The range is not a contradiction but a matter of definition - and the bottlenecks are the same everywhere.
What it is about
Depending on the definition, between 20% and 54.5% of companies use AI, large companies far more often than small ones. What holds them back is the same in every survey: a lack of expertise and unclear legal consequences.
The figures come from 2025 and 2026 and from serious surveys - they simply measure different things.
Why the numbers are so far apart
- Eurostat (2025): 19.95% of EU enterprises with 10 or more employees, 55.03% of large enterprises. Official statistics (ICT survey 2025, enterprises with 10 or more employees) with a narrow definition, excluding the financial sector.
- Destatis (2025): 26% in Germany, 57% of companies with more than 250 employees - the same official ICT survey 2025, enterprises with 10 or more employees in Germany.
- ifo (2026): 54.5% use AI in their business processes. A different method (ifo business survey, May 2026, sample size not stated), a broader definition, one year later.
Size is decisive: in large companies AI is everyday practice, across all companies not yet.
What holds companies back
- Companies that considered AI but do not use it mainly cite a lack of expertise (Eurostat 70.89%, Destatis 71%) and unclear legal consequences (52.52% and 58%). The base is only companies that considered AI (official statistics, Eurostat 2025, Destatis 2024).
- Only 40% of employees know of a policy on generative AI at their workplace, only 47% have received training (University of Melbourne with KPMG, survey November 2024 to January 2025, over 48,000 respondents in 47 countries, self-reported). KPMG, itself an AI consultancy, co-authored the study.
- According to RAND (2024), more than 80% of AI projects are estimated to fail. The figure comes from outside estimates, not RAND’s own measurement; the 65 interviews with experienced data scientists and engineers mainly cite a lack of data quality.
What we make of it
The bottleneck is rarely the technology, but knowledge, rules and data. That is why we start with one page of rules and capable teams - see AI governance without bureaucracy, AI adoption: we make your teams capable, not dependent and Start. Prove. Scale..
The studies in detail
Selection: only studies with a disclosed method - experiments, official statistics, legal texts, independent institutes. Vendor surveys about their own product and consultancy forecasts are not included here.
- 19,95%
Eurostat · 2025 · Official statisticindependent
In 2025, 19.95% of EU enterprises (10+ employees) used AI technologies, up 6.47 percentage points from 2024; among large enterprises it was 55.03% (Eurostat).
Base: Unternehmen ab 10 Beschäftigten in der EU (ICT usage survey), Daten extrahiert Dez. 2025
Limits: Excludes the financial sector and firms with fewer than 10 employees. Eurostat's definition of AI use is narrower than in industry surveys.
Use of artificial intelligence in enterprises ↗We use it in: AI adoption: we make your teams capable, not dependent · What the EU AI Act means for your engineering
- 26%
Statistisches Bundesamt (Destatis) · 2025 · Official statisticindependent
In 2025, 26% of enterprises in Germany used AI technologies, including 57% of those with more than 250 employees (2024: 20%).
Base: Enterprises with ≥10 employees, official ICT survey 2025
Limits: Quote is a table row; the 2024 value (20%) comes from the Destatis press release PD24_444
Unternehmen mit Nutzung von Technologien der künstlichen Intelligenz nach Beschäftigtengrößenklassen ↗We use it in: Start. Prove. Scale. · Agentic development in everyday enterprise work · Agentic AI: 85% on the slide, 36% at best
- 54,5%
ifo Institut · 2026 · Surveyindependent
54.5% of companies in Germany use AI in their business processes, up from 40.9% a year earlier (ifo survey, May 2026).
Base: ifo-Konjunkturumfrage, Mai 2026
Limits: Sample size is not given on the page. Different method from Destatis/Eurostat, so the numbers are not directly comparable.
Mehr als die Hälfte der Unternehmen nutzt Künstliche Intelligenz ↗We use it in: AI adoption: we make your teams capable, not dependent · From shadow AI to official approval: a roadmap
- 70,89%
Eurostat · 2025 · Official statisticindependent
Among EU enterprises that considered but did not use AI, 70.89% cite a lack of expertise and 52.52% unclear legal consequences.
Base: EU enterprises that considered using AI (official ICT survey 2025)
Limits: Base is only enterprises that considered AI
Use of artificial intelligence in enterprises - Statistics Explained ↗We use it in: AI adoption: we make your teams capable, not dependent · What we learned from six GenAI projects · Start. Prove. Scale.
- 58%
Statistisches Bundesamt (Destatis) · 2024 · Official statisticindependent
German enterprises that considered but did not use AI cite a lack of knowledge (71%), unclear legal consequences (58%) and data protection concerns (53%).
Base: Enterprises with ≥10 employees that considered AI, 2024
Limits: Base is only enterprises that considered AI; 2024 data
Jedes fünfte Unternehmen nutzt künstliche Intelligenz ↗We use it in: AI governance without bureaucracy · What the EU AI Act means for your engineering · What we learned from six GenAI projects · Start. Prove. Scale.
- 40%
University of Melbourne & KPMG · 2025 · Surveysurvey, self-reported
Only 40% of employees say their workplace has a policy or guidance on generative AI use (KPMG/University of Melbourne, 48,000+ respondents in 47 countries, 2025).
Base: über 48.000 Personen in 47 Ländern, Nov. 2024 - Jan. 2025
Limits: Global and self-reported. KPMG co-authored the study.
Global study reveals trust of AI remains a critical challenge reflecting tension between benefits and risks ↗We use it in: AI governance without bureaucracy · From shadow AI to official approval: a roadmap
- 47%
University of Melbourne & KPMG · 2025 · Surveysurvey, self-reported
58% of employees intentionally use AI at work, but only 47% have received AI training (KPMG/University of Melbourne, 48,000+ respondents, 47 countries, 2025).
Base: über 48.000 Personen in 47 Ländern, Nov. 2024 - Jan. 2025
Limits: Global and self-reported. 'Training' is not further defined.
Global study reveals trust of AI remains a critical challenge ↗We use it in: AI adoption: we make your teams capable, not dependent
- >80%
RAND Corporation · 2024 · Studyindependent
According to RAND, by some estimates more than 80% of AI projects fail, twice the rate of non-AI IT projects; a main root cause is a lack of high-quality data (65 expert interviews).
Base: 65 experienced data scientists and engineers interviewed
Limits: The 80% figure cites outside estimates; the interviews are qualitative
The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed ↗We use it in: What we learned from six GenAI projects · Start. Prove. Scale.
