Stack ranking compares people with each other and often forces a fixed share into each rating bucket, whatever the team actually delivered. Calibration compares draft ratings with shared standards and evidence, and changes a grade only when the evidence does not support it. If your CEO has asked you to “rank the team” or “cut the bottom 10%”, the useful move is to keep the bar and drop the quota.
This page is for People leaders who need to push back on, or reshape, a ranking request. It draws the line between stack ranking, forced distribution and calibration, then sets out a quota-free way to raise the bar. Use it with the performance calibration guide, the calibration meeting guide, and the talent density calculator. Disclosure: Ash Taherian is Product Sales Executive at Revolut People, which sells performance management software. Treat product mentions as coming from an interested party.
What is stack ranking?
Stack ranking is a way of ordering employees against one another, from highest to lowest, rather than grading each person against a published standard for the role and level. In many companies it is paired with a forced distribution: a rule that a fixed share of people must fall into each rating bucket. When the people at the bottom of that order are fired or pushed out, critics often call the combination rank-and-yank.
The two ideas are easy to mix up, but they are not identical. You can put a team in rank order without imposing a quota. You can also force a share of ratings into each bucket without producing a full 1-to-N list. What they share is a relative rule: someone has to sit at the bottom, even if the evidence says the whole team met the bar.
That is a different job from performance calibration. Calibration is the step where draft ratings are checked against shared standards and evidence, so that a rating means the same thing across managers. A distribution can appear in that room as a prompt. It should not set the outcome in advance.
Why companies dropped it
The best-documented corporate exit is Microsoft’s. In November 2013, Microsoft announced that it had ended its stack-ranking review and compensation system, and that managers were no longer required to grade employees against one another on a scale of 1 to 5, as reported by SHRM (Stephen Miller, 20 November 2013, as checked 13 November 2026). SHRM wrote that the most contentious part of the approach was its targeted, or forced, distribution, which meant managers had to designate a small percentage of their direct reports as underperformers. The system had been in place for Microsoft’s 2011 performance reviews.
Lisa Brummel, then Microsoft’s head of human resources, wrote in a 12 November 2013 email to employees, quoted by SHRM as reported by the Seattle Times: “No more curve. We will continue to invest in a generous rewards budget, but there will no longer be a pre-determined targeted distribution. Managers and leaders will have flexibility to allocate rewards in the manner that best reflects the performance of their teams and individuals, as long as they stay within their compensation budget.”
That was one company, not a census. SHRM also recorded that the practice still had defenders, including Jack Welch, who argued in a Wall Street Journal opinion piece that “rank and yank” was a pejorative for a practice he called differentiation. Commonly recycled GE percentage splits are not quoted here, because they often appear without a primary source in front of the reader.
The retention cost of a hard cutoff is easier to source. Bond, in Cut to the Curve, Management Science (volume 72, issue 4, DOI: 10.1287/mnsc.2023.01204, as checked 13 November 2026), studied a multinational firm that enforced a strict recognition cutoff at the top of a ranking. Employees downgraded from the top ranks because of that cutoff were at least 34% more likely to leave voluntarily, even when the organisation tried to manage fairness concerns and paid them larger bonuses than their top-ranked peers. That is the same finding cited on the 9-box grid guide.
Leniency is still a real problem. Managers sometimes go easy on weak performance for rational reasons, including fear of pushback — see the playbook’s note on HBR and lenient reviews. A quota is a blunt fix: it can punish a strong team, and it can push out people you would rather keep. McKinsey, in Performance management that puts people first (15 May 2024, as checked 13 November 2026), describes a shift among companies from forced distribution to distribution guidance — a prompt, not a curve the room must hit.
