Algorithms increasingly shape hiring decisions, yet field evidence on how they affect efficiency, fairness, and trust remains limited. We study these issues using administrative data from a large Italian job-matching platform in the hospitality sector, Restworld, which oversees the full hiring pipeline from application screening to final placement. Our data cover more than 250,000 applications and exploit a staggered transition from fully human-driven shortlisting to hybrid and AI-driven regimes. We first document substantial disparities in human shortlisting decisions: migrant applicants are significantly less likely to be shortlisted, even after conditioning on rich worker and job characteristics. These penalties emerge at the screening stage and vary widely across human reviewers, highlighting the role of individual discretion. We then compare human and algorithmic shortlisting. The AI system selects candidates with stronger experience, job stability, and pre-assessment signals, while humans place greater weight on CV presentation and native-language indicators. The introduction of AI substantially expands the shortlisted pool and increases interview and hiring rates for both natives and migrants. Finally, we outline ongoing survey and field experiments that study employer trust in AI, including randomized variation in the visibility of algorithmic recommendations and the role of keeping humans in the loop.
Co-authors: Silvia Barbareschi and Francesca Miserocchi
