Why the 9-box grid is dead, and how to remove bias from performance reviews

The 9-box grid (performance on one axis, “potential” on the other) is still a fixture of talent reviews. It is quick, visual and familiar. It is also one of the easiest ways to bring bias back into a process you have worked hard to make fair. This guide explains why, what to use instead, how to remove bias at the manager-to-employee level and at the calibration level, and why it pays to standardise one process across departments.

Why the 9-box grid is dead

1. Potential is an opinion, not evidence. Performance can be tied to goals and deliverables. “Potential” rarely has written criteria, so it defaults to the manager's impression: confidence, visibility, similarity to the people already at the top. The axis with the least evidence ends up deciding who gets developed.
2. It multiplies rater bias. Ratings already say a lot about the rater. In the study of 4,492 managers by Scullen, Mount and Goff (Journal of Applied Psychology, doi:10.1037/0021-9010.85.6.956), as summarised by Buckingham and Goodall in Harvard Business Review, 62% of the variance in ratings could be accounted for by individual raters' peculiarities of perception. A 9-box asks the same rater for two subjective judgements and then combines them.
3. Boxes become labels. Once someone is a “low-potential” box, it follows them, often without them knowing. The grid is a snapshot, but it gets treated as an identity.
4. It rewards the most persuasive manager. In a talent review, people move boxes based on who argues best, not on what is written down. That is calibration theatre, not calibration (see calibration theatre).
5. It isn't comparable across teams. A “high performer” in one department and in another were judged against different, unwritten bars. You can't use the grid to compare, promote or pay fairly across the company.
6. It mixes up potential and readiness. What you usually need to decide is whether someone is ready for the next level. That is answerable with evidence: do they already show the next level's skills?

What to use instead

Removing bias at the manager-to-employee level

• Standard scorecards with observable statements. Managers answer yes/no statements about behaviour (“Writes clear acceptance criteria for every story”), not 1–5 impressions. Because the questions are the same across the company, there is less room for personal standards.
• Rate each area separately, so one strong impression doesn't colour everything (the halo effect).
• Evidence for every rating: a goal result, a deliverable, a piece of feedback. Goals that update from tools such as Jira or a CRM give a record that doesn't depend on memory.
• Collect evidence during the cycle, in 1:1s and notes, to avoid recency bias.
• Add a second perspective: input from other managers who work with the person.
• Brief managers before each cycle with examples of what each level looks like.

Step-by-step version: how to reduce manager rating bias. Check your own process with the review bias checker.

Removing bias at the calibration level

• Start from calculated grades, not a blank grid. The meeting reviews results produced by the scorecards rather than building them from scratch.
• Run the three checks: generous graders, missing or thin evidence, and outliers or sudden jumps.
• Compare rating spreads by manager. If one manager rates half the team at the top, ask for the evidence and the strongest performers.
• Test borderline cases against the bar for the level, not against other people in the room.
• Record every change: original grade, new grade, reason, evidence, who approved it.
• Use distributions as a prompt, not a quota. Research on capped rating curves is mixed: one study found employees nominated for a capped top tier but not awarded it were at least 34% more likely to leave voluntarily (Bond, Cut to the Curve, Management Science).

Full running order: how to run a performance calibration meeting, plus the calibration meeting template. In Revolut People, calibration sits on top of the scorecards, and calibration flags help spot bias before grades are final. See Revolut People performance management.

Why standardise a performance process across departments

• Fairness people can see. If Sales and Product use different scales and cadences, employees reasonably ask whether a grade in one means the same as in the other.
• Comparable promotion and pay decisions. Leadership can only compare grades across departments if they were produced the same way.
• Calibration across functions. Senior leadership review only works when every department arrives with the same kind of evidence.
• Internal mobility. Levels and skills that mean the same thing everywhere make moves between teams easier to judge.
• Less admin. One calendar, one template, one set of manager training.

RecommendationStandardise the structure, tailor the content. Keep the scale, levels, cadence, scorecard format and calibration rules the same everywhere. Let each function leader define the role skills and goals for their own teams.

Why traditional performance management fails

Most failures come back to four problems: unclear standards, goals that don't align with company priorities, evaluation bias in ratings, and the admin burden of pulling a review together. The 9-box is a symptom of the first and third: when there is no written bar, a grid of impressions fills the gap. Find your weakest area with the performance health check, and read why annual reviews fail.

Sources: Scullen, Mount & Goff (2000), Journal of Applied Psychology; Buckingham & Goodall, Reinventing Performance Management, HBR (2015); Bond, Cut to the Curve, Management Science.

FAQ