
Bulgaria Ranks Last in the EU on Generative AI, and Most Firms Say They Don't Need It
Four in ten Bulgarian companies use generative AI. That sounds like progress until you see the rest of the chart: every other EU country is higher. In the EIB Group Investment Survey 2026, published by the European Investment Bank on 9 November, Bulgaria comes 27th of 27, with 40% of firms already using tools such as ChatGPT, Gemini, Claude or Copilot. Greece is at 45%, Italy at 48%, the EU average at 67% and Denmark at 89%.
The ranking is the headline. The more useful part is a question the survey asked the firms that do not use AI at all: why not? Their answer is not what most people expect.
What the survey measured
The EIB has run this survey every year since 2016, with Ipsos. The 2026 wave covers about 12,000 firms across the EU and a comparison sample of 800 in the United States. The AI question is specific: does the company make systematic use of generative AI tools to improve processes, and in which areas? A firm where one person sometimes asks a chatbot for help with an email is not what is being counted.
For Bulgaria, the EU overview report shows:
- 40% of firms already use generative AI;
- 23% do not, but are planning or evaluating it in the next 12 months;
- 32% have no plans to use it.
A year ago the 2025 country overview put Bulgarian usage at 22%, against an EU average of 37%. So adoption has grown, but the EU average grew faster, and the gap to the average is wider than it was.
The EU-level pattern explains some of it. Across the EU, 27% of small and medium-sized firms have no plans to adopt generative AI, compared with 9% of large firms.
The answer to "why not?"
Among Bulgarian firms that do not use generative AI, the main reasons, according to the same report, are:
- they see no need for it (35%);
- they are uncertain about its potential benefits (32%);
- they lack skilled staff (16%).
Together, the first two reasons cover two thirds of the non-users. The EU picture is similar: 43% see no business need and 28% are unsure what it would bring. This is not mainly a story about money, infrastructure or access to tools. Most of these companies could open a ChatGPT or Copilot account tomorrow. What they lack is a clear picture of what it would do for their business.
That matters, because the two problems have very different fixes. You solve a shortage of money with financing. A shortage of understanding needs something else: someone has to show, on the company's own work, what changes.
Using AI is not the same as getting something from it
The second chart that deserves attention is about firms that already use AI. The survey asked when they expect most of the benefits to show up.
Across the EU, 21% of users say the benefits are already visible. In the United States, the figure is 35%. In Bulgaria, it is 16%, and another 31% expect to see the benefit within a year.
In other words, even among the Bulgarian companies that are counted as AI users, only about one in six can say it is already paying off. The rest are waiting.
The EIB report adds a pattern that may explain part of the gap. European firms use generative AI mostly for internal processes (50% of all EU firms) and much less for growth-related work, such as marketing and sales (34% in the EU against 50% in the US) or product development (17% against 23%). And 69% of EU users say the main benefit they expect is operational efficiency, while only 11% expect it to drive sales and revenue growth (19% in the US).
The EIB's broader conclusion is about adoption in general, not only AI: 20% of EU firms report significant investment in innovations that are new to the firm, compared with 44% in the US. Europe is not short of inventions. It is slow to take what already exists and put it to work.
My reading
Here is where I move from the data to interpretation.
If a third of Bulgarian non-users say "we don't need it" and another third say "we don't know what it would give us", then a campaign telling them to "adopt AI" will not move them. "Adopt AI" is not a decision anyone can make on a Monday morning. It has no owner, no deadline and no way to tell if it worked.
The same applies to many companies that already count as users. A few licences and some enthusiastic employees will put a firm into the "already using" column of a survey. It will not put it into the "benefits already visible" column.
What does work is narrower. Pick one recurring task that costs real time every week, measure how long it takes today, change how it is done with AI, and measure again after a few weeks. If the number moved, you have a business case written in your own data. If it did not, you have lost a few weeks, not a budget.
A hypothetical example
Imagine a Bulgarian distributor of building materials with a small sales office. Every day, customers send enquiries by email and phone: is a product in stock, what is the delivery time, can they get a quote for a list of items. Two people in the office spend a large part of their day searching the stock system and retyping the same information into replies.
If the owner were asked in the EIB survey, the honest answer might well be "we don't need AI". The business works. Nobody has tried to describe that office work as a process that could change.
Now suppose the company takes that one task. For two weeks, the team simply logs how long each enquiry takes to answer. Then it sets up an assistant that drafts the reply from the enquiry and the stock data, and a person checks and sends it. After another four weeks, they compare. The owner now has a concrete answer to "what would AI give us?" for one process, and a reason to decide on the next one, or to stop.
This is also where the "lack of skilled staff" reason gets smaller than it looks. The team does not need to become AI engineers. It needs to learn how to describe its own work clearly, which is the part AI training for a specific team should focus on.
What the numbers do not tell us
The survey is representative and the sample is large, but the EU overview gives the Bulgarian results mostly in charts, not in a detailed country table. The EIB usually publishes separate country overviews later, and the 2025 Bulgarian edition included sector detail, such as low usage in construction. When the 2026 country overview appears, it will be worth checking which sectors are pulling the average down.
The survey also does not tell us how well firms use AI, only whether they use it systematically and where. A firm that counts as a user may be using it badly, and a non-user may have made a perfectly reasonable choice for now.
Still, the main signal is clear enough. For most Bulgarian companies that stay out of AI, the main barrier they name is not price and not technology. It is the absence of one concrete, measured example from their own work. That is a much cheaper problem to solve.


