The nine-box grid, line by line

For roughly fifty years now, the nine-box grid has been the default ritual for talent reviews at large companies. The grid plots employees on two axes — performance (this year) and potential (estimated upside) — into a 3x3 grid of nine cells. The conversation at the top of the calendar talent cycle is essentially: who moves into the top-right cell, who falls out of it, who languishes in the middle, and who needs to leave from the bottom-left. It is a ritual that has survived multiple performance-management reformations, including the period when “performance management” itself was supposedly being abolished.

This is a teardown.

The performance axis

The horizontal axis is supposedly objective — last year’s results, scored against goals. In practice it is heavily contaminated by three things. It captures the rater’s relationship with the employee far more than the work, because managers anchor on their interaction sample rather than on outputs. It captures role-luck: an employee assigned to a struggling product line scores below an employee assigned to a rising one, regardless of contribution. And it captures the manager’s own political position, because managers downgrade reports who threaten their standing and inflate reports who don’t.

It does not, in any reliable way, capture how well an employee actually did the work.

The potential axis

The vertical axis is worse. “Potential” is, by construction, a prediction about a future role the employee has not yet held. There is no reliable signal for this in the data managers actually have. So managers fall back on proxies: communication style, executive presence, similarity to existing leadership, tenure with a powerful manager. These proxies are heavily collinear with demographic background and with conformity to the dominant cultural code of the firm. The “potential” rating measures fit-to-the-current-culture far more than it measures upside.

This is also why nine-box outcomes correlate poorly with whether those employees actually perform in the larger roles they get promoted into. The selection mechanism is not measuring what it claims to measure.

What survives the teardown

Two things justify keeping the grid alive, and neither is what HR will tell you in the meeting.

First, it forces a conversation. Once a year, senior leaders are required to discuss specific employees by name, defend their ratings, and react to disagreement. This conversation is genuinely valuable — independent of the framework — because nothing else in the corporate calendar forces it. If you removed the grid and put nothing in its place, that conversation would stop happening.

Second, it produces a written record of who was seen as promotable when, by whom. That record matters in disputes and in legal review. It is, in this sense, a documentation system disguised as an assessment system.

What does not survive

Everything else. The grid does not predict who will succeed in the next role. It does not reliably surface high-potential talent the system would otherwise miss. It does not, when audited, distribute people in proportion to their actual performance. It does not survive contact with hybrid or matrixed reporting structures. And it almost never includes the people the firm most needs to retain, because the rating process is run by managers who are themselves competing for budget.

If you must run a talent review, run it. But run it for the conversation and the documentation — which is what it is for — and stop pretending the framework underneath it is doing the work.