One-Way Interviews Cost You Half Your Applicants

Swap a live interview for a one-way recorded one, and roughly half your applicant pool disappears.

I'm not pulling that number from a vendor deck. Economists Mallory Avery, Edwin Ip, Andreas Leibbrandt and Joseph Vecci ran a pre-registered field experiment on more than 3,000 real applicants for real vacancies, randomly routing each one into an asynchronous audio interview, an asynchronous video interview, a live online interview, or no interview step at all. Every response then got scored twice: once by professional recruiters, once by a commercial AI recruitment tool of the kind already used across much of the Fortune 100. Their results are published as Monash Economics Working Paper 2026-05.

The Numbers

In the group with no interview step, 85% of applicants stayed in the process. In the asynchronous group, only 40% completed the step, a 45 percentage point drop. The live online interview, for comparison, only cost the funnel 17 points.

That comparison is really the whole finding. Both formats ask a candidate to sit an interview. Only one of them asks them to do it alone, into a camera, on their own time, and that difference alone is worth about 28 points of your pipeline.

Who Actually Drops Out

The easy defence of a one-way interview is that it filters for motivation. If someone won't record five minutes of video, how serious were they to begin with?

That theory doesn't hold up against the data. The study scored candidate strength four different ways, covering education, experience and CV quality, and on every measure the pattern ran the opposite direction: the people who walked away from the recording were, on average, the stronger candidates. A gender split showed up too. Women finished the async step 5.1 points less often than men, a gap that wasn't there in the group given no interview at all.

The filter is real. It's just not filtering on the thing most teams assume.

Where the AI Came Out Ahead

This is where the study stops confirming what you'd expect.

The researchers followed those applicants for the next twelve months. The commercial AI scorer predicted who actually succeeded on the job noticeably better than the professional human recruiters did, and it scored women and underrepresented racial minority candidates higher than the humans scored them, while the humans skewed toward men and White and Asian candidates. The paper documents several cognitive biases in the human ratings that simply don't show up in the AI ones.

I want to be careful here. I'm not arguing that AI should run your interviews. I studied AI at university and use it weekly, and I still haven't met a tool that can tell me whether someone can reason through a problem they've never seen before. But most of the debate about automated hiring has focused on the scoring being the risky part. This study suggests the scoring held up fine. The damage came from the format wrapped around it.

Why People Actually Leave

A separate vignette experiment in the same paper dug into why applicants drop out, and the answer wasn't laziness or technical friction. Candidates were reading the interview format itself as a signal: how competitive the process looked, and how much the company valued their time.

That's worth sitting with, because it's a rational read. A recorded, one-way interview says, in effect, that you haven't earned a human hour yet. If a strong candidate already has two live interviews booked elsewhere that week, the recorded one is the one they skip. That's exactly why it's the qualified candidates who leave. Deterrence sorts by who has other options, and having other options tends to correlate with being good at the job.

The Honest Trade-Off

Live interviews don't scale for free. They cost engineering hours, they're hard to schedule, and if you're screening four hundred applicants for one role, you can't put a senior engineer in front of all of them. Anyone who tells you async screening has no upside hasn't run high-volume hiring.

Choosing async isn't the mistake. Choosing it blind is.

What to Do Instead

Three changes, all small enough to start this week.

Report completion rate next to invite count, on the same slide, every month. If you invited 200 people to a one-way interview and 84 finished it, your real funnel starts at 84, not 200. Most dashboards only count the people who completed, which makes a shrinking pool look stable.
Use audio instead of video where you can. The study found video versions had 6 percentage points lower completion than audio-only. That's a free improvement.
Put the async step after a human touch, not before it. A short recruiter call first changes what the recording means. It stops reading as "we haven't invested in you yet" and starts reading as "we've already started."

And if you're hiring for a senior or scarce role, ask whether the async step belongs there at all. Its entire justification is volume. When you're chasing a handful of genuinely qualified people, a step that removes the most qualified ones isn't a filter. It's a leak.

The Question

If you run one-way interviews, you can answer this today: how many people did you invite last month, and how many finished?

If nobody on your team can produce both numbers, that's the finding.

Source: Avery, M., Ip, E., Leibbrandt, A., & Vecci, J. (2026). A Brave New World of Hiring: A Natural Field Experiment on How Asynchronous Interviews and AI Assessment Reshape Recruitment. Monash Economics Working Paper 2026-05. Pre-registered as AEARCTR-0013356.