AI adoption starts
with the work

Why does AI adoption stay low after a tool rollout, and what actually works?

What actually works is offering three to five concrete scenarios per role as a ready-made starting point, such as a button, a template or a pre-filled prompt. Adoption spreads through colleagues, with champions per team, and the advice is to measure by role or team instead of using one adoption rate for the whole organisation.

Does low AI adoption mean employees aren't ready for it yet?

Almost always not: the conclusion that employees aren't ready yet is, for low AI adoption, almost always wrong. According to the article, the problem lies in how the rollout was designed, not in people's willingness. When in doubt about what's allowed, employees do nothing, and that's sensible behaviour when the rules are unclear.

Is one adoption rate for the whole organisation enough to track AI adoption?

No. One adoption rate for the whole organisation hides, for AI adoption, where it does and doesn't work. A measurement by role or team looks at weekly use, which scenario is used, and what happens to people who tried it once and didn't come back.

Are complaints about a tool's limitations a bad sign for AI adoption?

No. Complaints about limitations mean the tool is being used, and count as a sign that AI adoption is on the right track. Silence usually means employees have given up.

Source: article by Dave Willemstein. Updated: October 2026.

The pattern that repeats everywhere

The rollout follows the textbook. Licences for everyone, a kick-off, a training session, an intranet page with tips. After three months: eleven per cent weekly active users. After six months: fourteen per cent, mostly the same people.

The conclusion that usually follows ("our people aren't ready for this yet") is almost always wrong. Those same people have learned to use new apps in their personal lives without any training. The problem sits in how the introduction was designed, not in people's willingness.

Why an empty input box doesn't work

A general AI assistant offers infinite possibilities and no starting point at all. For someone with half an hour between two meetings, "you can do anything with it" is the same as "I don't know where to start, so I'll do it the way I always have."

There's a second barrier too, discussed less often: uncertainty about whether it's allowed. Can I put customer information into this? What if the output is wrong and I used it anyway? Will I be judged for using this, or for not using it? When in doubt, people do nothing. That's not resistance, it's sensible behaviour when the rules are unclear.

What does work: scenarios instead of possibilities

The approach that makes the biggest difference in practice is also the dullest: replace "here's a tool" with "here are four tasks you do often, and here's how to handle them now."

Concretely, for a consulting organisation:

• Turning a call recording into an action list in the standard format.
• Rewriting a standard quote for this specific client's context.
• Searching past files for similar questions.
• Reducing a long document to its three key decisions.

Each scenario becomes a ready-made starting point: a button, a template, a pre-filled prompt. The user doesn't have to think anything up, just add their own material.

How do you choose those scenarios? Not by asking "where could AI help?", because that produces speculation. Instead, ask "which task from last week cost you more time than it was worth?" That produces a usable list within an hour per team.

Adoption spreads through colleagues

Training delivers knowledge. Behaviour change comes from the colleague two desks over who shows how they did something in five minutes instead of an hour.

What helps with that:

• Champions per team, not per organisation. Someone who does the same work as you is more credible than an expert from the innovation department.
• A place where people share working examples. Low-threshold, with real examples from their own work. This is worth more than a manual and needs less upkeep.
• Room to practise without risk. An environment where mistakes have no consequences, explicitly flagged as such.

Measure by role, not in total

One adoption rate for the whole organisation is a useless number. It hides the fact that the legal team is at 70% and the field team at 4%. And those two need a completely different answer.

Break your measurement down by role or team, and look at three things: how many people use it weekly, which scenario they use, and what happens to people who tried it once and didn't come back. That last group is your most valuable source of information, and the one least often asked.

The conversation that usually gets avoided

For every AI application that saves time, there's someone for whom that task was part of their craft. Sometimes it's the dullest task and everyone's glad. Sometimes it's exactly the work someone was good at, and where they drew their sense of value from.

It doesn't help to skip over that with "AI replaces no one, it just supports you." People hear that more often than they believe it. What does help: naming explicitly what's changing, what replaces it, and who decides that. Even when the answer is uncomfortable.

At best, that conversation leads somewhere different from what was planned. I've seen a team, after exactly that conversation, choose a smaller scope (the tool was used only for preparation, not for the assessment), which meant the time saved was smaller but usage was much higher. Net, that delivered more than the original plan.

Three signs your adoption is on the right track

• People use it for tasks that weren't on your list. That means they've grasped the principle and you're past the scenario stage.
• Complaints about limitations start coming in. Complaints mean usage. Silence usually means people have given up.
• New employees get it explained by colleagues, not by IT. Then it's become part of how work happens here. And that's the only goal that counts.