easy to collect and tell you almost nothing about capability. A hundred percent attendance at a programme nobody applies is still a failure.

Better measures are about behaviour and results:

  • How many people use AI tools in their daily work, a month and a quarter after training?
  • How many use cases moved from idea to something people actually use?
  • What did those use cases change: time saved, quality improved, customers served faster?
  • How many teams are now improving their own processes without outside help?

Track these over quarters, not weeks, and report them alongside the business results AI is meant to deliver. It keeps everyone honest, including us.

Common mistakes we see

Well-intentioned programmes tend to go wrong in predictable ways:

  • Tools before purpose. Licences are rolled out to everyone with no guidance on what to use them for, so usage spikes and then fades.
  • Policy as a wall. A long list of things people must not do, with nothing about what they can and should do. People either ignore it or stop experimenting.
  • Ignoring the anxiety. Nobody addresses the obvious question of what AI means for people’s roles, so it gets answered by rumour instead.
  • No time to practise. People are told to learn AI on top of a full workload, and quite reasonably, they don’t.
  • Success measured by headcount trained. Which tells you how busy the trainers were, not whether anything changed.

What the first 90 days can look like

Every organisation is different, but a useful first quarter often follows a shape like this:

Weeks 1–3: Listen and plan

Baseline survey and interviews, a short session with leaders, agreement on two or three priority functions and the outcomes that matter to them.

Weeks 4–8: First cohorts

Awareness sessions for everyone in scope, plus practitioner cohorts in the priority functions, each working on a real use case from its own backlog.

Weeks 9–12: Ship and spread

Cohort projects go live with real users. Champions are named and given time. Results are shared openly, and the next wave of functions and use cases is chosen based on what worked.

By the end of the quarter you should have a handful of things in production, a group of people who can explain them, and a clear plan for the next quarter. That is a far stronger position than a thousand completed e-learning modules.

The real goal

We say this to every client, and we mean it: the best outcome of an enablement programme is that you need us less.

Not because we did a poor job, but because your people can now see opportunities, build and run solutions, and lead the change themselves. That is what lasting capability looks like. Everything else is a training day.

Key takeaways

  • Match depth to role, from awareness for everyone to leadership for a few.
  • Learn on real use cases from the team’s own backlog, and ship something.
  • Give champions time, recognition and a direct line to the programme.
  • Get leaders using AI visibly, early.
  • Measure adoption and outcomes over quarters, not attendance.

Case studies & examples

References & further reading

  1. AI competency framework for teachersUNESCO, 2024
  2. AI competency framework for studentsUNESCO, 2024
  3. The Future of Jobs Report 2025World Economic Forum, 2025
  4. EU AI Act, Article 4 (AI literacy), as amendedAI Act Explorer

Links checked on 1 November 2026. External sources are listed for further reading; Delphilens is not affiliated with them.