Your AI rollout raised the ceiling. Now raise the floor.

6 months after the licences went out, most organisations can name the 3 or 4 people who are flying. They hand AI whole jobs. They’ve built little routines for themselves that nobody asked for. Someone in leadership has started forwarding their work around as an example.

Everyone else uses it as a faster search box.

That split is what a licence rollout produces. A licence raises the ceiling, because the people who already know how to delegate get a far better assistant than they had before. It does nothing for the floor, because nothing in the rollout taught anyone else how.

The split shows up everywhere

EY’s 2025 Work Reimagined Survey asked 15,000 employees and 1,500 employers in 29 countries how they use AI at work. 88% of employees use it, mostly for basic tasks such as search and summarising. 5% use it in advanced ways that change how they work.

So if your power users are a small group, your organisation is normal. The interesting question is what you do about everyone outside that 5%.

Why the ceiling went up on its own

Delegation is a skill people bring with them. Someone who has spent years briefing colleagues, agencies or junior staff already knows the moves. Give the context, say what good looks like, say what to leave out, check what comes back.

Hand that person an AI assistant and they use the same moves on day one.

Someone who has never had to brief anyone types a question, gets an average answer, and decides the tool is average. Both of them got the same licence and the same launch email.

2 ways companies raise the floor

The first is to hire for it. Some companies now ask candidates to do a piece of real work with AI during the interview, and expect every role (marketing and sales included) to use these tools as part of the job. Over a few years of hiring, the floor rises, because everyone who joins is already above it.

The second is to build one internal assistant that everyone asks. It knows the company’s documents, data and processes, sits where people already work (Slack or Teams), and answers for the whole organisation. People who’d never write a good prompt still get a good answer, because someone wrote the context in once.

Both work. Both also assume things most organisations don’t have.

Hiring your way there takes years of open headcount, and it does nothing for the people already on the payroll. An internal assistant needs engineers to build it and keep it current, and most organisations don’t have spare engineers for an internal product.

The third way: teach it on their own work

The people outside the 5% are already employed, already doing their jobs, and already have a licence. The fastest way to move them is to teach them the moves the power users brought with them, on the work they do every week.

The “own work” part matters more than it sounds. A generic course teaches the tool on prepared examples. People nod along, go back to their desks, and send the same one-line prompts on Monday, because nothing in the course looked like their actual Tuesday.

Practising on their own reports, emails and spreadsheets is what makes it stick. They leave with something they’ve already used once on real work, and the second time is easier.

A test for whether your floor has moved

Here’s a question for any manager. Could you say this to your team and have it be fair?

Before you bring me something, try it with AI first.
Only bring me the part it couldn't handle.

If half the team wouldn’t know where to start, the floor hasn’t moved yet. If they’d all manage it, it has.

I like this test because it checks the floor where it matters, in the everyday handover between a manager and a team. Licence counts, login numbers and course completion rates all tell you about the ceiling, or about nothing at all.

If you’re looking after a rollout where a few people are flying and everyone else isn’t, this is what I do on the Power User Gap: find where AI can do real work in your teams’ workflows, then teach the people who haven’t worked it out yet, on that work.