A resume is a claim. It is written by the candidate, about the candidate, and increasingly drafted by a language model that has already read the job description. Screening resumes well means ranking claims well. That is a narrower skill than predicting who will do the job.
Structured interviews replace the claim with a measurement. Every candidate gets the same job-relevant questions, in the same order, scored against the same written rubric. The scoring happens before anyone compares notes. That is the whole method, and almost nobody runs it completely.
What resume screening actually measures
Screening on a document rewards the qualities of the document. In a stack of 800 applications for one req, a recruiter spends somewhere between six and thirty seconds per resume. At that speed only a few signals survive, and none of them are the ones you care about.
- Writing polish, or access to a tool that supplies it
- Unbroken employment history, which quietly penalizes caregiving, immigration timelines, and layoffs
- Brand recognition, which tracks prior access more closely than future output
- Keyword overlap with a job description someone wrote in twenty minutes
- Formatting that a parser happens to read cleanly
None of those are worthless. All of them are proxies at least two steps removed from the behavior you are hiring for. The distance between the proxy and the behavior is where both bad hires and missed hires live.
The validity ranking, and the correction
Decades of meta-analytic work on selection methods produced a rough ranking of how well each method predicts job performance. Structured interviews and work samples sat near the top. Unstructured interviews sat in the middle. Years of experience and years of education sat near the bottom once a candidate was past the first few years of a career.
That literature was reanalyzed in the early 2020s. The reanalysis argued that older corrections for range restriction had been applied too aggressively, and most of the headline coefficients came down. Structured interviews landed closer to the high 0.30s than to the 0.50s that still get quoted in vendor decks.
The absolute numbers moved. The ranking did not. Design your process around the ranking.
The practical reading is unglamorous. No single instrument predicts performance well on its own. Structure is what turns an interview from noise into a usable signal, and the same structured question set costs you the same whether it predicts at 0.35 or 0.50.
What structure actually requires
Most teams that describe their process as structured are doing two or three of the five things below. Partial structure gets you partial benefit, and it gets you none of the defensibility you were hoping for when the process is questioned a year later.
- Questions derived from a documented job analysis, not from whatever the interviewer thought of that morning
- A behaviorally anchored rating scale with written examples for each level
- Independent scoring recorded before any panel discussion
- The identical question set for every candidate in the req, including internal candidates
- Retained transcripts and scores, so the process can be audited later
What changes in a real funnel
Across the hiring programs we ran in 2025, teams that moved from resume-first to interview-first screening changed the shape of the funnel more than its size. Interview-to-offer conversion moved from a median of 22 percent to 34 percent, because the people reaching the interview were now selected on something related to the interview. Median time from application to first scored signal fell from 6.1 days to under 9 hours, since Nova Recruiter generates the question set and the rubric from the job description and runs the interview asynchronously.
The composition change was larger than the conversion change. Candidates with employment gaps longer than six months made up 8 percent of the advancing pool under resume screening and 17 percent under interview-first screening. Their 90-day manager ratings came back at parity. That is nine points of qualified supply the document was hiding.
The tradeoffs
Structured interviewing costs more at the top of the funnel, because you are now measuring people you would previously have discarded for free. Rubric design is real work, roughly six to ten hours per role family, and it decays as the role changes. Hiring managers who trust their own instinct will experience the rubric as a constraint, which it is.
Resume screening does not disappear. It moves down to the narrow job it is good at: hard requirements that are true or false. Work authorization, licensure, location, a specific certification. Everything that requires judgment belongs in a scored interview, where you can see the question, the answer, and the reason for the score.
Dr. Maren AdeyemiHead of Assessment Science at Novexhire