Authored with Victor Alfonso Naya, Guillaume Bied, Bruno Crépon, Philippe Caillou, Christophe Gaillac, Elia Pérennes and Michèle Sebag)

We examine the properties of a recommender algorithm currently under construction at the Public Employment Service (PES) in France, before its implementation in the field. The algorithm associates to each offer-job seeker pair a predicted ``matching probability’‘ using a very large set of covariates. We first compare this new AI algorithm with a matching tool mimicking the one currently used at the PES, based on a score measuring the ``proximity’‘ between the job seeker’s profile or preference and the characteristics of the offer. We detail and discuss the trade-off Designing labor market recommender systems: the importance of job seeker preferences and competition (with Victor Alfonso Naya, Guillaume between matching probability and preference score when switching from one system to the other. We also examine the issue of congestion. We show on the one hand that the AI algorithm tends to increase congestion and on the other hand that this strongly reduces its performance. We finally show that the use of optimal transport to derive recommendations from the matching probability matrix allows to mitigate this problem significantly. The main lesson at this stage is that an algorithm ignoring preferences and competition in the labor market would have very limited performances but that tweaking the algorithm to fit these dimensions substantially improves its properties, at least ``in the lab’’.