Strip away the ritual and ask what the standard interview actually measures: composure under social evaluation, fluency in talking about work, and resemblance to the interviewer's mental image of "good." All real traits — and almost none of them the job. Decades of selection research keep delivering the same verdict: the unstructured conversation-style interview barely predicts performance, while interviewers' confidence in it remains serenely untouched.
Gut feel is pattern matching on the patterns you already prefer.
From Ceremony to Sample
The reframe that fixes most of it: every interview hour should be a sample of the work, or of evidence about the work. Three formats carry nearly all the predictive weight:
1. The work sample
Have them do a slice of the actual job: critique this real (anonymized) plan, debug this code, draft this client reply, work through this dataset. Keep it short, paid if substantial, and identical across candidates. One hour of watching someone do the work outperforms five hours of hearing them describe it.
2. The structured behavioral interview
Past behavior in similar situations remains the best conversational evidence. The structure is the active ingredient: same questions for every candidate, anchored to the role's defined competencies, with probing follow-ups ("what did you specifically do? what happened next?") that separate participants from protagonists. The moment interviewers freestyle, you're back to measuring chemistry.
3. The realistic dialogue
Reserve genuine conversation for what conversation is good at: letting the candidate sample you. Honest answers about the team's challenges, the role's frustrations, the manager's style. Selection runs both directions, and the best candidates are sampling harder than you are.
Process Hygiene That Multiplies Everything
- Score independently, before the debrief. The first strong opinion voiced in a debrief anchors everyone; written scores submitted beforehand preserve the panel's actual information.
- Assign each interviewer a lane. Four people asking "tell me about yourself" is one interview with redundancy. Divide the competencies; cover the map.
- Decide what disqualifies before you meet anyone. Criteria invented after the fact bend toward whoever charmed the room.
- Time-box the chemistry check. Likability matters a little — as one input, not the verdict. Label it honestly in the scorecard and cap its weight.
- Close the loop with reality. Twice a year, compare interview scores against actual performance of past hires. This is the audit almost nobody runs — and the fastest cure for confident bad judges, who are otherwise promoted by their own confidence.
The Candidate-Experience Dividend
Here's the happy alignment: the predictive process is also the respectful one. Candidates consistently rate work samples and structured interviews as fairer than vibe-chats — they can show what they can do rather than perform charisma. You get signal; they get dignity; the charming empty suit gets, at last, an honest score.