42% of Europe’s smaller firms use AI, mostly as an experiment. Two EU studies show where it stalls.

The European Commission published two studies on 14 October, at the start of European AI Innovation Month. One is a Flash Eurobarometer survey of small and medium-sized companies and mid-caps. The other is an analysis by the Commission’s Joint Research Centre (JRC) of how AI use has grown in ten sectors the EU treats as strategic. They measure different things, but they arrive at the same picture: plenty of companies have started, and few have taken AI into the work that actually makes them money.

What the survey found

The survey asked about AI together with algorithmic management, meaning systems that help managers organise staff or decide on tasks, schedules, performance or staffing. 42% of the companies already use such tools, and the Commission says most of them do so in a limited and experimental way. Another 21% plan to start.

Customer service and client relations lead the uses at 47%. Next come monitoring performance and productivity (36%) and logistics, planning and staff scheduling (30%). 61% say AI helps them make decisions, yet 90% say people still take those decisions, and only 2% of decisions run fully automated. Almost half, 49%, do not trust AI or automated systems for decisions about staff.

The barriers are close to a three-way tie: high investment costs (50%), legal uncertainty (49%) and lack of expertise (48%).

Where adoption stops

The JRC analysis uses Eurostat’s enterprise survey for 2021 to 2025. In the ten strategic sectors, from energy and manufacturing to tourism and agri-food, the share of firms using AI rose from 7% to 18%. Across the rest of the economy it rose from 8% to 20%. Energy went from 18% to 34% and electronic communications from 12% to 30%, while mobility, tourism and agri-food trail the field.

The JRC attributes most of the growth to cheap, general-purpose tools that write or generate text and audio for marketing and sales. Those tools are easy to start with, and that is exactly the problem the report describes. In many firms AI still has to reach core processes such as production, where the JRC sees the biggest productivity gains. Its data also show that firms which reorganise how they work around AI are more likely to use it in production, and that buying ready-made tools does not by itself move a company in that direction.

Asked what holds them back, firms in most strategic sectors name a lack of relevant expertise first. The JRC picks out two dominant bottlenecks: scarce specialist skills, and regulatory uncertainty, including worries about data protection and privacy.

Rules on AI at work are coming

The Commission links the survey to the upcoming Quality Jobs Act, which it says will give companies more clarity and better protect workers’ rights where algorithmic management is used. That matters for the 36% who already use AI to monitor performance. A tool a company adopted as an experiment can end up making staff decisions, and those decisions will draw closer scrutiny.

What an Estonian company should do on Monday

• ✳Count your experiments honestly. If AI in your company means a few people writing emails and marketing copy, you sit in the 42%, and you have not yet reached the part the JRC links to real productivity gains
• ✳Pick one core process, in production, delivery or finance, and decide what AI should change there. Then change the way the team works around it. The JRC data point to reorganisation, not more licences, as what moves AI into core work
• ✳List every place where AI or software already influences decisions about staff: performance tracking, shift planning, recruitment screening. Write down who makes the final call in each case, before the Quality Jobs Act asks
• ✳Treat lack of expertise as the first problem to solve, because it is the one firms name most. Train the people who own the process, so the knowledge stays in the company and does not leave with one enthusiast
• ✳Answer the legal uncertainty with a short internal rule on what data may go into which tool. It costs a day of work, and it removes one of the three biggest reasons companies give for holding back

Both studies point the same way. Starting with AI is now cheap and common. The companies that pull ahead will be the ones that move it out of the side experiments and into the work their customers pay for.