AI is already used in business, but integration and results remain uneven

Artificial intelligence is taking up an increasingly prominent place on corporate investment agendas, but that alone does not show that the way people work has changed. Access, frequent use and integration into processes are distinct stages: making tools available to staff is not the same as incorporating them into operations and decision-making.

Deloitte’s global State of AI in the Enterprise 2026 study, based on more than 3,200 business and technology leaders in 24 countries, reflects this difference: around 60% of workers have access to AI solutions approved by their organization, but fewer than six in ten employees with access use them daily. In addition, only one in four organizations has moved at least 40% of its AI initiatives into production. Tool availability does not equal sustained use or production deployment.

Each percentage needs to be read in light of its definition and population. “Using AI” can refer to different practices, while a spending forecast expresses future intent, not completed investment. For that reason, adoption rates and business survey results should not be treated as direct measures of productivity across the whole economy. Access, investment and impact are different indicators.

In its presentation of the study in Spain, Deloitte says 85% of surveyed companies expect to increase AI investment in the next fiscal year; nearly a third anticipate increases of more than 20%. The report also finds that 68% identify efficiency and productivity as the main benefit they have obtained. The first figure is a spending expectation; the second is a reported benefit, not a measure of productivity across Spanish businesses as a whole.

Deloitte’s result comes from its business survey, not a census of every company in the country. Other measurements may use different definitions and samples, so their adoption rates are not necessarily directly comparable. CaixaBank Research describes AI adoption by Spanish businesses as rapid, but still limited and uneven. Taken together, the picture is one of progress with differences between companies, not a uniform transformation.

For a business leader, the national figure is no substitute for an internal assessment. It is more useful to establish whether the organization provides access, whether staff use the tools regularly and whether they have been incorporated into specific processes. Maturity shows up in the work actually done with AI, not just in licenses or announced budgets.

Deloitte reports that only one in four organizations has moved at least 40% of its AI initiatives into production. This figure describes project progress in the global sample; it does not mean that the other organizations have no systems in production at all. It does show that moving from testing a solution to operational deployment remains a challenge for many of the organizations studied. A pilot and a production system are different stages.

The report also says that 84% of organizations have not redesigned jobs or operational workflows to integrate AI. This finding helps explain the gap between having tools and changing business activity: when adoption amounts to adding technology on top of existing processes, organizational integration is still pending. Redesigning work remains a significant task, alongside technology adoption.

The Inter-American Development Bank offers a regional comparison, not an estimate that applies to Spain or Europe: in Latin America and the Caribbean, its analysis distinguishes tool use from reported economic returns and from integration into core operations. The comparison helps separate concepts; it does not make results across regions equivalent. The spread of a technology and the capture of value should be assessed separately.

Business integration also raises questions about skills and management. Deloitte reports that, in Spain, 59% of organizations are training their general workforce to improve practical AI knowledge, while 54% have begun reskilling initiatives. These figures describe measures reported by surveyed organizations; on their own, they do not indicate each program’s reach or results. Training is among the business responses to AI deployment.

The same report says that 70% of surveyed Spanish organizations express high or very high concern about the use of their proprietary data, and 49% identify regulation and governance as one of the main barriers to adopting generative AI. These are survey responses, not a count of incidents or a ranking of risks. Data management and governance are part of the adoption debate.

For a specific decision, these issues translate into questions about how tools are used and supervised: what information is involved, who is accountable for the process and how its use is controlled. The need to answer them depends on the business use and the data involved. A case-by-case assessment avoids treating all applications as if they had the same requirements.

Before deploying a solution, a company can establish a baseline for the process it wants to change. Depending on the objective, it might track time per task, volume handled, errors or response time. It should then compare the same indicator and account for relevant integration, licensing, training and operating costs. Without a prior measure, it is difficult to attribute a change to the initiative.

It is also useful to distinguish an operational improvement from a financial result. A faster task can be meaningful, but its economic value depends on how it affects the company’s work and costs. Return cannot be inferred from an investment forecast or from a general perception that AI improves productivity. The metric should match the project’s specific purpose.

To decide whether to continue or expand a pilot, leadership can identify who owns the process, what result is expected and which criteria will be used to assess progress. Deloitte’s figures on initiatives in production show why it is worth distinguishing experimentation from deployment. A decision to scale requires information about the specific use case, rather than relying solely on general interest in the technology.

Differences between organizations are not explained by budget alone: bringing in AI also involves changes to processes, training and governance, all of which feature in Deloitte’s analysis. However, the cited figures do not make it possible to attribute results to a single factor or conclude that a company’s size determines its ability to benefit. Broad explanations based on one survey should be treated with caution.

Before expanding an initiative, leaders can check four things:

• Process: which activity the initiative is meant to change.
• Indicator: which result will be tracked and its starting point.
• Responsibility: who will oversee deployment and results.
• Costs: what resources it requires beyond the tool.

These questions do not guarantee a return. They help clarify what is being evaluated and distinguish a test from an operational application.

In 2026, Deloitte’s and CaixaBank Research’s figures describe active investment and uneven adoption, not a definitive measure of AI’s effect on productivity across the Spanish economy. For each company, the useful decision is to see whether a particular application becomes part of the intended process and what result it produces. Evidence of use and integration is more informative than a budget expectation, although economic impact needs to be measured separately.