Managing Artificial Intelligence Sourcing And Oversight In European Enterprises

Abstract

Enterprise adoption of artificial intelligence in Europe accelerated sharply between 2023 and 2024, but adoption counts say little about what kind of AI is spreading or about who remains able to scrutinise it. This paper uses the Eurostat survey on ICT usage in enterprises to separate two questions that are usually merged. The first asks whether the way a firm acquires AI predicts its use of technologies that automate workflows and support decisions. The second asks whether acquisition mode predicts whether the firm monitors those technologies for bias against individuals. The analysis covers 133 country by size class by year cells from 24 European economies, together with the 2024 oversight module in the 11 countries where it is disclosed. The first question yields a null. Firms developing AI with their own staff do use workflow-automating AI more, but the association is a composition effect. The coefficient falls from 0.845 in a pooled specification to 0.058 once country and firm size are held constant, and a wild cluster bootstrap confirms it cannot be distinguished from zero. The second question yields a large and stable difference. Because bias-monitoring firms use more acquisition routes than adopters generally, their sourcing profile is rescaled before comparison. On that basis, firms building or modifying systems internally are over-represented among monitors, at ratios between 1.22 and 1.49, while firms purchasing ready-to-use commercial AI are under-represented at 0.69 and fall below parity in every one of the eleven countries. The gap between internally capable and externally sourced acquisition averages 0.584. Acquisition mode appears to shape the capacity to govern artificial intelligence far more than the choice to automate work with it.