How to use the 9-box grid: running a talent review step by step
A practical guide from the 9Box team · Last updated 22 August 2026
Placing people on a grid takes twenty minutes. Running a review that changes what anyone does afterwards takes preparation. This is the process.
New to the method? Start with what the 9-box grid is and what each box means. Have one specific question? The 9-box grid and talent review FAQ answers twelve of them.
What you need before you start
- A defined population that can sensibly be compared — one function, one level band, or one leadership team's reports.
- Written definitions of low, medium and high on both axes, circulated before anyone is rated.
- Initial placements from each manager, with one piece of evidence each.
- A grid built and ready to project, so the meeting starts from real positions rather than a blank page.
- Ninety minutes, a chair who can break ties, and someone capturing actions.
Before the session
1. Decide the population and be able to say why
A grid should cover a group of people who can sensibly be compared: one function, one level band, or one leadership team's direct reports. Mixing a graduate cohort with a senior leadership team produces a meaningless grid, because "high potential" means something different in each. If you need both, run two grids.
Also decide the purpose, out loud, before you start. Succession planning, development budget allocation and identifying retention risk are three different jobs and they lead to three different conversations. A grid built for one and used for another is where most of the damage gets done.
2. Define both axes in writing, and circulate the definitions first
This is the step teams skip, and skipping it is the single biggest cause of a wasted session. Write down what low, medium and high mean on each axis, in language specific to your organisation. Give an example of what high potential looks like and — more usefully — what it is often confused with. If nine managers hold nine private definitions of "high potential", the calibration meeting becomes a two-hour argument about vocabulary.
3. Have managers place their own people first, with evidence
Ask each manager to place their team independently, and to bring one concrete piece of evidence per placement. "She led the migration and it landed two weeks early" is evidence. "He's very strong" is not. Pre-placement means the session starts from real positions rather than a blank grid, and the act of writing down evidence catches a good proportion of gut-feel ratings before they reach the room.
4. Build the grid before the meeting, not during it
Get the roster in, get the initial placements on, and have it ready to project. In 9Box that means creating a grid named for the population and cycle, importing your people from CSV or adding them individually, and dragging the pre-placements into position. People who haven't been placed yet stay in the "To place" list until the room decides.
Who should be in the room
Keep it small — under ten people. Calibration is a discussion, and discussions stop working past that size.
- The managers of the population, all at the same level. They own the placements and do the talking.
- Their common leader, chairing. Someone has to break ties, and it has to be someone whose decision everyone accepts.
- An HR partner, facilitating. This is the most important seat in the room. The HRBP's job is not to rate anyone — it is to hold the definitions steady, ask "what's the evidence?", surface patterns across the grid, and be the person willing to say "we've placed every woman in the medium potential row, let's look at that again."
- Someone taking notes, capturing development actions as they are agreed rather than reconstructing them afterwards.
Do not invite people to sessions where their own team is discussed by their manager. Do not invite a wider audience "for visibility" — attendance changes what people are willing to say, and candour is the whole point.
A ninety-minute agenda
Budget roughly two to three minutes per person, plus twenty minutes at the top and twenty at the end. This is a thirty-person grid; scale the middle block and leave the bookends alone. If the population will not fit, split it and run two sessions rather than shortening the discussion.
| Time | What happens | Who leads it |
|---|---|---|
| 0–10 min | Purpose of this grid, and the definitions restated in full. | HR partner |
| 10–20 min | Ground rules: evidence required, no quotas, challenge is expected, nothing leaves the room unagreed. | Chair |
| 20–45 min | The extremes — proposed Stars and proposed Underperformers, while attention is highest. | Each manager in turn |
| 45–70 min | Everyone else, moving placements as the discussion lands. | Each manager in turn |
| 70–80 min | Read the shape of the finished grid and ask the pattern questions out loud. | HR partner |
| 80–90 min | Actions, owners and dates. Export the grid before anyone leaves. | Chair and note-taker |
The two ways this goes wrong are both about pacing: spending forty minutes on the first three names and then racing the rest, or booking an hour for eighty people and calling the result calibrated.
Running the calibration
5. Restate the definitions, every time
Two minutes at the top. Nobody remembers them from last cycle, and half the room did not read the email.
6. Work the extremes first, then the middle
Start with the proposed Stars and the proposed Underperformers. These carry the biggest decisions and provoke the most disagreement, and you want them discussed while the room is fresh and attentive rather than in the last ten minutes. Then work through the rest.
7. Make every placement earn its position out loud
The manager states the placement and the evidence. The room can challenge. The chair decides. Placements move as a result — if nothing moves during a calibration session, the session did not calibrate anything and you have just held a meeting.
8. When two people are hard to separate, compare them directly
Abstract debate about whether someone is "medium" or "high" goes nowhere. Naming two specific people and asking "is Priya genuinely operating above Marco here?" resolves it in a fraction of the time. Comparison against a real person is far easier than comparison against a written definition.
9. Look at the shape of the finished grid, not just the placements
Before you close, step back and read the distribution. An empty high-potential row in a large team is a succession problem. A crowded top-right corner usually means ratings inflation rather than an exceptional team. And check the pattern questions: does "high potential" cluster by demographic, by who is loudest, by who happens to report to the most persuasive manager in the room?
Common pitfalls
- Rating potential on performance. The most common single error. A strong performer gets nudged up the potential axis because it feels harsh not to. Both axes then say the same thing and the grid collapses to a ranked list.
- Recency. Someone's last six weeks stand in for their year. Ask for evidence from the first half of the period specifically.
- Forcing a distribution. Quotas turn assessment into trading. Let the shape be what it is and interpret it.
- The manager who wins every argument. Persuasiveness is not evidence. This is the chair's problem to manage, actively.
- Confusing "not promotable" with "not valuable". Trusted Professional and Effective are healthy boxes describing people the organisation runs on. If your room treats the bottom row as a failure state, your definitions are wrong.
- No actions. A grid with no owner and no date attached to anything is a two-hour meeting that produced a picture.
- Never revisiting it. Placements made twelve months ago and never revisited become permanent facts about people.
What to do with the output
10. Turn every placement into something with an owner and a date
Development action, stretch assignment, coaching, a conversation that has been avoided, a role change. Written next to the person, owned by a named manager, with a date. This is what makes the difference between a talent review and an exercise.
11. Extract the three lists that matter
Retention risk (usually your Stars and Growth Employees), succession gaps (senior roles with no credible successor), and the difficult conversations now overdue. Each goes to a different person and moves at a different speed.
12. Decide deliberately what employees are told
There is no universally right answer, but there is a wrong one: telling people nothing while assuming they will never find out. Most organisations share the development actions and the conversation behind them without ever quoting a box name — the actions are what's useful to the employee, and the label is what causes the damage. Whatever you choose, choose it explicitly and apply it consistently, because inconsistency is how the labels leak.
13. Export it, store it properly, and treat it as personal data
Export the grid and keep it with your talent records. A 9-box grid contains identifiable people alongside written performance judgements — it carries the same handling obligations as any other performance record, including retention limits and the possibility of a subject access request. Do not circulate it more widely than the decisions require. In 9Box, grids are private to the account that made them, and the PNG or PDF export is a flat file — no drag targets, so a name cannot be nudged into a different box on the way round.
14. Run it again, and compare
The grid gets genuinely useful on the second cycle, when you can see who moved and ask why. Keep each cycle as a separate grid with its own review date so the comparison is possible.
Set your grid up before the meeting
Import the roster, pre-place your people, and project it live while the room calibrates. Private to your account. $49 a month or $490 a year — and while we finish building billing, signing up gives you the whole thing.
Build your first grid