Causal Inference in Econometrics Using Machine Learning: Estimating the Effect of AI and Automation Adoption on Firm Productivity in Europe

Abstract

Artificial intelligence and automation are becoming central to how European firms work, compete, and organize their production. But despite the rapid growth of these technologies, there is still a key question: does adopting AI genuinely make firms more productive, or are already-productive firms simply more inclined to adopt it? This study addresses that question by combining established econometric methods with newer causal machine learning techniques. Using a large panel of European firms from 2010 to 2023, built from Orbis and EU KLEMS, AI adoption is identified through both investment measures and text based disclosure indicators. Across multiple empirical approaches including fixed effects, difference in differences, and instrumental variable models the results consistently show productivity gains of roughly 3% to 6% among AI adopting firms. Double Machine Learning produces a similarly robust estimate of around 4.5%. Event study evidence further indicates no pre adoption improvements, with productivity gains emerging gradually afterward. The effects, however, are uneven. Larger firms, those with more advanced digital systems, and firms employing a higher share of skilled workers benefit noticeably more from AI adoption. In contrast, firms lacking strong digital foundations or sufficient human capital see smaller gains. The results indicate that adopting AI does boost firm productivity, but the size of the benefit depends heavily on the presence of complementary skills and digital infrastructure.