Research by: Daniel Broby & Florian Gerth
Executive Summary
The paper examines how the rapid diffusion of artificial intelligence (AI) challenges established methods for measuring productivity. Traditional decomposition techniques, such as Griliches-Regev and Foster-Haltiwanger-Krizan, assume homogeneous firm responses and smooth substitution between inputs. Broby and Gerth argue that these assumptions are increasingly untenable. AI alters production by automating routine and cognitive tasks, changing the marginal productivity of labour and capital. It does this by complementing intangible assets like proprietary data and organisational know-how. These factors create heterogeneity across firms and introduce non-linear effects that standard models fail to capture.
The paper reviews leading decomposition frameworks and their limitations when applied to AI-driven change. Classic approaches disaggregate industry productivity into within-firm improvements, reallocation between firms, and net entry or exit. While these methods illuminate structural dynamics, they risk misattribution when technology adoption is uneven or when intangible inputs dominate. For example, averaging firm productivity across two periods, as in Griliches-Regev, can obscure the role of rapid AI-induced divergence. Similarly, Foster-Haltiwanger-Krizan’s covariance term may overstate allocative efficiency if intangible capital is unobserved.
Building on this critique, the authors present a new conceptual framework linking AI adoption to productivity growth channels. They emphasise within-firm effects from improved technical efficiency and deployment of intangibles, and between-firm reallocation as market share shifts toward AI-capable firms. Net entry dynamics are also affected, as entrants may be AI-native while legacy firms fail due to integration challenges rather than low productivity per se.
The central methodological innovation is an extension of the Cobb-Douglas production function to include AI-specific capital and complementary intangibles. The modified function is:
This formulation allows the marginal impact of AI to vary with organisational readiness and intangible investments, directly addressing the heterogeneity of technology diffusion.
Broby and Gerth contend that empirical analysis must move toward decomposition methods that reflect these structural realities. Techniques by Diewert-Fox and Melitz-Polanec, which benchmark entry and exit against contemporaneous survivors and model the joint distribution of productivity and market share, are highlighted as more suitable. Such approaches improve the measurement of within-firm gains and reallocation effects in economies undergoing rapid AI adoption.
The article concludes that recalibrating productivity metrics is crucial for credible economic analysis and sound policy. Without methodological updates, productivity gains from AI may be underestimated or misattributed. This can lead to flawed conclusions about economic growth and competitiveness. The work provides a structural foundation for integrating AI’s impact into productivity statistics and offers guidance for economists, statistical agencies, and policymakers seeking to understand the digital economy’s transformative effects.
Keywords: total factor productivity; decomposition techniques; artificial intelligence
To cite this article: Broby, D., & Gerth, F. (2025). Artificial intelligence and the limits of the measurement of productivity. Annals of Social Sciences & Management Studies, 12(2), 555831. https://doi.org/10.19080/ASM.2025.12.555831
To access the article: DOI: 10.19080/ASM.2025.12.555831
About the Journal
| The Annals of Social Sciences & Management Studies are an international journal which focuses to spread a vast knowledge to the readers. ASM includes various topics which cover every part of Sociology and management studies like Philosophy & Public Affairs, new econometric techniques, estimation, testing, forecasting, and policy analysis, business ethics, business-government relations, corporate governance, corporate social performance, and environmental-management issues, didactical, methodological and pedagogical subjects, industrial metabolism, product stewardship, industrial symbiosis, dematerialization and decarbonization, finance, economics, mathematics, and statistics, political science, sociology, economics history, philosophy, psychology, social anthropology, legal and educational disciplines, evaluation studies in such fields as child development, health, education, income security, manpower, mental health, criminal justice, and the physical and social environments, historical investigations, philosophical reflections, and empirical research, strategic management, sociology, economics, political science, history, information science, communication theory, and psychology, foreign policies, international relations, international and comparative political economy, security policies, environmental disputes and resolutions. | |
| Publisher | Juniper Publishers |
| Review System | Peer-reviewed |
| Chartered Association of Business Schools Academic Journal Guide 2024 | Not Ranked |
| Scimago Journal & Country Rank | Not Ranked |
| Scopus | Not Ranked |
| Australian Business Deans Council Journal List | Not Ranked |
| Journal Citation Reports (Clarivate) | Not Ranked |