Stack ranking vs forced distribution vs calibration
Use this table when you need to show a leadership team the line between a ranking request and an evidence check.
| Question | Stack ranking | Forced distribution | Calibration |
|---|---|---|---|
| What it does | Orders employees against each other, from highest to lowest. | Requires a fixed share of people in each rating bucket, whatever the team delivered. | Compares draft ratings with shared standards and evidence, so a rating means the same across managers. |
| The decision rule | Relative place in the order, often with a cutoff at the top or bottom. | The quota. Someone must fill the bottom (or top) share. | Change a grade only when the evidence does not support it. |
| What a distribution is for | Often the basis of the ranking itself. | The rule the room must hit. | A flag that triggers a second look, never a quota. |
| Typical pressure | “Rank the team.” | “Cut the bottom 10%.” or “No more than X% at the top.” | “Does this rating stand up against the standard and the evidence?” |
| Main risk | A strong team still has a bottom, and people may compete for a safer place in the order. | Under-recognition and avoidable leavers. Bond found a 34% higher voluntary exit risk after a strict top rank cutoff, in one firm. | Calibration theatre: the meeting happens, but grades never move and thin evidence is left in place. |
| When to use it | Rarely as a standalone system. A rank order can be a temporary sorting aid, not the grade. | Do not use it as a quota. A planned share can be guidance only. | The default alternative, after draft ratings are in and before results are published. |
What to do instead: calibrate without quotas
If the request is really about higher standards, you can raise the bar without a curve. The playbook’s alternative is evidence-based ratings and calibration that flags problems without forcing a share into each bucket.
- Publish the standard. Agree what good looks like by role and level, and put the rating scale in writing before anyone drafts a grade. That is the job of the performance management framework, not of a ranking algorithm.
- Link goals to the review. People should be graded against the work they were asked to do, with a record of what changed during the cycle.
- Collect evidenced draft ratings. Managers write to the standard, with checkable notes, not a rank order. Self-reviews and peer input can add evidence; they should not become a popularity contest.
- Calibrate before any conversation. Run the three checks in the calibration meeting guide: generous graders, thin evidence, and outliers. Look at draft ratings side by side by team, role and level. The performance review bias guide and the bias checker help you spot patterns worth questioning.
- Change a grade only when the evidence does not support it. Record the reason. If the evidence is thin, send it back to the manager. Do not move a person down, or up, to make a percentage look right.
- Then hold the conversation and act. Publish results only after calibration. Use the rating, not a forced rank, to inform development, pay or promotion decisions you have already said the cycle will feed. A merit matrix is the usual way to turn that calibrated rating, and position in range, into a recommended merit increase.
In Revolut People, calibration is a structured process after the review period, in which managers and leaders review and adjust performance ratings for consistency. Admins can choose one of four recommended-grade logics, including Ranking Grade. The product documentation for Calibration Recommended Grade: Ranking Grade (as checked 13 November 2026) says: “Ranking Grade works by calculating individual scores, ranking employees from highest to lowest, and then mapping them into grade brackets according to a predefined distribution.” That is a recommended grade, not a final one. Calibrators assign the grade. If you use a distribution-based recommendation, treat it as a starting point for the three checks, not as a curve the room must hit. The same documentation describes system-generated calibration flags — concerns, strengths and insights, populated overnight — which surface extra information, such as line-manager leniency or a grade not supported by deliverables. A flag does not decide the grade. Those points are from Revolut People’s public product documentation, as checked 13 November 2026. This page does not claim that the product applies a quota.
If your calibration sessions look thorough but grades never move, start with calibration theatre and the meeting guide, not with a tighter curve.
When a distribution check is useful
A draft distribution is useful when it makes you look again. It is harmful when it tells you what the answer must be.
Use the spread of ratings as a flag in these cases:
- One manager’s team is almost all at the top, or almost all in the middle. That may be right — a carefully hired team can be strong — but it needs evidence against the written standard, not a defence of the manager’s reputation.
- The top or bottom of the scale is empty across several teams, which may be central tendency or leniency rather than a true picture of the work.
- Two people with similar evidence have landed far apart, or a rating is an outlier for the role and level.
