Identifying circular public procurement with large language models: An application to firm-level circular innovation

Policy makers increasingly recognize circular public procurement as a demand-pull instrument for stimulating the transition to a circular economy. However, empirical studies on circular public procurement have been hampered by a fundamental measurement challenge, as public procurement databases do not contain structured ways of identifying circular projects. This paper presents the first application of an LLM-based semantic similarity approach to identify circular procurement at scale. Adapting a bibliometric text-embedding approach from circular economy research, we show its application in comparing tender descriptions to a reference corpus of circular economy scientific abstracts, generating circularity scores for each award. We then apply the identified circular public procurement awards in an empirical study of firm-level adoption of circular economy innovation, matching the classified tenders to German data from the Community Innovation Survey. The results show that firms winning circular procurement are more likely to introduce circular economy innovation after three to five years, while no significant results are found at shorter or longer time horizons. Overall, this paper demonstrates the potential of using LLMs to identify circular public procurement and study its effectiveness in enabling the circular transition.