Why async AI interviews cost you your best lateral hires

Updated 4 November 2026

When you are hiring someone who is already doing the job, your interview process has to earn their time. A senior candidate may be speaking with several employers while managing a full-time role. Asking them to record answers alone, with no conversation or chance to clarify, can give them a reason to leave your process before a recruiter has assessed them. A 2026 field experiment involving 3,296 applications found that fewer applicants continued after a one-way recorded interview than after a live online interview or no interview step. The result is not a forecast for your roles, but it is a reason to look closely at the format you ask experienced candidates to complete.

The decision for a lateral hiring team is not whether to screen candidates. You still need to understand their experience, technical or functional ability and fit for the role. The question is how to get that evidence without making a strong candidate complete a one-way task that feels like a dead end.

What the evidence means for your lateral hiring process

The paper, A Brave New World of Hiring, is a 2026 working paper by Mallory Avery, Edwin Ip, Andreas Leibbrandt and Joseph Vecci. It is a field experiment, not a peer-reviewed journal article. The researchers advertised three technology-related roles in the United States and randomly assigned applicants to different next steps. The study is useful because it observes what real applicants did, rather than asking people to imagine how they might respond.

The 3,296 applications were assigned to one of three broad routes: confirm interest without an interview, attend a live online interview, or complete an asynchronous interview. The asynchronous group included audio and video interviews. Applicants were asked to answer set questions without an interviewer present. The researchers then compared how many people continued with their application.

That is what asynchronous means here: candidates record answers on their own time, and nobody is present in the conversation. It is different from a live AI interview, where the system asks questions and follows up in real time. The distinction matters. The study should not be used as evidence that every AI-led interview has the same effect as a one-way recording.

Random assignment makes the comparison more informative: applicants were not simply choosing their preferred format. But the experiment still has boundaries. It took place in one market, covered three advertised roles and tested particular interview experiences. A different role, candidate pool, invitation or interview length could produce a different result.

The study is not a direct test of lateral hiring, senior technology recruitment or candidates approached by a recruiter. It used applicants responding to US job advertisements. Its continuation rates show how the interview format affected willingness to continue in that experiment; they are not a forecast for your own pipeline, and they do not identify the quality of every person who left.

How many applicants continued with each format

The figures below show the share who completed the next step assigned in the experiment: confirming continued interest, attending a live interview or completing a recorded interview. They are not measures of interview quality, job performance or the number of strong candidates ultimately hired.

The groups were not equally sized. The live-interview group contained only 94 applications, compared with 667 in the no-interview group and 2,535 in the asynchronous group. That makes the live estimate less precise, so avoid treating the 68% figure as a stable benchmark for your own hiring.

For your hiring process, the useful question is not whether to remove screening. It is whether a one-way recording gives you enough additional evidence to justify the friction it creates. If you need to understand how someone thinks, clarify an achievement or probe a role-specific skill, a live exchange may give you more useful information while letting the candidate take part in an actual conversation.

Experienced applicants were especially likely to leave after a one-way interview

Among applicants in the top 10% by experience, 89% continued in the no-interview group and 36% continued after an asynchronous interview. In this paper, the top 10% by experience means applicants with at least 14 years of relevant work experience. The paper also reports a similar pattern for applicants in the top 10% by CV score: 92% versus 44%. These are groups defined within the study, not a universal definition of a high-quality candidate. The results suggest that the format may deter some applicants with stronger measured experience or CVs; they do not tell us the quality of each person who left.

Use the calculator below only to understand the scale of the difference if the study's rates were applied to a shortlist of your size. It is an illustration, not a forecast. Your own continuation rate may be very different.

For example, applying the study's 89% and 36% rates to five experienced applicants would produce roughly four continuations without an interview and two after an asynchronous interview. Do not read that as a prediction that two of your five candidates will disappear. Use it as a prompt to measure your own funnel by format and seniority.

This is particularly relevant in IT services hiring. A lateral candidate may have a current project, client commitments and a narrow window to speak with a recruiter. You need to check technical depth, delivery experience and availability, but the first screening step should make that conversation easier to start, not harder to complete.

Candidate experience is also a fairness issue

Women were 5.1 percentage points less likely than men to complete an asynchronous interview in the study (p=0.01). The estimated difference in the effect of moving from no interview to asynchronous was smaller and did not meet the conventional 0.05 significance threshold (p=0.055). Treat the first result as evidence of a disparity in this sample, not proof that every asynchronous process will have the same gender effect.

A p-value is one way researchers describe how compatible a result is with a chance-only explanation under a statistical model. It is not the probability that the finding is true. The 0.055 estimate is close to, but above, the commonly used 0.05 threshold; it should be described as suggestive rather than conclusive.

Does the one-way format matter more than AI scoring?

The researchers used a 2×2 design within the asynchronous interviews: candidates recorded audio or video, and each recording was evaluated either by a human or by AI. This let the paper compare the format and the scoring method separately. Video recording reduced continuation by about 6 percentage points compared with audio (p=0.003). AI evaluation was associated with a smaller 3-point reduction, which was not statistically significant (p=0.170).

The regression implies approximate continuation rates of 44% for audio scored by a human, 41% for audio scored by AI, 38% for video scored by a human and 35% for video scored by AI. These are approximate model-implied estimates, not raw rates reported in the headline results. The study found a 6-percentage-point difference between video and audio, while the 3-point difference between AI and human scoring was not statistically significant.

For employers, the practical point is to evaluate the candidate experience separately from the scoring method. A live AI interview is not automatically equivalent to a live human interview, and the study does not establish that AI scoring has no effect. It does show why you should ask whether the candidate can have a real-time exchange, explain an answer and respond to relevant follow-up questions.

