How to write an interview scorecard

The four parts of a usable scorecard

1. The three or four attributes this round is responsible for assessing.
2. A rating scale with a written definition of what each level looks like.
3. An evidence box per attribute, separate from the rating.
4. One overall recommendation, with the rating decided after the evidence is written.

Choose attributes the round can actually observe

Each interview in a loop should own a small number of attributes, and no attribute should be assessed by every interviewer. If four people all rate “communication”, you get four correlated impressions and no coverage of anything else. Assign deliberately: the coding round owns problem decomposition and code quality, the system design round owns trade-off reasoning, the hiring manager round owns ownership and motivation. Write on the scorecard which attributes this round is *not* responsible for, so interviewers stop rating things they only glimpsed.

Define the rating levels in words

A 1–4 scale with no definitions produces a 3 for almost everyone, because 3 means “fine, I suppose.” Definitions force a decision. For a problem-decomposition attribute, a usable set reads something like: 1 — could not break the problem into parts without substantial help; 2 — broke it down after prompting, missed an important case; 3 — broke it down unprompted, handled the main cases, needed a nudge on edge cases; 4 — broke it down unprompted and identified a constraint the interviewer had not mentioned. Those are observable. “Exceeds expectations” is not.

Use an even-numbered scale

A four-point scale with no midpoint forces interviewers off the fence. On a five-point scale the middle option absorbs every candidate the interviewer is unsure about, which is precisely the population where a decision is most valuable. If you want a no-signal option, make it explicit — “did not assess” — rather than letting the midpoint do that job silently.

Separate evidence from rating, and write it first

• Evidence box first, rating second. Writing the quote or the decision before choosing a number keeps the number tethered to something.
• Evidence means what the candidate said or did, not how it felt. “Chose to denormalise and could explain the write-amplification cost”, not “solid on databases”.
• An attribute with no evidence gets no rating. That is a coverage gap for the loop to fill, not a reason to guess.

One overall recommendation, not an average

Averaging attribute scores produces false precision and hides the thing you most need to see: that a candidate was outstanding on the attribute that matters most and weak on one that does not. Ask for a single recommendation — strong no, no, yes, strong yes — and require a sentence explaining it. The sentence is what gets read in the debrief.

Design mistakes that make scorecards useless

• Too many attributes. Three or four per round. Twelve guarantees the form is completed from memory afterwards.
• Attributes that are really personality traits. “Culture fit” as a rated attribute is where bias enters a structured process wearing a lanyard. Replace it with observable behaviours, such as *gave credit specifically* or *disagreed without escalating*.
• Shared visibility before submission. If interviewers can see each other's scores, you have one opinion recorded four times.
• No deadline. Scorecards completed two days later are memory, not evidence. Require submission before the debrief, ideally within the hour.

Keep the same scorecard across the whole search

Changing the scorecard midway makes earlier candidates incomparable with later ones, which usually surfaces at the worst moment — when someone asks how the third candidate compared to the first. If the scorecard is wrong, fix it, but then be explicit that the scores before and after are not interchangeable. See also how to design an interview loop.

Where Interviewer Mode fits

Disclosure: we make CoPilot Interview, which includes Interviewer Mode for the hiring side and a real-time assistant used by candidates. We have an interest in you using the first, and our view of the second is that interviews should be designed to measure something an assistant cannot fake. Weigh the advice below accordingly.

Frequently asked questions

What should be on an interview scorecard?

Three or four attributes the round is responsible for assessing, a rating scale with written definitions for each level, an evidence box per attribute kept separate from the rating, and one overall recommendation with a sentence explaining it. Anything longer gets completed from memory afterwards.

What rating scale should an interview scorecard use?

A four-point scale with written definitions and no midpoint. Five-point scales collect every uncertain candidate in the middle, which is exactly where a decision is most useful. Add an explicit “did not assess” option rather than letting the midpoint serve that purpose.

Should interviewers see each other's scorecards?

Not before they submit their own. Shared visibility turns four independent assessments into one opinion recorded four times, which removes the main benefit of having a panel. Open them up at the debrief, not before.

How many attributes should each interview assess?

Three or four, and they should differ between rounds. If every interviewer rates communication, you get four correlated impressions and no coverage of anything else. Assign attributes deliberately across the loop and say on the scorecard which ones this round does not own.

Is culture fit a valid scorecard attribute?

Not as written. “Culture fit” is where bias enters an otherwise structured process, because it rewards similarity to the existing team. Replace it with observable behaviours you can cite evidence for, such as giving credit specifically or disagreeing without escalating.