AI adoption and productivity are positively associated in the cross-country analysis reported by Bendo and colleagues (2026). The authors also report that stronger digital infrastructure strengthens this relationship. Their paper draws attention to the conditions surrounding AI adoption, although its limited sample and reporting leave substantial questions about how confidently we can interpret the results.
For educators and institutional leaders reading productivity claims, the useful starting point is to examine what was actually measured. In this study, productivity means GDP per worker. It does not mean better teaching, improved student learning, or verified gains within an individual organization.
What counts as adoption, and what counts as productivity?
Adoption can describe very different practices. A business might use AI in one function or integrate it across several operations. The measure used here captures firms reporting AI use in at least one business function. It therefore tells us little about the depth, quality, or purpose of that use.
The productivity measure also needs attention. GDP per worker describes national economic output relative to the workforce. It can support comparisons between economies, but it cannot tell us which particular businesses became more productive after introducing AI.
These distinctions affect the questions we can answer. National indicators may reveal patterns worth investigating. Explaining what happens within a workplace requires evidence about its tasks, people, resources, and outcomes.
This is also why my discussion of whether AI increases productivity needs to consider how work changes alongside output. Different studies can use the same word while examining quite different experiences and measures.
What the AI adoption and productivity study found
Bendo and colleagues combined secondary data from sources including the OECD AI Policy Observatory, Stanford AI Index, World Bank, and McKinsey. They selected five economies: the United States, United Kingdom, China, India, and the Philippines.
The analysis used descriptive comparisons, correlations, and regression models to examine AI adoption, digital infrastructure, and productivity. Infrastructure measures concerned internet penetration, connectivity, and cloud-computing readiness; productivity was represented by GDP per worker.
The authors report a positive association between AI adoption and productivity. Their models also suggest that the association is stronger where digital infrastructure is more developed.
Separate industry comparisons show higher reported adoption in technology and finance than in the sectors the paper groups as less data-intensive. These findings concern patterns in the assembled indicators. They do not establish that adopting AI caused productivity to increase.
The study’s small selection of economies limits generalization, and combining sources introduces questions about comparable definitions and reporting periods. The authors acknowledge limited diagnostic and robustness testing.
The regression table does not report the number of observations, and the methods do not clearly explain how many matched observations entered each model. This makes the statistical claims difficult to assess independently. The paper also does not directly examine organizational practices or mechanisms within individual firms.
Reading productivity evidence before making decisions
The authors recommend investment in digital infrastructure, workforce capabilities, and further research using larger samples and firm-level evidence. Their analysis makes infrastructure a relevant question, but it cannot determine which investment would deliver a particular benefit in a university or school.
I would use this paper as a prompt to examine the evidence behind an adoption proposal. Start by defining the intended outcome. Faster completion of administrative work, better decisions, and stronger student understanding require different measures. A national productivity indicator cannot substitute for evidence about those local outcomes.
Then ask what the proposed tool needs to work effectively in that setting. Connectivity and data availability are relevant starting points from the paper. Staff preparation, institutional support, and the suitability of the task would need their own examination. My review of AI and work in higher education provides a more directly relevant route into that institutional discussion.
For researchers, the reporting questions are especially important. A reproducible cross-country analysis needs identifiable source indicators, dates, the rules used to match them, and a clear observation count. Readers should be able to understand whether a model uses country averages, repeated country-year observations, or another structure.
Further work would also need to examine alternative explanations. Economies with greater resources may have both stronger infrastructure and higher productivity. Productive organizations may be better positioned to adopt AI. Establishing the direction of these relationships requires a design capable of testing it.
Within AI and research methodology, I would read this paper as an exploratory analysis whose claims need more transparent reporting and further testing. Its most useful contribution to institutional discussion is the question it raises about the conditions surrounding adoption. Those conditions deserve examination before productivity gains are promised.
Reference
Bendo, M. C. D., Alviar, D. C., Ulgado-Rosas, M. V., Clarin, A. V., Carpio, L. S., & Gomez, A. M. (2026). Artificial intelligence adoption, digital infrastructure, and business productivity: A cross-country empirical analysis using secondary data. International Review of Management and Marketing, 16(6), 292–302. https://doi.org/10.32479/irmm.24231
