Both numbers are true and they measure different things. Nearly 89% is a US self-report about any use of an AI tool. For 2025 Eurostat says 17.0% of small firms in the EU use at least one AI technology, and 15.6% in Slovakia. Of those firms, 4.1% automate a workflow with AI.
That 89% figure reached me three times in one week. Twice in a newsletter, once from a client asking whether he was already behind. So I downloaded the raw Eurostat tables and read the methodology behind them. My name is Tair Khamitov and at DevNova I decide which process in a client's company is worth automating and which is not. The gap between those two numbers is not a lie. It is a definition, and it costs you money.
Contents
1. Where the 89% figure comes from
On LinkedIn it circulates as a fact. In newsletters as a headline. At conferences as a warning: "89% of small businesses already use AI, and you don't?"
The number is real and it has an author. It comes from research by ICIC (Initiative for a Competitive Inner City), run in partnership with Intuit, with more than 3,700 responses. It reads exactly like this: nearly 89% of surveyed small businesses reported using AI tools, for drafting emails, creating marketing content and analysing data.
Three things get lost on the way into a Slovak newsletter. It is a US survey, not an EU one. It is a self-report, the company said about itself that it "uses AI". And "uses" means anything at all, including one email drafted in ChatGPT.
None of that is a fraud. ICIC states its methodology. The loss happens in transit: by the third forward, all that survives is a bare "89% of small businesses", with no country, no sample, no definition.
If you run a cafe in Petrzalka and you read that 89% of your competitors do AI, you will do one of two things. You will panic. Or you will open a chatbot, write one Facebook post and tick the box. Both reactions are wrong, and both come from a number that was never about you.
The 89% figure is not a lie. It is the answer to a different question than the one you are asking.
2. What Eurostat actually measures
Eurostat asks differently. In its enterprise ICT survey it asks companies whether they use at least one of eight specific AI technologies: text mining, speech recognition, generating text or code, generating images and audio, image recognition, machine learning for data analysis, automating workflows and decision making, and autonomous movement of machines.
For 2025 the result was that 20.0% of EU enterprises with ten or more employees use at least one of them. Among small firms (10 to 49 employees) it is 17.0%. Among medium firms 30.4%, among large ones 55.0%.
The growth is real and fast. In 2023 it was 8.1%, in 2024 already 13.5%. That is +6.5 percentage points in a single year. Nobody here is claiming nothing is happening.
| What is actually measured | Who was asked | Result | Source |
|---|---|---|---|
| Does your business use AI tools? | small businesses, US, self-report (n = 3,700+) | nearly 89% | ICIC & Intuit |
| Have you used generative AI in 3 months? | individuals, EU | 32.7% | Eurostat 2025 |
| Does the firm use 1 of 8 AI technologies? | EU firms, 10 to 49 employees | 17.0% | Eurostat 2025 |
| The same, Slovakia | SK firms, 10 to 49 employees | 15.6% | Eurostat 2025 |
| Does AI automate a workflow or decision? | EU firms, 10 to 49 employees | 4.1% | Eurostat 2025 |
| Do SMEs use generative AI in core activity? | SMEs using genAI, OECD | 29% | OECD 2025 |
Sources: Eurostat, ICT surveys of enterprises and households 2025 (online code isoc_eb_ai), published 2026; OECD, AI adoption by small and medium-sized enterprises, December 2025; ICIC and Intuit, AI in Business. Eurostat covers firms with 10+ employees in selected sectors (NACE C to J, L to N, 95.1). The rows are not interchangeable. Each one measures a different population and a different question.
And here is the limitation you need before you hold that number against your own company. Eurostat only collects data from firms with ten or more employees. A cafe with three people, a two person studio, a sole trader: they are simply not in that sample. So even 17.0% is not exactly "your" number. It is the closest hard number that exists about small firms in the EU.
3. Why the numbers differ: three different questions
The gap between 89% and 17% does not exist because somebody lied. It exists because these are three different questions.
- The question for a person: "Have you used generative AI in the last three months?" In 2025, 32.7% of EU residents said yes.
- The question for a firm: "Does your company use at least one of eight AI technologies?" Small EU firms: 17.0%.
- The question for a process: "Does AI automate a workflow or assist decision making at your firm?" Small EU firms: 4.1%.
The same owner can answer yes to the first question and no to the third. He will not have lied once.
Eurostat knows about this problem and writes about it openly. In its chapter on methodological challenges it admits that defining AI is hard: the technology is often invisibly embedded in other software, and the word itself is frequently used as a buzzword that does not reflect technological reality. That is exactly why the household survey was deliberately narrowed to the conscious, intentional use of generative AI.
The OECD confirms it from the other side. In its discussion paper for the G7 it reports that in the US about 40% of people aged 18 to 64 had used generative AI at work or at home by late 2024. People use AI. Companies have it inside a process far less often.
