Using AI is not implementing it

On October 14, the European Commission's Joint Research Centre published an analysis of artificial intelligence adoption across the 10 sectors the European Union regards as strategic, from energy to manufacturing, health, and tourism [1]. The headline is upbeat: the share of firms using AI in those sectors rose from 7% in 2021 to 18% in 2025. Across the European economy as a whole, it rose from 8% to 20%. More than double in four years.

The interesting part is the mechanism. According to the report itself, the growth comes mainly from accessible, low-cost general-purpose tools that analyze or generate text and audio for marketing and sales [1]. Put differently, much of what Europe counts as adoption consists of producing commercial texts and materials with standard tools. It is useful. It is not yet a transformation of the business.

Where the value is

The study is blunt: in many cases AI has not yet reached core processes such as production, which is where the largest productivity gains lie. Firms that adapt the tool and integrate it into their production, workflows, or infrastructure tend to obtain more value added [1].

Two findings should unsettle anyone who believes the problem is choosing the right vendor. The first is the correlation between reorganization and productive use: the firms that invest most in structural organizational change to integrate AI are also those that use it most in production. This suggests prioritizing organizational change, without establishing a causal sequence [1][2]. Energy, telecommunications, and defense lead that combination. The second is almost a paradox. Manufacturing, with low overall adoption, relies more on off-the-shelf solutions than the climate sector, and this does not translate into greater integration in its production processes. Buying the ready-made tool does not, on its own, lead to using it where it matters.

Speed is uneven too. Capital-intensive sectors are moving faster: energy firms using AI rose from 18% to 34%, and electronic communications firms from 12% to 30%. Mobility, tourism, and agri-food lag behind, with uses still concentrated in marketing, sales, and other accessible tasks [1].

Expertise is the main bottleneck

The main obstacle firms report in almost every sector is neither cost nor technology but a lack of expertise, followed by regulatory uncertainty and concern over data protection [1]. It is worth reading carefully. What is scarce is rarely someone who knows how to ask an assistant for something, but someone who understands both a business process and what the tool can do within it. That profile cannot be bought with a license.

What holds beyond Europe

The figures are European and should not be transposed as they stand to other markets. The mechanism, however, travels. A license is a purchase; an implementation is a redesign. The question a board should ask is not which assistant to contract, but which process it is willing to redo: how a quotation is prepared, how a maintenance order is planned, how a loan is approved, how a purchase is decided. If the answer is none, the company will appear in the statistics as an adopter without having demonstrated an improvement in its results.

The European report proposes watching three things: how far the firm has reorganized to take advantage of AI, how much it uses AI in production, and how far it depends on standard rather than custom-made solutions [1]. It is a reasonable self-assessment for any management team. The first two tell you whether AI has changed anything; the third helps assess how closely the solution fits the process. Provider dependence also requires examining portability, contracts, and control of data.

Sources

ONLINE SUBSCRIPTION

Receive every new analysis and invitation to open events.

This is a separate public category: it does not confer membership, admission or access to the private portal.

Subscribe for free

Would you like to discuss how this affects your organization?

Request a private conversation