How to run an interview that produces evidence instead of impressions. Read it before any interview round, for staff hiring or programme selection.
Interviews are where hiring goes wrong most quietly. The candidate talks for 30 minutes, everyone leaves with a feeling, and the feeling gets a score. Senior candidates are the most practised interviewees, so the more senior the role, the more an unstructured impression flatters them.
Two findings shape this pattern:
- Structured interviews (same job-related questions, anchored scoring) are the strongest single predictor of job performance in a century of selection research (r = .42, Sackett et al. 2022 ). An average interviewer with a good script beats a gifted one improvising.
- The same research shows structured interviews vary most in how well they work (±.24). Leading questions, stacked questions and shallow probing throw the result away. A 45-minute preparation on the technique below captures most of the gain.
"Tell me about a time when…" predicts; "what would you do if…" does not. Memory is episodic, and a story gives you something to probe. Open with:
"Share three or four accomplishments you were proudest of in the last three years."
Each part of the wording does a job. Three or four: the candidate proposes, you choose which to unfold, and the set shows what motivates them. Proudest: emotion points to drivers. Three years: detail (who said what, what you were thinking) fades after about 18 months.
A variant for early-career or programme candidates: "Three examples from the past three years where you started something that made a difference. Give each a movie title." Then pick one.
Alternate them:
- : What led up to it? Who was involved? What happened next? What was the outcome?
- : What did you do? What did you say? What did they say? What were you thinking?
Something the person did, said, thought or felt:
- , not on reflection afterwards
- , not "we" (ask who did what)
- , not in generalities
A story can feel like a 4 on your rubric and never demonstrate the competency. "We made the call; leadership was supportive" leaves the question open: decided before or after leadership weighed in? If you did not ask, you are scoring impression.
- : "So you brought the stakeholders together first?"
- : "What would you do in the director's shoes?"
- : "How do you approach leadership?"
- : "Why did you do that?" invites justification, not recall
- : three questions in one breath; the candidate answers the easy ones
Ask one question. Then be quiet.
Bias is not stoppable; it is manageable. The usual ones: familiarity ("people like us"), halo and horns, confirmation, primacy and recency, anchoring on the first strong candidate.
- , with someone who sees differently from you. Where you can, pair across nationalities: cultural fit is where familiarity bias hides.
- for every candidate. Write the rubric before you meet anyone.
- Average the scores first and discuss only large divergences. Mechanical combination beats a debrief that talks the room into a favourite, by up to 50% in accuracy ( Kuncel et al. 2013 ).
- "Sounds like the right call" tells the candidate what impresses you; they manufacture more of it and your halo grows. Use neutral acknowledgements.
- when you notice a bias reaction. Write it in your notes.
- , in the invitation. Otherwise you select for people who improvise well, which is a personality trait, not the competency.
- and build on them. Re-asking what they already wrote wastes their time and yours.
- : 7 minutes of 30, 10 to 15 of 45.
After the shortlist is final, copy your interview template per candidate and pre-fill: profile summary, scores from earlier rounds, strengths, remaining doubts, and one or two questions aimed at those doubts. This is the step most often skipped for lack of time, and it is where an AI assistant helps most: give it the application data and the template.
Two formats, 15 to 30 minutes each:
- : interviewer, candidate telling a real story, observer scoring the interviewer on neutral questions, probing depth, and who talked most. Rotate.
- Attach your real guide and rubric, have it invent a fictional candidate of similar seniority, run your questions, then ask for a debrief that scores you on the observer checklist and quotes your questions back. Ask it to answer like a polished senior candidate: "we" instead of "I", reflection instead of what they thought at the time, with one or two specifics buried that you only reach by probing the chronology. Voice mode is closer to the real pacing than text.
What a first session typically catches: stacked questions, evaluating out loud, context before the question, and the missed pin-down from step 3. What usually works and is worth keeping: hearing "we" and asking "what was your role in that?"
- Every finalist gets the same core questions. Note any deviation and why.
- Do not ask about health, pregnancy, family plans, religion, origin, age or other protected characteristics, and steer the conversation away if the candidate raises them. Hard requirements (a work permit, a language) belong in the application form, not the interview.
- Interview notes and scores are applicant data. Store them with the application and delete them on the same schedule. Retention rules are in Advertising a vacancy .
Record for your own organization:
- Your opener wording and interview length per round
- Your pairing rule (always two interviewers? across which differences?)
- Which questions you share in advance, and where in the invitation
- Where your interview templates and rubrics live