Then do the second look: read the evidence, ask for what is missing, and change the grade only if it does not stand up. Do not add a person to the bottom group to hit a share, and do not cap the number of people who can be rated above the bar if the evidence supports them.
The playbook publishes its own guideline weights, not an industry benchmark, and they are a starting point for questions rather than a rule. See the framework evidence and caveats. McKinsey’s “distribution guidance”, as checked 13 November 2026, is the same idea: a prompt, not a forced curve.
Fictional example. A product team of eight has been hired carefully over two years. Draft ratings, with evidence notes, put six people at “performing” or “exceeding” and two at “developing”, with a plan already in place for those two. A rule that 10% of every team must be below the bar would require someone who met the standard to be marked down. In a quota-free calibration, the room keeps the six ratings that stand up and spends its time on the two people who need a clearer plan, not on inventing a third.
Raising talent density without rank-and-yank
CEOs who ask for a ranking are often asking for higher talent density: a higher share of people who meet or exceed the bar. That is a fair ambition. Rank-and-yank is a poor way to get there.
The playbook measures talent density as high performers divided by headcount. Hiring below the bar dilutes it; a high performer leaving cuts it; evidenced exits below the bar can raise it. The walk-through is in how to measure talent density, and you can run the numbers in the talent density calculator.
A forced cutoff can work against that goal. Bond’s study, as checked 13 November 2026, is a warning that people held just short of a top rank by a strict curve may leave even when you pay them well. You also risk marking down a strong team to satisfy a percentage. Raise the bar in hiring, be honest in ratings, calibrate without a quota, and act on evidenced underperformance.
If you are holding that conversation now, book a 15-minute review and bring the ranking request or the draft distribution. You walk Ash through it and leave with a way to keep the bar without a forced curve, whichever vendor you pick.
FAQ
What is stack ranking?
Stack ranking is a way of ordering employees against one another, from highest to lowest, rather than grading each person against a published standard for the role and level. It is often paired with a forced distribution, a rule that a fixed share of people must fall into each rating bucket.
Why did companies stop using it?
Not every company did, but some well-known users dropped the forced curve. Microsoft ended its stack-ranking system in November 2013, as reported by SHRM (as checked 13 November 2026), and no longer required managers to grade people against one another on a 1 to 5 scale with a pre-determined targeted distribution. Research also points to a retention cost: Bond (Management Science, as checked 13 November 2026) found that employees downgraded from the top ranks because of a strict cutoff were at least 34% more likely to leave voluntarily.
What is the difference between stack ranking and calibration?
Stack ranking orders employees against each other and often forces a fixed share into each bucket, whatever the team actually delivered. Calibration compares draft ratings against shared standards and evidence, and changes a grade only when the evidence doesn't support it. A distribution can be a useful flag in calibration, but it should trigger a second look, never set a quota.
What should companies use instead of stack ranking?
Clear standards by role and level, goals linked to reviews, evidence-based ratings, and calibration that flags generous graders, thin evidence and outliers without forcing a curve.
Is a distribution check the same as a forced curve?
No. A distribution check is a prompt: if a team is bunched at the top or in the middle, or if two similar records sit far apart, you look at the evidence again. A forced curve is a quota: a fixed share must land in each bucket, even when the evidence does not support it.
Does a recommended grade that uses a distribution mean we have to hit a quota?
No. In Revolut People, Ranking Grade is one optional method for recommending a grade: it ranks a composite score and maps it to a predefined distribution. Calibrators assign the final grade. Treat that recommendation as a starting point for an evidence check, not as a curve the room must hit. Those points are from Revolut People’s public documentation, as checked 13 November 2026.
Related guides
Performance calibration: what it is and what calibration software should do · how to run a performance calibration meeting · performance management framework · employee performance rating scales explained · 9-box grid: how it works, a free template, and when to replace it · performance review bias · calibration theatre · how to measure talent density · talent density calculator · HBR and lenient reviews · year-end performance review run sheet · free tools · People Performance Playbook home