Joseph is currently an Associate Professor in the Department of Economics at the University of Gothenburg, Sweden. He completed my PhD in 2016 at Monash University, Australia. His primary research interests are in the fields of development economics and behavioural economics.
Title: The Future of Hiring: How Automated Screening Tools Shape Job Applications and Hiring Decisions
Authors: Mallory Avery, Edwin Ip, Andreas Leibbrandt, Joseph Vecci
Abstract: Automated screening tools—such as AI-based video interviews—are rapidly reshaping the recruitment landscape. Globally, over 60% of large employers now report using some form of automated screening in hiring. With low marginal costs, these tools make it easier and cheaper to collect and process extensive applicant data. However, additional screening stages impose costs on applicants and there is little evidence on how they perform in terms of predicting applicant quality relative to human assessors. In this project, we conduct a large scale field experiment to examine the impact of automated screening tools on job application behavior and then compare how these tools perform relative to alternative methods. Over 3,000 real job applicants are randomized into receiving different forms of automated screening including asynchronous audio or video interviews, synchronous Zoom interviews and a control group with no additional screening. Candidates job interviews are then evaluated by a international HR firm and one of the largest AI recruitment firms in the US. The CV of all candidates are also assessed. We find that, compared to a no-screening control, asynchronous Audio- or Video-interview stages are associated with a decrease in continuation by 44%-53%; alternatively, a more traditional synchronous zoom interview sees continuation decrease by only 9% compared to control. Women are generally more deterred than men by these additional screening stages, though asynchronous Video-interview screening. When comparing candidate evaluations—that is, the scores assigned to applicants’ interview responses—we find that AI produces substantially more diverse shortlists. Women and underrepresented minorities receive higher interview scores from AI than from human evaluators, leading to 20% more women and 13% more under represented minority candidates in the top-ranked pool. Furthermore, when we follow up with candidates one year later, AI evaluations are over three times more predictive of applicants’ subsequent career success than those of human recruiters.