How to run an interview debrief.

Before the meeting: opinions are formed independently.

Google's re:Work hiring research, the finding behind its "Rule of Four," showed that four interviews predicted a hire with about 86% confidence, and each additional interview improved the decision by only about 1%. But that only holds if each interviewer forms their opinion independently. That means no sharing notes, scores, or impressions with each other before the interviews are done. The debrief is the time to come together, compare notes, and share what the team learned. If interviewers talk to each other along the way, the panel stops being several independent reads and becomes one opinion with several signatures.

Independence needs a few things in place:

• Every interviewer had an assignment. Each person covered specific areas, with questions prepared for them, so the panel’s coverage adds up instead of overlapping on first impressions.
• Every candidate was scored on the same scorecard for the role. One set of criteria, one scale, with a written description of what each score level looks like. If you do not have a scorecard yet, start with the complete interview scorecard template and build one before the interviews, not after.
• Scores and notes were submitted before the debrief. Written down, in the system, before anyone hears anyone else’s view.

The scorecard is not there to replace anyone’s judgment. It puts every candidate on the same scale, so the panel can compare what happened rather than compare impressions. Each interviewer still brings their read; the scorecard gives that read something to attach to.

A worked example: a 4 and a 2 on the same competency.

Here is an illustrative debrief for a customer success manager role, with a panel of three. The company and candidate are invented; the mechanics are the point.

The panel: Maya (hiring manager, covering role outcomes and ownership), Tomás (peer CSM, covering the skills: renewal conversations and account planning), Priya (support lead, covering collaboration across teams). All three submitted scores and notes independently, and nobody compared notes beforehand. On the collaboration competency, Maya scored a 4 and Priya scored a 2.

Maya leads the meeting but holds her own read for last. She opens with the role’s key competencies, then the first kickoff question: does anyone have questions for the other interviewers?

Priya does. “Maya, your notes mention the candidate kept teams aligned on a renewal save. Did they say what they actually did after looping support in?” Maya checks her notes: the candidate said “I looped in support early and kept everyone aligned,” and Maya had not pushed further. Priya shares what happened when she did push: she asked twice for a case where the candidate stayed involved after a handoff, and did not get one. Every collaboration example the candidate gave was about routing the problem to someone else.

That takes the meeting to the fourth question before Maya has argued for her 4 at all: has anyone changed their recommendation? Maya has. “You asked the follow-up I didn’t. My 4 was resting on a claim, and your interview tested the claim.” She updates her score, and says so out loud, because a hiring manager visibly changing her mind on new information is what makes it safe for everyone else to.

Next steps: the panel’s picture is now coherent. Strong renewal skills from Tomás’s role-play, real ownership of a number from Maya’s interview, thin evidence on cross-team collaboration. Rather than rejecting on an area the panel only probed once, Maya decides on a final conversation targeting collaboration directly, with the specific probe named in advance. If it happens again after that conversation, the reference calls ask about it too.

Notice that no score decided anything. The 4 and the 2 did their job by pointing at the exact spot where two interviewers learned different things, the discussion resolved which read was better evidenced, and the decision-maker ended up with something better than a feeling to decide on.

Where Yardstick fits.

Everything above works with a shared document and discipline. What makes it routine instead of heroic is having the inputs ready without anyone assembling them. In Yardstick, the interview guide for the role is designed with AI before anyone is scheduled: which areas each interviewer covers, the questions they ask, what to listen for, and a scorecard with written anchors, edited by you before the first interview. Every interviewer scores against the same scorecard for the role, and those ratings roll up into one grid, so the debrief opens with the evidence already lined up instead of scattered across six documents. You can compare candidates on the same evidence side by side, and the built-in chase missing scorecards agent sends configured reminders to interviewers, escalating to the hiring manager at the final step, so the debrief does not start with two scorecards missing.