Ask the owner and you get a high number. Ask the process and you get a low one.
4. Where Slovakia stands
In that table Slovakia looks different from what the usual "we are behind" narrative would suggest.
For 2025, 18.0% of Slovak firms with ten or more employees use at least one AI technology. Among small firms it is 15.6%. The EU average is 20.0% and 17.0% respectively. So we are below average, by two points and by 1.4 points.
And now the part that surprised me. Czechia is at 17.6% (small firms 13.4%). Hungary 10.4% (8.7%). Poland 8.4% (5.8%). Slovakia is ahead of all three of its V4 neighbours.
At the top of the ranking sits Denmark with 42.0%, at the bottom Romania with 5.2%. Austria, where a lot of Slovak firms go for inspiration and for clients, sits at 30.0%.
Where Slovakia genuinely lags is not at the bottom. It is at the top. Slovak small firms (15.6%) sit 1.4 points from the EU average (17.0%). Slovak large firms (43.7%) sit 11.3 points below the EU average (55.0%). The gap is eight times bigger on the large firm side. I expected the opposite.
For a small business owner one practical thing follows. Your competitors in your size class are not doing much better. The space is still open. The wider digitalization context is in my guide to digitalizing a small business.
5. The 4.1% nobody quotes
Of all the numbers in this article, the most important one is the one that never shows up in a newsletter.
Eurostat breaks the AI technologies down one by one. Among small EU firms in 2025: text mining 9.9%. Generating images and audio 8.1%. Generating text or code 7.0%. Machine learning for data analysis 3.8%.
And the technologies that automate a workflow or assist decision making: 4.1%. The year before, 3.0%.
Read that again. The most common "AI" in a small company is working with text. The rarest is AI that does something on its own.
The OECD arrived at the same place by another route. Among small and medium enterprises that use generative AI, only 29% use it in their core activity. The rest use it for peripheral tasks. And AI adoption inside firms' core business functions ranged from 1.9% (Japan) to 6.1% (the United States) across the G7.
So the gap between 89% and 4.1% is not a gap between optimism and pessimism. It is the gap between trying something and adopting it.
I see the same pattern in my own measurements, from another angle. When we went through 93 live websites built by AI builders, the mean SEO score was 96 out of 100, while the median mobile load of the main content was 6.6 seconds, with 70% of them in the "poor" band. AI touched that website; a business outcome did not come out of it. One of those sites we later rebuilt in custom code: from 11.4 seconds down to 1.5. The whole measurement is in are AI websites good for SEO (dataset on AI-Builder Audit 2026), the before and after numbers on the project page.
Trying is not adopting. Surveys measure the first, invoices measure the second.
6. Why the gap matters to an owner
A false number is not just an inaccuracy. It leads to two specific bad decisions, and both cost money.
Panic. The owner believes he is behind 89% of his competitors and buys an AI tool without knowing which process it replaces. The tool runs, nobody uses it, the subscription leaves every month. We see this pattern on audits regularly.
The alibi. The opposite direction. Once a month the owner opens a chatbot, rewrites one email and tells himself the company "already does AI". Box ticked. Inside the processes, NOTHING changed.
Both reactions share the same flaw. They decide by somebody else's percentage instead of their own number.
And now the part that AI articles rarely say. Automation does not always make sense. If you process five orders a day and answer one email an hour, a well built workflow will save you less than it costs to build and maintain. I will say it BLUNTLY: below a certain volume of repetitions, AI is more expensive than a person. The threshold is not a percentage from a survey. It is the number of repetitions per month.
7. Measure your own AI adoption in 10 minutes
Instead of comparing yourself to a survey, measure your own number. It takes about ten minutes and you need paper.
- Step 1. Write down the processes that repeat at least 20 times a month. Not the ones where "AI could be used". The ones that actually repeat.
- Step 2. Next to each, write how many minutes one repetition takes. Multiply by the number of repetitions. Now you have hours per month.
- Step 3. Mark the ones where the input is text or data and the output is text or a decision. Those are the candidates. Leave the rest alone.
- Step 4. For each marked one, ask: if it stopped working tomorrow, would anyone notice within a week? If not, it is not adoption. It is an experiment.
Your real AI adoption is the count of processes that survive question four. If you end up with zero or one, you are not behind 89% of the market. You are exactly where most small firms in the EU are, and Eurostat says the same thing from the other side with its 4.1%.
That number has one property a survey percentage does not. You can raise it by a decision, not by a feeling.
8. One measurable process instead of a percentage
This is how we do it at DevNova. One bot, one use case, a measurable result. No "AI will take over everything".
- 1. Process audit. We go through the list from the previous section and pick one process. The one with the highest repetition count, not the most interesting one.
