Adoption: AI only counts when your team uses it

There is a moment that repeats itself in a great many AI projects. The system is ready, the tests went well, the demo impressed everyone. Three months later, the team has drifted back to the old way of working and nobody can quite say when it happened. Recent research helps explain why.

Real use is lower than it looks

In the United States, an NBER study found that 23% of workers had used generative AI at work in the previous week, but only 9% used it every day. Measured in time, AI was involved in just 1% to 5% of all working hours (Bick, Blandin and Deming, 2025). Trying something out is easy. Building it into the working day is another matter.

The employer’s role

The most detailed study on the subject comes from Denmark. Anders Humlum and Emilie Vestergaard surveyed around 25,000 workers across eleven occupations exposed to AI and linked their answers to company administrative records (Humlum and Vestergaard, 2025). Three findings stand out. The first is the weight of the employer:

When the employer encourages use, adoption rises from 47% to 83%. And the reported benefits are 10% to 40% larger.

The second is training, which the study singles out as particularly effective at narrowing the adoption gaps between groups of workers. The third is that the time saved is modest on average, about 3% of working hours. The big gains appear when a process is redesigned around the tool.

Many people also work without support. A Microsoft and LinkedIn study of 31,000 workers reports that only 39% of AI users had received training from their company (Microsoft and LinkedIn, 2024). It is vendor research, but the signal matches the Danish one.

Why teams don’t adopt

• If using AI means opening a new tool, it competes with everything else for the team’s attention, and it loses.
• When people cannot see where each answer comes from, they check everything by hand. And if they are going to check everything, why use the system at all?
• The work went up instead of down, with one more dashboard, one more alert and one more record to fill in.
• Nobody on the team took part in building the system.

What works

Explicit encouragement from leadership. The Danish study shows this is the factor with the greatest effect. Telling the team that the tool is there to be used, and why, almost doubles adoption.

Hands-on training, in the real process. Sessions work through the team’s own cases, instead of a generic afternoon on AI.

Put AI where the team already works. If AI sits in email, in the ERP or in Teams, nobody has to learn a new interface.

Always show the source. Every number comes with its origin and every proposal with its reasoning, so the team knows what it can trust.

Measure adoption the way you measure everything else

How many people use the system each week, what share of cases goes through it, how many proposals are accepted without changes. These numbers deserve the same tracking as any other phase of the project, and if they fall, that is the moment to act, without waiting for the final report. Adoption is where it is decided whether, a year from now, the AI is still at work.