Choice as Signal: Designing AI Adoption in Labor Market Screening
with Pëllumb Reshidi
SSRN

We study the design of human-AI screening systems in a hiring environment where applicants choose between interviewing with a human recruiter or an AI voice agent, a choice that creates a new and informative signal. Once firms condition on screener choice, welfare reversals emerge: choice benefits firms and high-ability applicants, but leaves low-ability ones worse off than under a predetermined assignment of the screening technology. Using data from a large-scale field experiment-where 70,000 applicants were randomly assigned to a human interviewer, an AI agent, or allowed to choose between them, we develop a structural estimation to quantify how choice-as-signal shapes match quality and welfare. This framework also allows us to evaluate alternative screening systems, with preliminary results suggesting higher welfare under joint human-AI screening than under either technology alone. Overall, we show that AI adoption in screening is a design problem rather than a simple human-versus-AI substitution decision.

Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews
with Luca Henkel
SSRN

Job interviews are a key stage in hiring through which firms collect information about potential employees, yet they often produce noisy signals of match quality. We study the impact of automating job interviews with AI voice agents. Partnering with a recruitment firm, we conducted a natural field experiment in which 70,000 applicants were randomly assigned to be interviewed by human recruiters, AI voice agents, or given a choice between the two. In all three conditions, human recruiters evaluated interviews and made hiring decisions based on applicants' performance in the interview and a standardized test. Contrary to the forecasts of professional recruiters, we find that AI-led interviews increase job offers by 12%, job starts by 18%, and job retention up to four months by 16-18% among all applicants. Analyzing interview transcripts reveals a key mechanism driving these results: AI agents achieve 'controlled variance'. They follow interview guidelines more consistently, cover a more uniform set of topics, and reduce interviewer-driven dispersion while remaining responsive within each conversation, which is associated with more hiring-relevant information collected from applicants. In response to AI-led interviews, recruiters score the interview performance of AI-interviewed applicants higher, but place greater weight on standardized tests in their hiring decisions. Applicants accept job offers with a similar likelihood and rate both the interview quality and the recruiter similarly in a customer experience survey. Moreover, when offered the choice, 78% of applicants choose the AI recruiter. Overall, our results provide evidence on the types of environments where AI automation may be most effective: by automating noisy stages of information collection, AI can improve human decision-making.