Small Teams, Big-Company Reach

A decade ago, a five-person firm that wanted to answer every enquiry within minutes, in several languages, at any hour, had two options. It could hire people it could not afford, or it could give up the ambition. Larger competitors held that ground because they could pay for a service desk, a content team and a back office.

Part of that advantage can now be rented by the month. AI does not turn a small business into a large one, but it changes which activities a small business can offer with a straight face. The evidence on how much it changes is uneven, so this article sets out what the better studies show, where small firms in Europe stand, and what remains out of reach.

In April 2025 Intuit QuickBooks commissioned a survey of more than 2,200 US businesses with up to 100 employees. It found that 68% now used AI regularly, up from 48% in July 2024. The most common uses were marketing (43%), customer service (36%) and administrative tasks (33%), followed by data processing and bookkeeping.

These are the functions that grow with headcount in a large company: a marketing team, a service desk, an administrative office. They are also the functions a small firm never has time for. When a model can draft the reply, the post or the invoice reminder, the owner reviews instead of writes, and the work gets done at all.

Two further capabilities follow from how the tools work. A model drafts in Spanish, Bulgarian or German at roughly the same cost as in English, which matters in markets where customers and visitors write in several languages. And it answers at three in the morning without overtime. No published survey yet measures either effect for small firms, so both points are inferences rather than statistics.

The U.S. Chamber of Commerce reported in August 2025, from a nationwide survey run with Teneo Research, that 58% of small businesses say they use generative AI. The figure was 40% in 2024 and 23% in 2023.

The OECD found a lower share in a different sample. Its 2024 survey of more than 5,000 small and medium-sized enterprises in Austria, Canada, Germany, Ireland, Japan, Korea and the United Kingdom showed generative AI in use in 31%. Those firms reported that it improves performance and helps compensate for skill gaps and labour shortages.

Vendor surveys are more enthusiastic. Salesforce’s sixth Small and Medium Business Trends report, based on 3,350 small business leaders, says 91% of SMBs using AI report a boost in revenue. That number comes from adopters describing their own results, with no comparison group, and should be read as sentiment more than measurement.

The strongest reason to expect a levelling effect comes from field studies that compare workers of different experience. Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied 5,179 customer support agents at one company as a generative AI assistant was introduced in stages. Productivity, measured as issues resolved per hour, rose 14% on average and 34% for novice and low-skilled workers, with minimal impact on experienced ones. The authors suggest the assistant spreads the practices of the best agents to newer staff.

A 2023 experiment led by Boston Consulting Group found a similar pattern among 758 of its junior consultants. On tasks inside the capabilities of GPT-4, those using it reached a level of performance 40% higher than those without it, and the lower a person’s baseline proficiency, the larger the effect tended to be. On the creative product task, the lowest-proficiency participants improved by 43%.

For a small firm the implication is practical. It rarely has a specialist for every function, so most of its staff are, in effect, novices at most of what they do. A tool that lifts the less experienced closer to the standard of the best is worth more to that firm than to a large one with deep benches of experts. This is an interpretation of two studies, neither of which examined small businesses directly.

Eurostat’s 2025 survey shows that 20.0% of EU enterprises with ten or more employees used AI, up from 13.5% in 2024. Size matters a great deal: 17.00% of small enterprises (10 to 49 staff) used it, against 30.36% of medium-sized and 55.03% of large ones. The statistic excludes firms with fewer than ten employees, so the smallest businesses are not counted at all.

Bulgaria sits near the bottom. Its overall rate was 8.55%, the third lowest in the EU after Romania (5.21%) and Poland (8.36%). An analysis of the Eurostat table published on the AI portal ai.bcci.bg gives the Bulgarian rates by size for 2025: 7.17% for small enterprises, 13.30% for medium-sized and 26.18% for large.

The EU’s Digital Decade programme sets a target that three in four EU companies use cloud computing, big data or AI by 2030. The target is a combined measure covering all three technologies, and Eurostat puts the 2025 value at 63%, twelve points short.

For a small Bulgarian firm, these numbers describe an unusually open field. Roughly nine in ten of its size-class peers report no AI use, so the firm that answers quickly in several languages and publishes steadily is likely to stand out locally. The same figures also show that large European companies are well ahead, and that the window will not stay open indefinitely.

The same BCG experiment recorded the downside. On a task that sat outside the frontier of what the model could do, participants using GPT-4 performed 23% worse than those working without it. BCG’s summary of the experiment is that people mistrust the tool where it adds great value and trust it too much where it is not competent.

That finding limits what a small firm should promise. A reply drafted at speed is valuable only if it is right, and the cost of a wrong answer to a customer is the same for a five-person firm as for a bank. Adoption statistics also say nothing about quality: a business that uses AI is not necessarily a business that uses it well.

The OECD reaches a related conclusion at policy level. It stresses the need for structured support to close digital and skills gaps between SMEs and larger firms, which suggests the levelling depends on skills and setup, and does not follow automatically from buying a subscription.

The sensible starting point is the function where customers wait longest, or where the firm currently says no: the language it does not serve, the hours it is closed, the enquiries that sit unanswered over a weekend. Pick one, write down what a good answer looks like, and decide in advance which cases a person must take over.

Then review a sample of the output every week for the first months. The studies above suggest that the gains are real for people who are still learning a task, and that errors are most likely at the edge of what the tool can do. A short routine of checking catches those edges before customers do. This is the kind of setup TriplePilot builds, though the test applies to any supplier.