Fighting Corruption with Artificial Intelligence: Searching for Suitable Public Procurement Data in the EU

Abstract

This study aims to provide a comparative overview of the suitability of public procurement data for AI-driven anti-corruption analysis in the European Union and its candidate or associated countries. Such an analysis goes beyond just the public procurement data itself. Algorithms that seek to flag clusters of corruption or anomalies in public procurement require contextual information that procurement notices do not provide, such as details of the (ultimate) beneficial ownership of companies, whether their directors have donated to the ruling party or have a political connection, and whether rival bidders sit on each others’ boards. Without such additional information, risk scores tend to collapse into simple indicators for corruption. Despite increasing digitalization of procurement processes, key challenges remain in terms of data completeness, standardization, and accessibility. Our research confirms that all EU Member States now publish procurement data of sufficient quality for training machine-learning models to detect corruption. We consider the full data landscape, connecting public procurement data with information about beneficial owners, political connections, media ownership, and complaints. Croatia, Estonia, and Latvia stand out among the EU-27 as front-runners, while the Netherlands, France, Austria, Germany, the Czech Republic, and Portugal also appear broadly AI-ready.