Three studies, three figures
In summer 2026 Bitkom Research surveyed 603 companies with 20 or more employees by phone (calendar weeks 28 to 33). Result per the press release of 14 September: 57 percent use AI, 38 percent plan or are discussing it, 4 percent say AI is not a topic for them. A year earlier use stood at 36 percent, two years earlier at 20 percent.
In spring 2025 KfW asked in its SME panel whether the company had used AI applications in 2022 to 2024. Result per Fokus Volkswirtschaft Nr. 533 of 11 February 2026: 20 percent, extrapolated to almost 780,000 companies and about five times the value for 2016 to 2018. Larger SMEs with 50 or more employees reach 36 percent, small ones 19 percent, and construction is last at 8 percent.
The KI-Index Mittelstand by Salesforce and the Deutscher Mittelstands-Bund reports 51.2 percent using or testing AI, up from 33.1 percent in 2024. We only know this figure from a trade blog. Sample and survey period aren't given there, so we couldn't verify them.
Why the figures don't contradict each other
Three differences explain most of the spread. First, the period: KfW asks about a multi-year window up to 2024 and surveyed in spring 2025, Bitkom asks in summer 2026 about the current state.
Second, company size: Bitkom covers companies with 20 or more employees. The KfW panel depicts the Mittelstand more broadly, including small businesses, and itself shows how strongly size matters: 36 percent for companies with 50 or more employees, 19 percent for smaller ones. Including small businesses gives a lower average.
Third, the definition: KfW gives respondents a detailed AI definition that includes text, image and speech recognition, and asks about eight kinds of technology. On the Bitkom page we find no definition of 'use'. The KI-Index also counts companies that only test AI. Whoever defines 'use' broadly, say down to occasional chatbot use by individual employees, reaches higher values than with productive use in a process.
The jump from 41 to 57 percent
Notably, Bitkom itself reported only 41 percent in March 2026 (604 companies, calendar weeks 2 to 6). In September it is 57 percent with a similar sample size. The two press releases give different comparison values for the respective prior year, 17 percent in March and 36 percent in September. We don't rule out that 16 percentage points of real growth happened in six months. Without information on question wording and sample composition, though, the press releases can't prove it. The series works as a direction, not as a yardstick.
What follows for a single company
The question 'how many use AI?' helps little with your own decision. More useful is the second figure from the Bitkom study, the barriers. Companies without AI cite missing technical know-how (85 percent), legal uncertainty (66 percent), hard-to-calculate costs (63 percent), lack of personnel (51 percent) and missing data (40 percent). Companies with AI cite data protection requirements (66 percent), legal uncertainty (56 percent), hard-to-calculate costs (54 percent), the risk of company data reaching the wrong hands (45 percent) and lack of employee acceptance (42 percent).
Our reading: the hurdle shifts after the start. Before, the know-how of how to begin is missing; afterwards data protection, data security and acceptance move to the front. Those are very different tasks. Anyone who plans to solve only the know-how problem, say through training, soon faces the data protection problem. This is a deduction from two lists of questions, not a finding of the study about the same company over time.
- Don't compare your situation with a single percentage, but with size class and industry. KfW shows a wide range, from construction at 8 percent to larger SMEs at 36 percent.
- Define what 'use' means in your own house: individual chat assistants, an automated process or an integrated product. Data protection, cost and responsibility depend on it.
- Clarify data protection and data types early, not after the pilot. Among users it is the most cited barrier.
- Plan for acceptance. 42 percent of users cite lack of acceptance in the team.
- Don't treat 'competitors already use it' as an argument as long as it's unclear what they use.
What the studies don't answer
None of the three figures says whether AI creates measurable benefit in the company. 'Use' is not a measure of success. The Bitkom page also gives no breakdown of the 57 percent by size that we could have used here. And a sample of around 600 companies with 20 or more employees says nothing about micro-businesses.