Recommender System in the Labor-Market

This project’s multidisciplinary team of economists and data scientists has partnered since 2018 with the French Public Employment Service (PES) to leverage its rich administrative data and operational infrastructure to design and test recommender systems for matching jobseekers and vacancies. This collaboration, known as the VADORE project (Valorisation de Données de Recherche d'Emploi), has developed AI-based tools to support PES caseworkers in identifying the most suitable candidates for employer-posted vacancies.

This project evaluates the impact of algorithm-assisted recommendations on recruitment outcomes, labor demand, and fairness. The researchers implement a large-scale randomized evaluation using a saturation design, with randomization first at the “micro-market” level (sector × geography) and then at the firm level within each market. The intervention targets pre-selection vacancies—where PES advisors are responsible for shortlisting candidates—and tests whether AI-generated suggestions improve the efficiency, speed, and quality of recruitment.

The researchers will assess the average performance of the recommendation system, its cost-effectiveness for both PES and firms, and its distributional effects. In particular, they examine whether algorithmic tools reproduce or amplify biases—such as under-ranking women or jobseekers from disadvantaged neighborhoods—and whether some profiles are over-proposed, creating congestion and reducing overall match quality.

The researchers will also test whether improved recommendations increase employer engagement and vacancy posting behavior. France Travail is the key implementing partner and will benefit from evidence on the value and risks of adopting algorithmic tools at scale.