Return of the Solow-paradox in AI? AI-adoption and firm productivity

AI-related technologies have become ubiquitous in many business contexts. However, to date empirical accounts of the productivity effects of AI-adoption by firms are scarce. Using Finnish data on job advertisements between 2013 and 2019, we identify job advertisements referring to AI-related skills. Matching this data to productivity data from ORBIS, we estimate the productivity effects of AI related activities in our sample. Our results indicate that AI-adoption increases productivity, with three important qualifications. Firstly, effects are only observable for large firms with more than 499 employees. Secondly, there is evidence that early adopters did not experience productivity increases. This may be interpreted as technological immaturity.Thirdly, we find evidence of delays of least three years between the adoption of AI and ensuing productivity effects (investment delay effect). We argue that our findings on the technological immaturity and the investment delay effect may help explain the so-called AI-related return of the Solow-paradox: I.e. that AI is everywhere except in the productivity statistics.