Among European companies that considered AI and decided against it, only one in five said it would not be useful. Seven in ten said they lacked the expertise. The brake on AI adoption is not belief. It is permission and practice.
We have spent fifteen years training and coaching more than 50,000 sellers and leaders across more than a hundred organisations. Over that time we have learned to be careful with one particular meeting: the one where everybody agrees.
The Week Everyone Agreed and Nothing Went Live
This month, at one of the enterprise organisations we work with, an AI rollout was ready to go. The business case was approved. The people who would use it wanted it. Nobody in the room needed convincing.
And then nothing happened for a while. The delay was a data processing agreement going back and forth over two questions: which company handles the data on whose behalf, and which AI model provider sits behind the feature.
Nobody was against it. Everybody was waiting. That difference matters, because the usual response to a stalled rollout is to make the case again, louder. Here that would have achieved nothing. The case had been made. What was missing was somebody who could say yes, and a clear answer to what exactly they were saying yes to.
It Isn’t One Organisation. It’s Most of Them
It would be easy to file that under one legal team being thorough. Europe’s own statistics say otherwise.
In 2025, 19.95% of enterprises in the EU with ten or more employees used at least one AI technology, up 6.47 percentage points on the year before. Eurostat also asked the companies that had considered AI but did not use it why not. They could give more than one reason:
| Reason for not using AI, among EU enterprises that considered it (2025) | Share |
|---|---|
| Lack of relevant expertise | 70.89% |
| Lack of clarity about the legal consequences | 52.52% |
| Concerns about violating data protection and privacy | 48.83% |
| The technology was not considered useful | 20.68% |
Source: Eurostat, “Use of artificial intelligence in enterprises”, data for 2025, extracted December 2025.
Read the bottom row first. Four out of five of these companies did not give usefulness as a reason. What they did give sits with people: we don’t know how to do this, we don’t know whether we’re allowed, and we don’t know who carries the risk.
The Same Hesitation, One Level Down
The pattern repeats inside organisations that have already said yes. Microsoft’s 2026 Work Trend Index surveyed 20,000 knowledge workers who use AI at work, across ten countries including the Netherlands, between February and April this year. The fieldwork was done by Edelman Data x Intelligence.
Two findings stand out. 45% say it feels safer to focus on their current goals than to redesign their work with AI. And only 13% of AI users say they are rewarded for doing that redesign.
There is also a gap in who feels free to experiment. Leaders are more likely than employees to say they feel safe suggesting new ways of working with AI: 81% against 67%.
Organisations wait for legal clarity. Their people wait for a signal that it is safe to try. Both are waiting for permission.
So the stall is not a technology problem at the top and a motivation problem at the bottom. It is the same problem twice. People who are convinced, holding back because nobody has told them clearly what is allowed, who owns the outcome, and whether trying will count in their favour.
What Changed in August
There is now also a legal reason to take the second half of that seriously.
Article 4 of the EU AI Act requires organisations that deploy AI systems to take measures to support the AI literacy of their staff, and of other people who use those systems on their behalf. That obligation has applied since 2 February 2025. According to the European Commission, supervision and enforcement by national authorities apply from 3 August 2026.
The Commission is explicit that this includes ordinary use. Staff who use generative AI to write or translate text should be informed about its specific risks, such as hallucination. No certificate is required; an organisation can keep an internal record of the training and guidance it has given.
In other words, “do our people know how to use this well?” stopped being only a question for the business. Since August it is also a question a regulator can ask.
Two Questions, Both About People
Strip the rollout down and two questions are left. Am I allowed? And do I know how?
Neither is answered by a better tool. The first is answered by a person with the authority to decide: someone in legal or procurement who owns the data processing questions, and someone in the business who owns the result. The second is answered by practice, guidance and leaders who go first.
What stalls a rollout is rarely a “no”. It is two open questions that nobody has put their name against.
What to Settle Before the Kick-Off
If you are about to roll out AI to a commercial team, three things do most of the work:
- Put the processing questions on the table in week one. Which company handles the data, on whose behalf, and which model provider sits behind each feature. Ask for it in writing before your legal team has to ask for it.
- Name who says yes. One person accountable for the outcome, not for the licence. When a question comes up in week three, everyone should know whose desk it lands on.
- Make trying visibly safe. Leaders share their own experiments, including the ones that went nowhere. And keep a simple record of what training and guidance people received: it helps them, and it is what the AI Act now asks for.
Final Word: Permission Is Part of Adoption
We often talk about AI adoption as if it were a question of enthusiasm, and plan it that way: a launch, a demo, a round of excitement. The numbers point somewhere else. Most people are already convinced. What they are missing is a clear answer to whether they are allowed, and the practice to do it well.
Both of those can be organised. Neither of them comes with the software.
Who in your organisation is waiting for a yes that nobody has been asked to give? Let’s find that out together before your next rollout.
