Core finding
Brian Jabarian and Luca Henkel, authors of this study, conducted a field experiment at PSG Global Solutions involving entry-level customer-service jobs in the Philippines. They randomly assigned 67,056 eligible applications to human interviews, AI voice interviews, or a choice between the two. Human recruiters reviewed interviews and test results and made hiring decisions in every condition. Applicants assigned to AI interviews were 12% more likely to receive job offers. The offer rate was 9.73%, compared with 8.70% in the human-interview group: a 1.03 percentage-point increase. Applicants assigned to AI were also 18% more likely to start work and 18% more likely to have employment lasting at least one month. Ostensibly because they were a better match and the algorithm correctly predicted it, but this was not confirmed in the study. Productivity measures available for a subset of hires showed no detectable decline. Transcript analysis points to a possible explanation. AI followed the interview guidelines more consistently while still adapting questions to individual answers. The authors call this controlled variance: a common structure with room for useful follow-ups. Human recruiters actually asked more follow-ups, but the difference was largely concentrated on answers that already contained sufficient information. AI-assisted transcript analysis counted around 15 distinct job-relevant details per AI interview, compared with 10 per human interview. Among survey respondents, reported experiences were broadly similar, although AI interviews were rated less natural.
Takeaway
Here, AI’s promise lay in combining consistent information collection with human hiring judgment. The findings suggest that an interviewer’s effectiveness depends on both covering important topics and recognizing when an answer needs clarification. For candidates, a reasonable practical implication is to prepare concrete explanations of your experience, motivation, and fit for the role. For employers, the design goal is a consistent interview that remains responsive to what each applicant says. These findings apply most directly to high-volume hiring with clearly defined requirements.
Better stories are possible — and they start with preparation.
AI Interview Training Course