A 9-box grid is easy to draw and easy to waste. Most teams fill one in, present it, file it, and wonder why nothing changed. The grid was never the deliverable. The meeting is, and specifically the argument the meeting forces. Here is how to run one that produces decisions instead of a nicer-looking spreadsheet.
What the meeting is actually for
The 9-box plots people on performance against potential, but the placements are not the output. The output is calibration: a room of leaders arriving at a shared, defensible standard for what "high potential" means here, applied evenly across teams.
You can see this by imagining the two failure modes. A grid where everyone agrees on every placement taught you nothing; either the room shares blind spots or nobody pushed. A grid filled in by one person alone is just their opinion in a 3x3 layout. The value lives in the disagreements, worked through with evidence, which means the meeting has to be built to surface them rather than smooth them over.
If you need a grid to run the session, the free 9-box builder on this site does the job without a login. Bring the placements; spend the meeting on what they reveal.
Before the meeting: placements come in cold
The single biggest determinant of a good review is preparation, and the rule is simple: everyone submits their placements before the meeting, independently.
Cold, independent placements matter because live placement collapses into groupthink. The most senior or most confident voice anchors the room, and everyone else drifts toward it. When each leader has already committed a placement in writing, the disagreements are real and on the record, and the facilitator can build the agenda around them.
The prep checklist:
- Circulate the criteria for both axes in advance. What counts as evidence of performance? Of potential? If people place against different definitions, you will spend the meeting discovering that instead of calibrating.
- Each leader places their own people and submits before the session.
- The facilitator collects the placements and marks the divergences: the same person placed in different cells by different raters, or clusters that look suspiciously generous or harsh.
- Pull last cycle's decisions. Which development actions completed? Who moved, and does the movement match what was predicted?
Walk in with the disagreements already identified, and the meeting has an agenda that writes itself.
The agenda, calibration first
A half-day shape that works:
- Ten minutes on standards, not people. Read the axis criteria out loud before anyone's name comes up. It feels remedial and it is the best defense against the definitions drifting mid-meeting.
- Ninety minutes on the disagreements. Work only the placements where raters diverged. This is the meeting. Each side names the evidence behind its placement; the room negotiates the standard, then the placement follows.
- Thirty minutes on the whole board. Step back. Which cells are crowded, which are empty? Is the top-right suspiciously full? Are there roles with nobody credible anywhere near ready?
- Forty-five minutes on decisions. Development assignments, retention moves, stretch opportunities. Every decision leaves with an owner and a date.
- Fifteen minutes reading the log back. Before anyone leaves, read the decisions aloud. Silence in the room now is commitment later.
Notice the shape: the agreements get almost no time, the disagreements get the most, and the meeting ends in owned actions. That ordering is the difference between calibration and a status update.
Facilitating the disagreements
The disagreement segment is where facilitation earns its keep, and the job is to keep the debate about evidence rather than advocacy.
A few moves that help:
- Ask for the situation, not the adjective. "Strong leader" is an impression. "Ran the recall response and kept the team shipping through it" is evidence the room can weigh. When someone offers an adjective, ask what specifically they saw.
- Name the anchor. When the room is drifting toward the senior person's placement, say so. "We are all moving toward Dana's read; does the evidence actually support that, or are we deferring?"
- Separate performance from potential out loud. The most common muddle is a top performer waved into the high-potential cell because they are good at their current job. Ask the two questions separately: how are they doing now, and what evidence says they could do something bigger?
- Let the standard win, not the loudest voice. The point of writing down criteria is that you can appeal to them. When two leaders are stuck, return to the definition rather than splitting the difference.
The traps that wreck a 9-box review
Four recurring failure modes, worth naming so the room can catch them:
- Potential that means polish. Without explicit criteria, "potential" quietly becomes "presents well to executives" or "reminds me of myself." Both track visibility, not capacity. The corrective is defining, before the meeting, what evidence of potential actually looks like.
- The top-right as a promotion promise. A placement in the future-leader cell is a development signal, not a commitment to promote. Treated as the latter, it sets people up to be pushed into roles they are not ready for, and sets the org up to be surprised when it goes badly.
- Placements that never change. A grid that looks identical year over year is not stable, it is unexamined. People grow, plateau, and shift. If nobody moves, nobody is really being reassessed.
- The meeting as theater. If the placements were decided beforehand by two people and the meeting exists to ratify them, everyone knows, and the calibration value is gone. The disagreements have to be allowed to actually change outcomes.
Leaving with decisions, not impressions
The test of a 9-box review is what exists when it ends. A review that produced a filled grid and a good feeling has to rediscover the same conclusions next cycle. A review that produced a decision log, with owners and dates against development, retention, and readiness calls, actually moved the organization.
So end every session the same way: read the log back, confirm each owner accepts their action, set the date for the next review, and record the reasoning behind the contested calls, not just the outcomes. That last part is what lets the next cycle start from decisions instead of from a blank grid.
The grid is a prop. The calibrated judgment and the owned actions are the product, and a meeting run this way is how you get them. When the reviews start spending their first half hour reconstructing who placed whom and why, that is the moment a shared spreadsheet stops being enough and 9-box grid software that keeps the history starts paying for itself.