Why companies that considered AI decided against it

Not because they concluded that AI would be of no use to them. According to Eurostat's isoc_eb_ai database for 2025, among EU enterprises with 10 or more persons employed that had ever considered using an artificial intelligence (AI) technology, 70.31% cited a lack of relevant expertise as a reason for not using it, 53.61% cited a lack of clarity about the legal consequences, and 52.72% cited concerns about breaching data protection and privacy. The answer that AI technologies are not useful for the enterprise was chosen by 17.79% — nearly four times less often than missing expertise.

Among those enterprises, the majority is not rejecting AI on the merits: the most frequently cited barriers are that there is nobody to implement it and that it is unclear what is permitted. The indicator is measured among enterprises that do not use AI. About everyone else — both those already using AI and those that never got as far as considering it — this survey says nothing on the question of why.

What these percentages actually measure

The figures above do not refer to all European enterprises. Eurostat publishes these indicators against several denominators; the unit used here (code PC_ENT_AI_EC) is the percentage of the enterprises which ever considered to use any of the AI technologies. That group is small: in 2025 it was 11.1% of the enterprises in scope in the EU and 4.89% in Bulgaria. Measured against all enterprises in scope, the same barrier — a lack of relevant expertise — gives 7.76% for the EU and 3.56% for Bulgaria. Both values are official; only the denominator differs.

The coverage is limited too: enterprises with 10 or more persons employed in NACE Rev. 2 aggregate C10–S951_X_K — all activities except agriculture, forestry and fishing and mining and quarrying, and without the financial sector. Micro-enterprises with fewer than 10 persons employed are outside the statistic.

One warning for the reader who checks the source: Eurostat's Statistics Explained overview article (data extracted in December 2025) shows different values for these same four reasons — 70.89%, 52.52%, 48.83% and 20.68%. Those are the reference-year 2024 values in the same database. The figures used here are taken directly from the isoc_eb_ai database for reference year 2025, in its version of 15 June 2026.

What changed between 2024 and 2025 in the EU

The comparison between the two years on the same indicator for the EU-27 reads as follows:

• lack of relevant expertise: 70.89% in 2024 and 70.31% in 2025;
• lack of clarity about the legal consequences: 52.52% and 53.61%;
• concerns regarding violation of data protection and privacy: 48.83% and 52.72%;
• AI technologies not useful for the enterprise: 20.68% and 17.79%.

Two observations do not make a trend, and the series has no data for 2022. With that caveat: the argument that the technology is not useful recedes by 2.89 percentage points, data-protection concerns rise by 3.89 points, and the expertise barrier is essentially unchanged (–0.58 points). The direction is not the same everywhere: in Bulgaria legal uncertainty moves the other way — downwards.

The Bulgarian picture

For Bulgaria the same four indicators read: lack of expertise 61.60% in 2024 and 72.69% in 2025; legal uncertainty 53.51% and 50.29%; data protection 38.39% and 47.68%; the answer that AI technologies are not useful for the enterprise 13.25% and 16.26%. In a single year the expertise barrier rose by more than 11 percentage points.

In 2025, 8.55% of Bulgarian enterprises with 10 or more persons employed in the same scope used at least one AI technology, against an EU-27 average of 19.95% — the third-lowest share in the Union, above only Romania (5.21%) and Poland (8.36%). The gap to Greece (8.93%) is only 0.38 percentage points, so the ranking itself is fragile. By size: 7.17% for 10–49 persons employed, 13.3% for 50–249 and 26.18% for 250 or more, against 17.0%, 30.36% and 55.03% in the EU. Even Bulgaria's largest companies stay below half the EU level for their size band.

The legal uncertainty now has firm dates

Regulation (EU) 2024/1689 (the AI Act) was published in the Official Journal on 12 July 2024 and entered into force on 1 August 2024. Under its Article 113, "It shall apply from 2 August 2026", but "Chapters I and II shall apply from 2 February 2025". That means the AI literacy duty in Article 4 and the prohibitions in Article 5 have bound enterprises for over a year.

In its "AI literacy – questions and answers" guidance (last updated 19 November 2025), the European Commission answers the question whether a company whose employees use ChatGPT for, for example, writing advertisement text or translating text has to comply with Article 4: "Yes, they should be informed about the specific risks, for example hallucination." Supervision and enforcement sit with the national market surveillance authorities; in one place the Commission's page gives 3 August 2026, while the application date in Article 113 of the Regulation is 2 August 2026.

It is worth knowing what the Regulation does not provide. Article 4 is not among the provisions listed in Article 99(4) — those whose infringement is subject to administrative fines "of up to EUR 15 000 000 or, if the offender is an undertaking, up to 3 % of its total worldwide annual turnover for the preceding financial year". And under Article 99(6), for SMEs, including start-ups, "each fine referred to in this Article shall be up to the percentages or amount referred to in paragraphs 3, 4 and 5, whichever thereof is lower" — that is, the lower of the two figures, not the higher.

The amending "Digital Omnibus on AI" regulation (procedure 2025/0359(COD)), which defers the obligations for high-risk systems, was adopted by the European Parliament on 16 June and by the Council on 29 June 2026 and signed by the Presidents of both institutions on 8 July 2026, but as of 22 July 2026 it has not been published in the Official Journal and is therefore not in force.

What follows from the data

If the leading barrier were a lack of usefulness, the task would be persuasion. The data say otherwise: three of the four most frequently cited reasons are questions of capacity — skills, legal orientation and data protection. Those call for training, internal rules and accessible legal guidance rather than more demonstrations.

Eurostat also reports that 22.50% of individuals aged 16–74 in Bulgaria used generative AI tools in the three months before the 2025 survey, against 32.66% in the EU; among 16–24-year-olds in Bulgaria the share is 49.98%, and among those aged 55–74 in Bulgaria it is 8.09%. This is a different survey, with a different population, and is not directly comparable with the enterprise statistics. Setting the two sets of numbers side by side does raise a question each company can answer for itself: 22.50% personal use against 8.55% enterprise use — does the firm have internal rules and training for tools that some of its staff already use?

On sources. The data on barriers and on enterprise AI use come from Eurostat's isoc_eb_ai database (version of 15 June 2026), and the data on individuals from isoc_ai_iaiu; those addresses are queries to Eurostat's dissemination interface and return exactly the values quoted. The legal provisions are quoted from Regulation (EU) 2024/1689 on EUR-Lex. Caveats: the isoc_eb_ai series has no observation for 2022; and from 2025 the aggregate for use of at least one AI technology (E_AI_TANY) covers eight technologies instead of seven, so part of the rise on 2024 reflects the widened definition. That second caveat concerns the adoption figures — 8.55%, 19.95% and the size breakdowns — not the barrier indicators.