Artificial intelligence has rapidly transitioned from theoretical promise to practical tool, quietly infiltrating business processes across sectors. Yet for all the attention, quantifying the true extent of AI integration in the real economy remains a challenge. This is partly a measurement problem. While surveys proliferate, most lack the granularity to distinguish between sectors where AI is genuinely embedded and those where it’s little more than a buzzword. For economists seeking a more precise picture, the ISIC framework, with its sectoral detail, offers both opportunities and complications.
The first step is to consider the existing ISIC structure. Certain subcategories seem obvious. Computer programming (ISIC 6201), data processing (ISIC 6311), and information service activities (ISIC 6399) are natural starting points. These sectors are not only early adopters but often the architects of AI solutions themselves. However, AI adoption is not confined to these obvious categories. Manufacturing, logistics, finance, even agriculture—each is experiencing its own version of digital transformation. The result is that AI investment, expenditure, and capability are spread unevenly across the ISIC landscape.
To monitor this phenomenon, economists are increasingly focusing on collecting firm-level data about AI-related expenditures and investments. The process, in theory, is straightforward: identify relevant ISIC codes for both core AI sectors and likely adopter industries; gather survey or administrative data from firms within those codes; and aggregate the results to identify trends, leaders, and laggards. In practice, there are obstacles at each stage.
Begin with the codes themselves. While ISIC provides a standardized structure, the boundaries are not always clear. For example, a company in ISIC 6201 may offer bespoke AI solutions for others, or it may simply develop generic software with limited AI application. Within manufacturing (say, ISIC 2829: Manufacture of other special-purpose machinery), firms may deploy AI in automation, predictive maintenance, or quality control, yet this activity may not show up explicitly in their ISIC classification. This blurring is inevitable, but it can be managed through careful supplemental data collection.
Next comes data gathering. National statistical agencies or industry bodies can survey firms within targeted ISIC codes, asking about AI-specific capital expenditure, operational spending, and the proportion of employees engaged in AI-related roles or projects. Question design matters here; generic queries about “digitalization” or “technology investment” often fail to capture what’s truly happening. It is better to break questions down: Does the firm deploy machine learning in production? Are neural networks or natural language processing tools part of daily operations? Is AI embedded in supply chain optimization or customer analytics? The more precise the questioning, the more meaningful the aggregated data will be.
Once firm-level data is collected, the aggregation phase begins. Here, ISIC codes provide the skeleton for analysis. By aligning reported AI expenditure with firm ISIC classifications, it becomes possible to compare rates of adoption across sectors, regions, and even over time. One can identify which industries are leading in AI integration—perhaps it is finance (ISIC 6419), which increasingly uses AI for fraud detection and algorithmic trading, or logistics (ISIC 5229), where route optimization and predictive warehousing are now commonplace. In other cases, results may surprise: agriculture (ISIC 0111) shows growing use of AI in precision farming and crop monitoring, a trend not always obvious to outsiders.
Care is needed when interpreting these aggregations. Not all reported expenditure translates to meaningful AI adoption. Some firms, especially in sectors newly exposed to AI, may label any digital investment as “AI” to signal modernity to investors or partners. Others underreport, either through lack of awareness or definitional confusion. Consistency in data collection, and explicit guidance on what counts as AI, are critical for maintaining data integrity.
There is also a temporal dimension. AI adoption is not a one-time event but an evolving process. Some sectors move quickly, others lag behind due to regulatory, cultural, or capital constraints. By regularly updating firm-level surveys and maintaining alignment with ISIC codes, economists can construct time series that reveal not only where AI is present, but how its footprint is changing. Sudden surges in AI expenditure in particular ISIC codes may indicate sectoral tipping points or the impact of policy interventions, while persistent stagnation can highlight structural barriers.
International comparability is another strength of the ISIC-based approach, provided countries harmonize their survey instruments and coding practices. Policymakers gain the ability to benchmark their economies against peers, identifying not just leaders and laggards in aggregate, but specific sectors where policy support might accelerate transformation. This level of detail is difficult to achieve through anecdote or generic digitalization metrics.
Of course, no classification is perfect. Some AI activity will slip through the cracks, especially as technology evolves and sectors converge. There is a case to be made for continued revision of the ISIC framework itself, perhaps with new subcategories or activity codes reflecting emerging AI-intensive business models. Until then, the combination of careful data collection, thoughtful aggregation, and honest interpretation remains the best tool for monitoring the march of AI through the economy.
Ultimately, ISIC codes provide both a map and a measuring stick. They enable economists to see where artificial intelligence is gaining ground, not in theory, but in the real, granular detail of industry activity. This clarity is essential, not only for academic understanding but for policymakers who must decide where to focus resources, remove obstacles, or simply keep pace with a technological revolution that refuses to stand still.