To stack rank is to put people in order against one another and to decide in advance how many may sit in each band. Employers usually arrive at it for a good reason, which is that ratings coming out of their cycle are not comparable between managers and something has to be done about it. Whether ordering people is the right fix depends on four things that are worth answering on paper before the cycle opens, because all four are difficult to change once the first round has run.
At what level does the order apply
Company, function or team changes everything about what the design means. A distribution enforced across a function of two hundred is a statistical statement; the same one applied inside every team of five is a rule that one person in five is bottom, which is not the same claim at all. Answer this first, because the rest of the design follows from it and nothing else can rescue the wrong answer.
What is being ordered
Results, behaviors, potential or some blend of them. Ordering on results rewards whoever had the better territory; ordering on behaviors is more defensible and much harder to evidence. A blend needs its weights written down before anyone rates, otherwise each manager chooses their own and the order is not comparable, which was the problem you were trying to solve.
What follows from the position
An order that changes nothing is an insult with no purpose. An order that decides pay needs an evidence standard; one that decides who leaves needs a much higher one, and in most jurisdictions it needs a documented process that stands on its own regardless of where the person sat in the order. Decide the consequence before the mechanism, because the consequence sets the standard of proof.
How often the order is redrawn
Annually is the usual answer and it has a quiet cost: a person's position is fixed for a year by one comparison. More frequent ordering is more work and more accurate, less frequent is neither. Whatever you pick, the ordering conversation has to happen at a point where the evidence is fresh, which is rarely the week the budget is due.
Questions people ask about stack rank
Does anyone still stack rank?
Yes, though usually in a softer form: a suggested distribution used to prompt calibration rather than a quota that must be met. That version keeps most of the comparability and loses most of the damage.
Is stack ranking legal?
The ranking itself is not the risk; what is done with it is. Any employment decision made from an order has to stand on the evidence behind the position, and that is a question for your own employment counsel rather than for a software page.
What if managers refuse to differentiate?
That is the real problem underneath most curve proposals, and a curve only hides it. Calibration with the evidence in the room, and a rating scale everybody has practised on real examples, address it directly.