Not All AI is Created Equal: AI Adoption and Firm-Level Productivity in Sweden

Sammanfattning

This study describes the development of AI adoption in Swedish firms between 2021 and 2025, and explores the associations between AI adoption and total factor productivity. Firm-level survey data from Statistics Sweden are used and three technology-specific AI indicators, text mining, natural language generation, and process automation, are regressed on total factor productivity in a two-way fixed effects model. The productivity analysis shows modest, but consistently positive, associations for all three technologies. Additionally, the different technologies show heterogeneity patterns that align well with the descriptive analysis. The language-based technologies show stronger productivity associations in service sectors, while process automation instead shows larger associations among large goods-producing firms. This suggests that associations are dependent on the specific AI technology being adopted, and on the firm in which it is used. The results therefore also highlight the limitations of measuring effects of AI as a single homogeneous technology.