88% or 20%? The Number Everyone Cites

One number carries half the AI debate: 88 percent of organizations use AI in at least one business function. Official European statistics put the figure for the same period at 20 percent. Anyone using AI adoption statistics should know which question they answer. Otherwise they plan on the wrong basis, in either direction.

Contents

• Two numbers, one apparent contradiction
• What each survey actually asks
• Why the gap is not a measurement error
• What this means for planning
• Which number we work with
• Conclusion: the question before the number
• Frequently asked questions about AI adoption statistics
• Sources

Two numbers, one apparent contradiction

The 88 percent comes from the McKinsey Global Survey on AI and is cited, among others, in Stanford University’s AI Index Report (Source: McKinsey, The State of AI, 2025; Stanford HAI, AI Index Report, 2026). It appears in almost every presentation on the topic. Including ours.

The 20 percent comes from Eurostat. For the 2025 survey year, 20.0 percent of EU enterprises with ten or more employees used at least one AI technology, up from 13.5 percent the year before (Source: Eurostat, 2025). Germany’s Federal Statistical Office reports 26 percent (Source: Destatis, 2025 survey).

The distance is substantial. It comes from methodology, not sloppiness.

What each survey actually asks

The two numbers do not measure the same thing. They differ on four points, and each one shifts the result.

For presentations:SVGPNG

Who gets counted? McKinsey surveyed 1,993 individuals across 105 countries between 25 June and 29 July 2025. The unit of observation is the person reporting on their organization, not the company (Source: McKinsey, The State of AI, 2025). Eurostat and Destatis record enterprises as the unit, from ten employees upward, within the annual ICT survey.

Who takes part? Participation in the McKinsey survey is voluntary; 38 percent of respondents work for organizations with over one billion US dollars in revenue. The official survey draws a sample and, in Germany, carries a legal obligation to respond.

How is it weighted? McKinsey weights by each respondent country’s contribution to global GDP. Official statistics extrapolate by size class and sector onto the actual business population.

What counts as use? For Eurostat and Destatis, using at least one of eight defined technologies counts, from text mining and image recognition to autonomous robots. McKinsey asks whether the organization regularly uses AI in at least one business function.

Stanford itself notes that the survey data it cites is self-reported and should be read as directional rather than comprehensive (Source: Stanford HAI, AI Index Report, 2026).

A free Self-Check shows where a company stands in a few minutes.

Why the gap is not a measurement error

Both numbers are correct. They answer different questions.

The 88 percent answers: how widespread is AI in the world of large, digitally active organizations, as seen by the people working there?

The 20 percent answers: how many enterprises with ten or more employees in the EU demonstrably use at least one AI technology?

“Someone in the company uses AI” is not the same as “AI is in the operation”.

Size dependency makes this visible. In Eurostat’s data, 55.0 percent of large EU enterprises use AI, but far fewer small ones (Source: Eurostat, 2025). In Germany, large companies sit at 57 percent, mid-sized at 36 percent, small at 23 percent (Source: Destatis, 2025). Surveying mostly large corporations necessarily produces a higher number.

Between the two poles sits the ifo Institute at 54.5 percent for Germany (Source: ifo Institute, 2026): a company survey, but not an official one with a duty to respond.

One final note on summaries: the AI Index 2026 states 70 percent GenAI adoption in its highlights, while the corresponding chapter, using the same data, reports 79 percent (Source: Stanford HAI, AI Index Report, 2026). Even within one report, the chapter is worth more than the summary.

What this means for planning

For an individual company the headline number is nearly worthless. It becomes useful only when matched to its own situation.

• Compare by size, not to the average. A company with 400 employees compares itself to the mid-sized class, not to a figure that includes corporations.
• Compare by sector. In the EU, 62.5 percent of companies in information and communication use AI, against 10.8 percent in construction (Source: Eurostat, 2025). There is no single benchmark.
• Time series beat snapshots. Eurostat and Destatis survey annually with the same methodology. That shows real movement: from 13.5 to 20.0 percent in one year at EU level.
• Adoption is not maturity. None of these figures says whether AI creates value. Only your own measurement answers that, as described in the article on measuring AI ROI.

The last point matters most. High adoption figures create pressure to act, but no orientation. That confusion is exactly what the Implementation Gap describes: reach grows faster than effect.

Which number we work with

We cite the 88 percent ourselves in several places. As a description of a trend it is usable: AI has arrived in large organizations.

For statements about the German mid-market we do not use it. There we work with Eurostat, Destatis and ifo, because those surveys treat companies as the unit, break results down by size class, and repeat annually with the same methodology.

And when the subject is effect rather than reach, none of these numbers counts. Then what counts is what a single workflow cost before and after. Why AI is rarely the actual problem is covered in AI Is Not the Problem.

Which benchmark fits a specific case is something we map out in the free diagnosis call.

Conclusion: the question before the number

88 percent and 20 percent do not contradict each other. They answer different questions, with different units of observation, for different populations.

Anyone adopting a figure should know three things: who was counted, who answered and what counted as use. All three appear in any serious source, usually in the methodology section nobody opens.

For your own planning that leads to an uncomfortable simplification: external figures are good for orientation, not for decisions. Decisions run on your own numbers.

Frequently asked questions about AI adoption statistics

Why do AI adoption figures differ so widely?

Because they measure different things. Surveys like McKinsey’s ask individuals about their organization, weighted by economic output. Eurostat and Destatis record enterprises with ten or more employees as the unit, with a defined population.

Which number is the right one?

Both, for their respective question. Survey data works for statements about large, digitally active organizations. Official statistics work for statements about a country’s business population.

How many companies in Germany use AI?

26 percent in the 2025 survey year: large companies 57 percent, mid-sized 36 percent, small 23 percent (Source: Destatis, 2025). The ifo Institute’s company survey reports 54.5 percent (Source: ifo Institute, 2026).

What counts as AI use in official statistics?

Use of at least one of eight defined technology categories, including text mining, speech recognition, image generation, image recognition, machine learning and AI-based process automation.

How does a company find the right benchmark?

Through size class and sector rather than the overall average. Which benchmark holds in a specific case is something we clarify in the free diagnosis call.

Sources

• Eurostat, 2025: Use of artificial intelligence in enterprises, Statistics Explained and 20% of EU enterprises use AI technologies, 11 December 2025
• Federal Statistical Office (Destatis), 2025: Use of ICT in enterprises: results of the 2025 survey
• McKinsey & Company, 2025: The state of AI in 2025: Agents, innovation, and transformation, 5 November 2025
• Stanford HAI, 2026: AI Index Report 2026, Chapter 4: Economy
• ifo Institute, 2026: More Than Half of Companies in Germany Use Artificial Intelligence, 5 June 2026