Keep candidate completion and assessment quality separate

For applicants who were unemployed when they applied, the researchers found that the AI assessment score predicted who was in work 12 months later, while the recruiters' scores did not show a reliable relationship. This is a separate finding about the predictive relationship in that sample. It does not mean the tool caused better employment outcomes or that its score can be transferred to another employer or product.

The AI assessment tool in the paper is not Fabric, so this result says nothing about Fabric's scoring accuracy or validity. For your hiring decision, keep two questions separate: will candidates complete the interview format, and does the assessment give your recruiters useful, job-related evidence? You need to test both.

Why applicants may decide not to continue

The researchers also ran a separate vignette experiment with 606 US-based people who said they would apply for one of the roles. Participants were shown a hypothetical invitation to a live or asynchronous interview and asked how they viewed the process. They believed more applicants were invited to asynchronous interviews and rated that process as less fair. This helps explain why candidates may hesitate, but it measures stated expectations in a hypothetical scenario, not the reasons for every real applicant's decision.

A 2026 Greenhouse survey offers a separate, self-reported view. It covered 2,950 job seekers across five markets; the figures in the next paragraph refer to the 1,200 US respondents. A survey can show what people say they dislike, but it does not measure the behaviour observed in the field experiment.

In the US sample, 38% said they had left a hiring process because it included an AI interview. Respondents could identify several reasons: 33% cited a pre-recorded video interview scored by AI with nobody present, 27% cited employers not explaining how AI would be used, and 26% cited AI monitoring during the process.

The survey also found that 70% said they had not been clearly told that AI would evaluate them, and 21% said they first learned this when the interview began. These are respondents' recollections, not an observed dropout rate. Still, the implication for employers is straightforward: explain the process before asking someone to take part.

For a lateral role, make the first assessment feel like a useful conversation about the job. Tell candidates what the AI interviewer does, how long the interview takes, which skills are assessed, whether a recruiter reviews the result and how they can reschedule or ask for help.

A checklist for reviewing your current interview format

Use these questions to review the interview process your team currently asks experienced candidates to complete. The aim is not to assume every recorded interview is unsuitable. It is to check whether the format gives you the evidence you need without losing candidates before a recruiter has spoken to them.

Start with the format: can candidates speak with the interviewer in real time, and can it ask relevant follow-up questions? Then check disclosure, duration, scheduling and recruiter review. If the process cannot answer those questions clearly, pilot a different format with one open lateral role before making it standard for senior or hard-to-fill positions.

Once the interview format is right, the rest of the experience matters too. See Fabric's guide to candidate experience for practical ways to make each step clearer for applicants.

What a live, two-way AI interview changes

The field experiment did not test Fabric or every type of AI interview. It found that applicants in this sample were more likely to continue after a live online interview than after a one-way recording. A live conversation gives candidates the chance to clarify an answer and respond to follow-up questions. That is the experience to look for when you want to assess an experienced candidate without asking them to speak into a camera with nobody on the other side.

Fabric's live AI interview is a two-way video conversation, not a recording of answers to a fixed list. It can ask follow-up questions based on what the candidate says. For a lateral hiring team, that means you can explore a candidate's experience and probe role-specific examples in the same interview. This describes the product experience; it does not mean Fabric will reproduce the study's 68% continuation rate. Measure completion in your own process.

The hiring manager sets the interview length, from 30 minutes to an hour or more. Candidates can join through a phone browser without installing an app. Fabric supports weak 4G connections and lets a candidate rejoin after a dropped connection. Test the experience on the devices and networks your candidates actually use, especially if you are hiring across locations.

Fabric supports AI disclosure and recording consent in more than 35 languages, and the recruiter selects the interview language. Candidates can reschedule. Still explain the process in the invitation in plain language; an in-product disclosure should reinforce that explanation, not replace it.

Recruiters can review the interview result. Make that review meaningful: give reviewers the role criteria and the evidence behind the assessment, and let them question a result rather than simply approve a score.

Questions lateral hiring teams should ask

Do candidates really drop out of one-way recorded interviews?

In this field experiment, 40% completed the asynchronous interview, compared with 68% who attended the live online interview and 85% who confirmed continued interest when no interview was required. The live group was small, and the result applies to the study's US job applicants and interview setup. It is a reason to measure your own completion rates, not a universal benchmark.

Is a live interview better than a recorded one for completion?

The study found a higher share completed the next step after a live interview than after an asynchronous one: 68% versus 40%. However, only 94 applications were assigned to the live group, so the estimate is less precise. Among applicants in the top 10% by experience, 77.78% completed the live interview and 36.46% completed the asynchronous interview; 89.04% of the control group confirmed continued interest.

Does AI scoring itself reduce completion?

The study's 2×2 design lets the researchers compare recording format and scoring method separately. Video was associated with about 6 percentage points lower completion than audio; AI scoring was associated with a 3-point reduction, but that difference was not statistically significant. In this sample, the format mattered more clearly than who scored the answer. The experiment still does not establish how candidates respond to a live AI interview.

What should candidates be told before an AI interview?

Tell candidates in the invitation that AI will be involved, what it will do, which parts of the interview are recorded, what the assessment covers and who reviews the result. The Greenhouse survey found that many US respondents did not recall receiving clear disclosure. Clear communication helps candidates decide whether and how to take part; it should not be left until the interview has started.

Try a live interview on one open lateral role

Choose one open role and compare your current screening step with a live, two-way interview. Track invitations, starts, completions, withdrawals, time to complete and recruiter decisions. Break results down by seniority and relevant candidate groups where lawful and appropriate. Ask candidates what made the process easy or difficult. Keep the format that gives your team useful evidence without unnecessarily losing people before the first recruiter conversation.