Confeti Labs
Process & Structure

The hiring work happens before the interview

Confeti5 min readStructured hiring

AI will speed up scheduling, screening, and notes. It will not fix your real bottleneck, because the days do not leak out of scheduling. They leak out of the work nobody did before anyone sat down to interview.

Ask a team why hiring is slow and you will hear about calendars. Scheduling is genuinely painful. But it is the visible problem standing in front of the expensive one.

The expensive one is upstream. Nobody defined the role concretely. The recruiter and the hiring manager never agreed on which signals matter. So the loop runs without a shared map, and that is where the time, and the quality, actually disappears.

The uncomfortable evidence about interviewsNo map

Start with how little an unplanned interview actually tells you. In a now-classic set of three experiments, researchers had people predict students' performance. Some interviewers got real answers. Others got answers generated at random, the interviewee replying yes or no on a coin flip.

The interviewers who heard nonsense formed impressions just as confidently as those who heard real answers. And the interviews actively made predictions worse, diluting the good data the predictors already had. In a final twist, people so trusted the ritual that they would rather have a random interview than no interview at all (Dana, Dawes & Peterson).

That is what an interview without a defined target produces: a confident story built from noise.

Structure is the whole ballgameFront-loaded

The fix is not charisma or more interviews. It is design, decided before anyone sits down.

The first construct-specific meta-analysis of interview validity, drawing on 245 coefficients across 86,311 individuals, found that an interview predicts a given competency well only when it is deliberately built to assess that competency. And the single biggest lever is scoring: moving from an unscored “who did you like best” to a structured rubric with defined criteria raises the predictive validity of rater evaluations by more than 50% (Wingate et al., 2025, citing Kuncel et al.).

What a scoring rubric is worth
Predictive validity, indexed to an unscored preference = 100 (Kuncel et al., via Wingate et al. 2025)
Moving from "who did you like best" to a defined scoring rubric lifts the predictive validity of rater evaluations by more than 50%. The work that earns that lift happens before the interview.

Question design matters too. In the same body of work, past-behavior questions, asking what a candidate actually did, outpredicted loosely-targeted situational questions for managerial roles. Same candidate, different question, completely different signal. None of that is improvised in the room. It is decided in advance.

The work that actually matters

So the leverage sits before the first interview, in two hard tasks.

First, define the role concretely: not a job-description template, an actual picture of what this person must be able to do. Second, align the recruiter and the hiring manager on the competencies you will test, write the questions to hit them, fix the scoring, and split them across the loop so each interview has a distinct job.

Get that right and the interviews become productive. Each one extracts specific signal. None of them waste a candidate's time, or yours, re-covering ground.

What the best teams do differently

They treat the loop as one connected evaluation, not five separate conversations. Before each round, the next interviewer knows exactly what has been covered and what signal is still missing. “Behavioral has been tested three times. Nobody has any evidence on how this person handles ambiguity. Here are the questions that would close that gap.”

That is the upstream work paying off in real time. No wasted rounds, no surprise gaps at the debrief, no decision built on whoever was loudest.

The honest caveat.The interview itself stays human. AI should not conduct it, and it should not decide. An interview is a two-way conversation, not an interrogation, and the relationship is the point. What AI should do is capture it, structure it, and surface what is still missing, so the humans spend their judgment where it matters.

So, will AI speed up your hiring?

It will speed up the visible parts: scheduling, screening, notes. If those were your real bottleneck, you will feel it.

But if your loop runs without alignment, faster tooling just gets you to a shaky decision faster. The leverage is upstream, in defining the role, the questions, and the scoring before anyone interviews. Fix that, and speed takes care of itself.

Signal that compounds across the loop

Confeti tracks, per candidate across the whole loop, which competencies are covered and which signal is still missing, then assembles the evidence into a decision packet. The upstream work stops living in someone's head and starts compounding with every interview.

See how it works

Common questions

Why is my hiring process so slow?+

Often the visible delay is scheduling, but the real loss is upstream: an undefined role and misaligned interviewers running a loop with no shared plan, which produces rework and stalled decisions.

Do interviews even predict performance?+

Only when designed well. In classic experiments, interviewers given random nonsense answers formed impressions as confidently as those given real ones, and unplanned interviews made predictions worse.

How much does scoring matter?+

A lot. A meta-analysis found that moving from an unscored preference to a structured criterion-based rubric raises the predictive validity of evaluations by more than 50%.

What is a structured interview?+

One where competencies are defined in advance, questions are designed to assess them, and scoring is consistent across candidates. It is the strongest known predictor of job performance.

Should AI conduct the interview?+

No. The interview should stay a human, two-way conversation. AI is best used to capture it, structure the signal, and surface what is still missing for the human to decide.

References

  1. Dana, J., Dawes, R., & Peterson, N. (2013). Belief in the unstructured interview: The persistence of an illusion. Judgment and Decision Making, 8(5), 512–520. link
  2. Wingate, T. G., et al. (2025). Evaluating interview criterion-related validity for distinct constructs: A meta-analysis (245 coefficients, 86,311 individuals). International Journal of Selection and Assessment. link
  3. Insights From an Updated Personnel Selection Meta-Analytic Matrix (2024). American Psychological Association / University of Minnesota. link