AI agents in Romania: 5 in 100 companies use them. What about the rest?

Romania closed 2025 last in the European Union for the use of artificial intelligence in business: 5.21% of companies with at least 10 employees use at least one AI technology in their activity. The EU average is 19.95%, and Denmark, at the top of the ranking, sits at 42.03%.

The figure is accurate, and on its own it says almost nothing useful. It does not say which companies are missing from the statistic, it does not say why they are not adopting, and it does not say what a five-person office could do tomorrow without an IT team and without a written strategy. So I took the official data by company size and put it in context, because that is where the market actually is and where the first step belongs.

What the figure measures — and who it leaves out

The Eurostat survey on the use of AI in enterprises starts at companies with 10 employees or more. Below that threshold, companies are not measured at all. In Romania, more than 90% of companies are micro-enterprises, and they employ roughly a quarter of the workforce. Small and medium-sized enterprises account for 99% of active companies and two thirds of the workforce.

In other words: the public discussion about “only 5% of Romanian companies” refers to a numerical minority of the country’s businesses. The segment that stands to gain the most from automation — the small firm where the owner answers the phone, WhatsApp and email at the same time — does not even appear in the statistic.

There is a second source of confusion. Studies announcing that “over 70% of Romanian companies use AI” measure something else: self-declared use of tools by employees. Three people in a company having an account on a conversational assistant does not mean the company runs a process on AI. Eurostat asks exactly that — AI technologies used in the enterprise’s activity — and that is where the gap between 70% and 5% comes from.

By segment: small, medium and large companies

Data by size class says more than any national average. Below are the shares of companies using at least one AI technology, in Romania and in the EU, for 2024 and 2025:

Size class	Romania 2024	Romania 2025	EU 2024	EU 2025
10–49 employees	2.57%	4.13%	11.21%	17.00%
50–249 employees	3.88%	7.79%	20.97%	30.36%
250+ employees	11.26%	20.75%	41.17%	55.03%
Total (10+ employees)	3.07%	5.21%	13.48%	19.95%

Three observations matter for anyone selling or buying automation:

1. The growth is real, but it starts from almost nothing. Romania moved from 3.07% to 5.21% in one year — an increase of over 70%. At that pace, reaching today’s European average would take about a decade, during which the European average will have moved as well.
2. The widest gap is not among small companies, but among medium-sized ones. A Romanian company with 50–249 employees adopts AI at nearly four times the EU gap: 7.79% against 30.36%. This is the segment with enough volume to feel the pain of repetitive work, but without an IT department to solve it.
3. Large companies are not behind for lack of money. At 250+ employees, one in five uses AI, against one in two in the EU. The problem there is different: integration with existing systems, governance, compliance, long procurement cycles. It is not solved with a subscription.

Why small companies don’t adopt: not money, but knowledge

Eurostat also asks companies why they do not use AI. Romania does not report this module, but the European answers are clear. Among companies that do not use AI:

• 70.3% say they lack the relevant expertise;
• 53.6% cite a lack of clarity about the legal consequences;
• 52.7% have concerns about data protection and privacy;
• 43.5% report difficulties with the availability or quality of the necessary data;
• 41.6% mention incompatibility with existing equipment, software or systems;
• 38.4% consider the costs too high;
• 24.6% raise ethical considerations;
• 17.8% consider AI not useful for their company.

The first reason is not money, it is knowledge — and it is nearly twice as important as cost. Public perception confirms it: in the 2026 Digital Decade barometer, 28% of Romanians name the lack of training or relevant skills as their main concern about generative AI tools.

In practice, inside a small company, the reluctance sounds like this: “it’s not for my business”, “it’s for big companies”, “it’s expensive”, “it will answer a customer wrong”, “I have nobody to work on this” or “I have no time to learn”. The first four are solved with education and a small pilot; the last two are exactly the problem a digital employee solves.

What a small office can actually do today

An AI agent is not a chatbot you talk to. It is a well-defined process that runs on its own, with controlled access to the company’s tools and a human who approves what matters. For a small office, the starting list is short and verifiable:

• WhatsApp and the phone. The first reply arrives immediately, at any hour: it takes the request, asks for the details it needs, books the appointment, and hands the conversation to a human when there is a decision to make. The owner is no longer the only person who can answer.
• Email. Automatic triage, documents extracted and archived, an alert only when something needs attention. The rest becomes a summary, not an inbox to read in the evening.
• Appointments and documents. Calendar synchronisation, reminders, documents and invoices prepared for approval.
• Online presence. A draft article or post every week, built from the company’s real activity, with the owner’s approval before anything is published.

In our own firm, measured on current operations: sorting the inbox used to take 30–45 minutes a day, coordination and status reporting 1.5–2 hours a day, and a code change 1–1.5 hours of manual work. The same pattern repeats in any small office: two to three hours a day of repetitive work that produces nothing new.

It is worth stating what stays with humans, otherwise the conversation is not serious: approving money, contracts, communication with the client, any commercial commitment, and any legal or tax decision. The agent prepares and executes on confirmation; it does not sign.

How to start without risk

The recipe that works is not “digital transformation”, it is a single process:

1. Pick the process with the highest repetitive volume, not the whole department.
2. Describe the procedure step by step, including what the human decides and what the agent decides.
3. Run a 2–3 week pilot, with separated access, an operations journal and human approval on anything sensitive.
4. Measure before and after — hours saved, things lost, response time.
5. Only then extend to the second process.

We do not promise financial ROI and we do not sell consulting by the hour. We work in measured hours saved, and whatever stays manual is stated at the start, not at the end.

The conclusion

Romania is not last in the EU because its companies are small, but because nobody has explained, concretely, what today’s technology can do in a five-person company. That 5% is a statistic, not a verdict. And for a small company, the first step is not a strategy — it is one process taken off the owner’s desk.

---

Sources

• Eurostat, Use of artificial intelligence in enterprises (data extracted December 2025), ec.europa.eu
• Eurostat, 20% of EU enterprises use AI technologies, 11 December 2025, ec.europa.eu
• Eurostat dataset isoc_eb_ai (AI use by size class, and reasons for non-use), ec.europa.eu
• European Commission, Romania’s 2026 Digital Decade Country Report, digital-strategy.ec.europa.eu
• National Bank of Romania — analysis of micro-enterprises and productivity; Agerpres, SMEs account for 99% of active companies (10 November 2025)
• The hours-saved figures come from Constantine Provisioning Group’s own operations.