AI adoption by EU businesses is growing, but the figures do not measure integration

An increase in the share of businesses reporting that they use artificial intelligence is a sign that the technology is spreading, but it does not, by itself, show how much work within those organisations has changed. Reported use and integration are different measures: a business may count as a user even if it applies a tool to a narrowly defined task, while another may have reorganised processes and responsibilities around it.

The Funcas document identified for this analysis is titled La adopción de las tecnologías IA por las empresas de la Unión Europea: evidencias y explicaciones desde la teoría de la difusión tecnológica. However, the evidence available here does not include the full text and does not allow independent verification of the specific figures of 7.6% in 2021 and 20% in 2025, the statistical population they cover or how they were calculated. It would not be responsible to present those figures as verified on the basis of the title and the retrieved excerpt alone. (https://www.funcas.es/documentos_trabajo/la-adopcion-de-las-tecnologias-ia-por-las-empresas-de-la-union-europea-evidencias-y-explicaciones-desde-la-teoria-de-la-difusion-tecnologica/)

The debate can be placed in context using an official Eurostat series available in the research: in 2024, 13.5% of EU businesses with ten or more persons employed reported using AI technologies in their business activity, compared with 8.0% in 2023. This indicator is defined by year and business size; it should not automatically be combined with another percentage without confirming that the populations, definitions and methods match. (https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20250123-3)

The Eurostat figure refers to businesses with ten or more persons employed and to the use of AI technologies in their business activity. This excludes businesses with fewer than ten employees from the population described, so the result does not, without qualification, represent all European businesses. The unit being observed matters as much as the percentage: a rate calculated for businesses of a certain size cannot be applied to microbusinesses without another source. (https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20250123-3)

The statistical release also breaks out specific technologies: in 2024, the most common categories were text analysis (text mining), the generation of written or spoken language, and speech recognition. The release assigns them different shares, showing that “AI use” groups together heterogeneous practices rather than one single type of implementation. The aggregate alone does not reveal the purpose or operational weight of each application. (https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20250123-3)

It is worth reading the label “use” as a survey category, not as certification of digital maturity. The available figure does not say whether a tool is experimental or stably deployed, how many people use it, what proportion of tasks it covers, or whether it connects to core systems. Those questions require additional variables and are not answered by the overall rate.

If the figures attributed to the Funcas document were comparable, the move from 7.6% to 20% would amount to an increase of 12.4 percentage points. In relative terms, the final share would be approximately 2.6 times the initial one. These are two different ways of expressing change: percentage points compare the rates directly, while the multiplier shows how the final figure relates to the initial one. But the arithmetic does not verify the source or methodological comparability.

Before using that change as a statistical trend, it would be necessary to confirm that both observations cover the same countries, sectors and business sizes, and apply an equivalent definition of AI technology. It also matters whether the percentage represents surveyed businesses, weighted businesses or a regional aggregation, as well as the reference period and response rules. Without these checks, the comparison may mislead, even if the subtraction is correct.

The Eurostat reference provides a verifiable comparison point for 2023 and 2024: 8.0% and 13.5%, respectively, among EU businesses with at least ten employees. These figures are consistent with a rapid expansion in reported use over that interval; they are not enough to validate a different series spanning 2021 and 2025. A matching direction does not establish continuity between series. (https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20250123-3)

An adoption rate essentially answers how many businesses report some use under the survey’s definition. It does not answer whether AI has become part of a routine process, whether human oversight exists, whether the organisation has updated its data, or whether the system produces better results than the previous alternative. Neither the intensity nor the quality of use can be inferred from a binary count of businesses using AI.

Nor is it valid to turn an increase in adoption into evidence of productivity gains. Demonstrating impact would require outcome data and an analysis that can distinguish the effect of AI from other changes, such as investment, training, sectoral composition or the economic cycle. A descriptive comparison can show that changes occurred at the same time; on its own, it does not establish causality. The evidence reviewed contains no estimate of business impact that would allow improvements to be attributed to AI use.

This caution does not make the figure unimportant: the spread of new technologies can be economically significant. It does, however, require separating questions that are often conflated: how many businesses use AI, how often they use it, which processes it transforms and what results it produces. Each question requires its own indicators and evidence, and answering one does not settle the others.

Assessing the spread of use requires comparable year-by-year series and a transparent definition of the population covered. The Eurostat release provides an example of a breakdown by technology type and country for 2024, as well as a comparison with 2023. It is useful for describing differences and changes within that specific framework, but it does not, on its own, measure organisational depth. (https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20250123-3)

To assess integration, one would need indicators such as the proportion of processes in production, the number or type of functions using AI, continuity of use, and whether organisational changes or training have taken place. Measuring outcomes would require metrics suited to the goal—for example, time, quality or costs—and a design that allows valid comparisons. The evidence provided does not include these indicators; listing them here defines what is missing, rather than claiming that AI has already improved those dimensions.

Business size and sector can also affect how an aggregate figure should be read. A single rate may conceal differences between activities with digitised data and processes and others where adoption is less common. The cited European statistic allows some geographic and technological breakdowns to be consulted, but explaining who adopts and why would require detailed tables and specific analyses. A gap by sector or business size should not be inferred without those figures.

The strongest conclusion is a limited one: Eurostat records a year-on-year increase in the share of EU businesses with ten or more employees reporting AI use, from 8.0% in 2023 to 13.5% in 2024. The Funcas document presents a longer-running series, but the evidence available for this article is not sufficient to certify its endpoints of 7.6% and 20%, or the details of their comparability. Growth in reported use differs from implementation maturity. (https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20250123-3)

The figure can therefore be presented as a sign of diffusion, not as a measure of deep integration or proof of business benefits. Supporting those conclusions would first require checking the series’ full methodology and then adding indicators of intensity, processes and outcomes. Until then, the useful approach for businesses and readers is to keep adoption, implementation and impact distinct.