Report · October 2026 · English summary

Bought is not adopted

More than half of German companies use AI. Hardly any get much out of it. Why that is rarely about the technology, and what the companies that make it work do differently.

Tobias Treppmann · summary of the German report of 17 October 2026 · 4 minutes · Full report in German

In September, the German digital association Bitkom asked 603 companies with 20 or more employees how they work with AI. The headline was good news: for the first time, more than half of them use it.

The more interesting figure came a few lines further down. Of the companies that use AI, not a single one says it uses the potential fully. Six in ten say they do not use it at all.

I know this gap from my own work. To see whether others experience it the same way, I read the current company surveys by Bitkom and the TÜV association and analysed 187 episodes of three German-language podcasts in which AI leads, managing directors and consultants talk about their practice. The studies show how widespread the problem is. The episodes show what it looks like day to day.

A licence is not an adoption

According to the TÜV association, 56 % of companies use generative AI in daily work; 27 % have trained their employees in it. Where the company offers no path, people find their own: in just over four in ten companies, employees use private AI accounts, or management assumes they do. Only just under a quarter have rules for it.

Process first, then the tool

Look at what AI users actually spend money on. The big items are infrastructure (51 %), preparing data (50 %) and integration into existing systems (41 %). Only one company in five names licences (21 %). The tool is the cheap part. The examples that work in the podcasts do not start with a big clean-up project. They start with one process that is clear enough to rebuild.

Someone has to own it

For mid-sized companies the answer is more sober than expected: not a new department, but one person with a mandate. Someone who sets priorities, sees the first process through and reports what it delivered. Whether that person comes from inside, joins for a while or is hired is secondary.

Bring people along

42 % of AI users say their employees lack acceptance; two thirds rate their people’s AI skills as low; 70 % now offer training. In the episodes, something else stands out: where adoption works, people experienced AI on their own task, with someone at their side, often in small pilot groups.

Measured too late, or not at all

45 % of AI users see a measurable contribution to business success. So more than half do not see it, or do not measure it. Where it is measured, it gets concrete straight away: time per case, share of people using AI daily, costs saved.

What follows

1. Choose one processNot a strategy: a workflow that happens often, costs time and is clear enough to rebuild.
2. Make one person accountableWith a mandate and time. Internal, interim or hired.
3. Enable people on their own workPeople try AI on their tasks, with guidance, not only in a classroom.
4. Decide up front what countsTime per case, usage, cost. Measured before the start and after.
5. After 90 days: review the result, then take on the next process.

Adopting AI is work with people, not with licences.