AI Is Entering the Business. When Does It Become Productivity?

AI adoption is growing, but introducing a tool and achieving results are separate steps. A European working paper offers insights into where the strongest signals are emerging and why management matters.

In my previous article on Italy’s 14,840 AI-focused companies, I asked where value was being created. The InfoCamere and University of Padua study described a young, expanding and highly fragmented ecosystem. Its growing supply of technological expertise still faces the challenge of building sustainable businesses and improving customers’ competitiveness.

The working paper The effects of AI on Europe’s productivity, produced by the Eurochambres Chief Economists Group, turns attention to the other side of that relationship: the businesses using AI.

That shift matters. The number of providers tells us something about entrepreneurial activity. Understanding its economic contribution requires looking at what changes inside customer businesses.

Adoption is increasing, but gaps persist

According to the Eurostat data reported in the paper, the share of EU enterprises using AI increased from 7.65% in 2021 to 19.95% in 2025.

Progress remains uneven across countries and sectors. Small and medium-sized enterprises consistently lag behind larger businesses.

Access to technology does not automatically provide the ability to use it effectively. A small business can purchase a subscription and begin experimenting. Embedding that technology into daily operations requires usable data, skills, time and clear ownership.

This is often where the first difficulty appears. An enthusiastic employee starts testing a tool, while the business problem and expected outcome remain loosely defined.

Several months later, the company knows more about the technology but still struggles to explain what has improved.

Where the strongest signals emerge

The European analysis finds a positive association between AI diffusion and labour productivity. Image recognition and process automation show statistically significant positive associations. Among business applications, marketing and sales display the strongest positive relationship.

For managers, this directs attention towards applications that fit into identifiable workflows with measurable outcomes.

Potential examples include handling commercial enquiries, classifying leads, updating customer records and reducing manual handovers between marketing and sales. These are practical illustrations, rather than specific outcomes established by the study.

Value needs to be assessed across the process. Producing more sales messages can increase activity without improving sales. Reducing response times for qualified enquiries may address a meaningful commercial problem, provided the company measures what happens afterwards.

An AI project should therefore begin with a straightforward question: which indicator will tell us whether this investment has been worthwhile?

A positive relationship is not proof of causation

The findings require careful interpretation.

The authors explicitly describe the research as associational rather than causal. AI adoption rates are measured at an aggregate country-and-sector level and linked to business financial data. The analysis does not directly measure the return achieved by an individual company adopting a particular solution.

Labour productivity is also measured through operating revenue per employee. This differs from the value-added-per-employee measure used in the Italian study discussed in my previous article. The findings should not be treated as directly comparable.

No coefficient in the working paper can therefore become a reliable promise about the return on a specific business project.

The evidence helps identify areas worth investigating. Evaluating an investment remains the company’s responsibility.

Italian businesses show interest, with important decisions still unresolved

The paper also includes findings from an Istituto Guglielmo Tagliacarne–Unioncamere survey of 4,500 Italian businesses employing between 5 and 499 people.

Within that sample, 16.1% report using AI tools on a stable basis, while a further 33.2% plan to introduce them within three years. These figures refer to a particular survey population and should not be directly compared with studies using different coverage.

Among current and prospective adopters, 24.8% have not yet decided which type of solution to implement.

This creates a substantial need for business guidance. Before selecting a provider, companies need to understand which activities consume resources, where errors occur and what information is available.

The survey’s analysis also finds that businesses using customised AI or a combination of standard and customised solutions are 5.1 percentage points more likely to expect turnover growth than those relying solely on standard tools.

These are expectations, not realised financial results. Nor does the finding establish that custom development is always the best option. It does, however, raise a useful question about the relationship between adapting technology to business needs and confidence in future growth.

Controls belong inside the implementation plan

Among the risks identified by the Italian businesses surveyed, 47.5% cite cybersecurity and privacy, while 33% highlight inaccurate or unreliable information.

The latter is an operational management issue. An answer can be clear and persuasive while still being wrong. If it enters a proposal, a customer communication or an internal decision without appropriate checks, greater speed can increase the cost of error.

Defining who checks outputs, which sources are acceptable and when specialist judgement is required forms part of implementation.

Verification time also belongs in the economic assessment. If an activity becomes faster but requires extensive corrections, its actual benefit may be smaller than an initial demonstration suggests.

The next question is how long the value lasts

A comment on my previous article raised an important question: how much of the value created today will last?

This working paper adds useful evidence about the relationship between AI diffusion and productivity. It does not establish the durability of any competitive advantage.

Companies need to keep evaluating results beyond the pilot. Do the benefits hold when volumes increase, employees change and tools are updated? Is the time released put to better use? Does the improvement reach the customer?

These questions bring management back to the centre of the discussion.

Choosing a relevant problem, organising implementation and reviewing outcomes over time determine whether technological potential becomes a business capability.

AI can contribute to productivity. Turning that contribution into lasting value requires a company able to manage its use.

Source: Eurochambres Chief Economists Group, “The effects of AI on Europe’s productivity”. This working paper is intended for research and discussion and does not constitute an official Eurochambres position. Findings should be interpreted within their respective samples and observation periods.