- 2. The metric before launch. We agree what gets measured: hours saved per month, or the share of enquiries resolved without a human. We write the number down before we start building.
- 3. We build one workflow. Chatbots from €120 setup plus €15 to €80 per month. Custom AI integrations and workflows from €480 setup. Delivery in 10 to 14 days, a simple chatbot from 7 days.
- 4. A monthly report. If the metric does not hold after 30 days, we switch it off and go back to step 1. An automation nobody measures is just one more tool in the panel.
Technically: Claude Haiku 4.5 for transactional use cases, our own backend, data in the EU region (Frankfurt), escalation to a human by email or Slack. Prices are public on the pricing page, finished projects with numbers on the work page.
A concrete example of one measurable process, a chatbot on the first line of e-shop enquiries, is written up in an AI chatbot for a Slovak e-shop. And if you want to know what part of AI regulation actually applies to you from August 2026, that is the EU AI Act for a small company.
Write the metric down before you start building. Otherwise you will fit it to the result afterwards.
If you want your processes reviewed from the outside, an AI audit is a concrete list of what in your company is actually worth automating and what is not. If you already know what you want, write it into the brief and we will reply within 24 hours.
Frequently asked questions (FAQ)
How many small businesses really use AI, 89% or fewer? It depends who you ask and what you ask them. The 89% comes from a US self-report survey (ICIC and Intuit) and counts one-off use as well. European statistics ask about specific technologies deployed inside the firm and get 17.0% (EU) and 15.6% (Slovakia). Neither number is wrong. They are different populations and different definitions of the word "use".
Why do AI adoption surveys contradict each other so much? Because the respondent is not the same. A survey of people measures who opened a chatbot. A survey of firms measures what is deployed. Process statistics measure what runs without a human. The gap between those three numbers is not measurement error. It is the finding itself.
What exactly does Eurostat measure when it talks about "AI use" in enterprises? It asks firms with ten or more employees whether they use at least one of eight listed AI technologies: text mining, speech recognition, generating text or code, generating images and audio, image recognition, machine learning, workflow automation, and autonomous movement of machines. The data table is called isoc_eb_ai.
How do I check for myself whether my company really uses AI or only experiments? Look at the invoice and at the calendar. If you pay for AI and nobody in the firm has a step in a process they cannot finish without it, it is an experiment. You recognise adoption because somebody reports the outage. Not because somebody opened the tool once.
Does low AI adoption in Slovakia mean opportunity or falling behind? For a small firm, mostly opportunity. Slovak small firms (15.6%) sit 1.4 points from the EU average (17.0%) and ahead of Czechia (13.4%), Hungary (8.7%) and Poland (5.8%). The lead you gain from one working process is cheaper here than in Denmark, where 42.0% of firms already use AI.
Is opening a chatbot once in a while enough to count as "AI in the company"? In a US self-report survey, yes. In Eurostat, no, and in your company's accounts, also no. The OECD confirms it: among small and medium enterprises that use generative AI, only 29% use it in their core activity.
What does one measurable AI automated process look like in practice? It has an input, an output and a number. For example: a bot answers an order status enquiry, resolves it without a human, and at the end of the month you know what share of enquiries went through without escalation. If you do not have that number, you do not have a process. You have a tool.
Does Eurostat cover sole traders and micro companies? No. The sample starts at ten employees and covers selected sectors (NACE C to J, L to N, 95.1). If you have three people, official statistics say nothing about you. The practical consequence: you have nothing to benchmark against, so the only honest number is your own. How many processes run without somebody starting them by hand.
How much does one AI automation cost? A chatbot starts at €120 setup and €15 to €80 per month, custom integrations and workflows from €480 setup. The more important question is not the price but how many repetitions per month the process has: at low volume even €120 does not pay back. Current prices are on the pricing page.
About the author
Instead of forwarding another number, Tair Khamitov opened the raw Eurostat table and the methodological annex behind it. In Bratislava he runs DevNova: websites, e-shops and AI automations for small and medium companies. He publishes prices openly on the pricing page and projects with numbers on /work. Company ID 54730775. Contact: b2b@devnova.eu · WhatsApp +421 951 584 412.
External sources
- Eurostat, Use of artificial intelligence in enterprises, methodology and headline results.
- Eurostat, table isoc_eb_ai, raw data by country and firm size.
- Eurostat, 20% of EU enterprises use AI technologies, news release for the 2025 wave.
- OECD, AI adoption by small and medium-sized enterprises, discussion paper for the G7, December 2025.
- ICIC, AI in Business, the original research behind the 89% figure (US).
- European Commission, AI Act, the EU regulatory framework for AI.
- Eurostat DESI Index, where Slovakia stands on overall SME digitalization.
- Slovak Business Agency, support and data on small and medium business in Slovakia.