Thomas Le Barbanchon

Title: How can AI improve search and matching? Evidence from 59 million personalized job recommendations

Abstract: We explore how artificial intelligence can be leveraged to help frictional markets to clear. We design a job recommender system that uses job seekers' history of job ad views to generate relevant personalized job recommendations. We deploy it at scale on the largest online job board in Sweden: recommendations are shown at the bottom of search engine webpage. We design a clustered two-sided randomized experiment to evaluate its impact on job search and labor-market outcomes. First, at the labor market level, the recommender system increases employment by 0.4\% within 5 months. Second, at the worker-level, treated job seekers are more likely to apply to recommended vacancies, at the expense of non-recommended ones: the total number of applications remains unchanged. Treated job seekers have 0.6\% higher employment within the 6 months following first exposure to recommendations, with no evidence of negative impacts on wages. Third, we study the heterogeneity of the employment effect at the job seeker-vacancy level and find that workers with lower job-finding probability and jobs with lower hiring probability are those benefitting the most from the recommender system.