Technology Adoption: A Rate Is Not the Same as Integrated Use

In discussions about digital transformation, “adoption” can mean trying a tool, using it for some tasks or incorporating it into a workflow. These are different situations, even when summed up with the same word. A rate only makes sense alongside the definition behind it.

When an organization reports using a technology, that does not in itself show who uses it, how often they do so or how many activities involve it. The figure describes the reach as measured by the survey; it does not automatically mean that use is daily or that working practices have changed.

Funcas’s study of artificial intelligence in European businesses draws on Eurostat’s EU-ICT-Firm survey. The measure records the share of businesses that say they use some form of AI technology, but not how many workers or hours are devoted to it. It counts businesses reporting AI use, not intensity of use.

It is therefore useful to distinguish between availability, trial, regular use and integration into processes. An aggregate figure can help estimate how widely a technology has spread among the organizations surveyed, without answering the other questions. Its precise meaning depends on which response is treated as adoption.

Two percentages can describe different realities if they cover different populations, sectors or periods. They can also vary according to which technologies are included and how the question is worded. To interpret a figure, it is necessary to know who is counted and what qualifies as use; expressing both as percentages does not make indicators comparable.

The unit of analysis also matters. One measure may count businesses, while another concerns workers, investment or technological capital. The percentage of organizations reporting AI use is not the same as the share of employees who work with it: each indicator answers a different question.

In a 2020 analysis, CaixaBank Research used an approximation based on capital in new technologies: it combined software and databases, research and development, computers, and telecommunications equipment, then related that capital to hours worked. This is a measure of technological capital per hour worked, not a rate of businesses using AI.

An observation should not be confused with a projection either. A survey describes responses collected over a given period; a model can calculate future scenarios under specified assumptions. If the period, population or method changes, a difference between results does not, by itself, prove that actual use has increased or fallen.

A survey, a forecast and a list of trends serve different purposes. The first collects information about a population; the second estimates scenarios conditional on a model; the third organizes topics its authors consider relevant. They are not equivalent measures of adoption.

Funcas’s study describes business adoption of AI in the European Union and analyzes differences between countries and regions using EU-ICT-Firm. It also estimates diffusion paths and sets out future scenarios. These scenarios are not a count of businesses that have already adopted the technology: they depend on the modeling approach and its assumptions.

A list of trends can provide strategic guidance, but it does not show that a technology is widely deployed. Revista Byte TI, for example, published a selection of business trends in 2023 that included automation, AI, cybersecurity and superapps. That selection reflects an interpretation of priorities, not a survey of market penetration.

The useful question is what claim each type of material supports: a survey can estimate the proportion of a population reporting use of a technology; a model can explore scenarios; a list can put topics of interest into context. A technology’s strategic importance does not prove that its use is widespread.

A clearly defined rate can describe the proportion of organizations that report using a technology within the population and period studied. Funcas’s analysis finds differences in adoption between European Union countries and places Spain in an intermediate position in the context examined. The result describes differences in diffusion, not a uniform situation for every Spanish business.

The measure described by Funcas does not provide information about intensity of use. It therefore cannot show how many processes have been automated or how many employees use AI. Nor is it equivalent to a productivity assessment: the study notes that intensity of use, expressed, for example, in workers or hours devoted to AI, is unknown.

Technological capital and reported use are also different ways of observing digitalization. CaixaBank Research used capital categories to approximate the economic significance of certain technologies; Funcas, by contrast, analyzes the share of businesses reporting AI use. The metric chosen determines which conclusion is valid.

Funcas’s study of AI in banking illustrates another distinct unit of analysis: it uses a survey of 2,018 banking customers in Spain and examines their perceptions, knowledge and receptiveness to AI-based services. Those responses describe customers and their relationship with banking services; they do not measure the share of businesses that integrate AI.

For a business, a sector-wide rate can provide context, but it does not by itself describe how a particular implementation works. A decision to expand a tool requires attention to the process it is applied to, the people who use it and the goal being pursued. A general indicator cannot replace an assessment of the specific use case.

Funcas’s study shows a specific limitation of the EU-ICT-Firm statistics: they make it possible to observe how many businesses report using AI, but not the intensity of that use. For a small or medium-sized business evaluating a tool, that proportion does not answer whether the solution fits its tasks or whether its implementation is achieving the organization’s intended goal.

An internal assessment can ask more specific questions than a national rate: which task has incorporated the tool, who uses it and what change is expected. These questions help describe an organization’s own deployment, but the answers should not be confused with the result of a survey covering a broader business population.

The usefulness of each indicator depends on the question. The share of businesses reporting use describes reported reach; a measure of technological capital reflects another dimension; and a customer survey concerns perception and receptiveness. Before comparing figures, confirm that they describe the same phenomenon.

Before repeating an adoption rate, identify the population, technology, period and unit of analysis. Also check whether it is an observation or a projection. These details set the limits of what can be claimed: a share of businesses reporting AI use is not, by itself, a measure of frequency, scale or integration.

An increase in reported use allows us to say that a larger proportion of businesses are users within the relevant measurement. It does not, by itself, demonstrate a productivity increase or explain how the technology is used. The spread of use and its effects are different questions.

To understand the scope of a figure, keep separate the measures used to answer different questions: reported use among businesses, technological capital, estimated diffusion paths and customer responses. They can offer complementary perspectives, but each has its own population and subject of analysis.

In short, an adoption rate is useful when it remains tied to its definition, population and period. The central rule is not to equate availability, trial, regular use and integration. This helps prevent a scenario from being presented as an observed result, or a measure of reach from being turned into a claim about intensity or impact.