Forced Ranking Psychology: What Stack Ranking Does to a Team

Forced ranking, also called stack ranking or forced distribution, requires managers to sort employees into fixed proportions: so many at the top, so many in the middle, so many at the bottom. It was among the most influential management ideas of the late twentieth century and, by the 2010s, one of the most widely abandoned.

The idea solves a real problem, which is that managers left to themselves rate nearly everyone highly. But it solves it by changing the question from “how good is this person?” to “is this person better than their colleagues?”. That shift has consequences that psychology would have predicted.

Where did forced ranking come from?

Its best-known advocate was Jack Welch, chief executive of General Electric from 1981 to 2001. GE’s “vitality curve” sorted managers into a top 20%, a “vital” middle 70% and a bottom 10%. The top were rewarded heavily. The bottom were expected to improve quickly or leave. Welch argued that it was kinder to tell people where they stood than to let them drift.

GE’s success in that period made the practice fashionable. By the early 2000s, a substantial share of large American companies were reported to use some version of it. It became known, less flatteringly, as “rank and yank”.

The problem it addressed is genuine. Rating inflation, described in our guide to performance review bias, makes appraisal systems useless for distinguishing performance. A quota is a blunt but effective cure for leniency.

What assumption does the curve rest on?

That performance within any group follows a bell curve, with a few stars, a few failures and a large middle. Two things are wrong with this.

First, statistics. A bell curve describes large random samples. A team of nine is neither large nor random: its members were selected, trained and, if they had been failing, managed. There is no reason for exactly one of Ingrid’s nine to be a poor performer. Applying a distribution designed for thousands to a group of nine guarantees misclassification.

Second, the shape itself may be wrong.

If performance often looks like that, a symmetrical curve with a fixed bottom 10% describes few real teams.

What does forced ranking do to behaviour?

It converts colleagues into competitors. When ratings are relative, one person’s gain is another’s loss, and people respond accordingly.

Less helping. Time spent improving a colleague’s work is time spent lowering your own rank. The engineer who declined to review calculations made a correct calculation of his own.

Hoarding. Information, contacts and credit become assets to protect.

Sabotage, mostly passive. The economist Edward Lazear showed in theoretical work that when rewards depend on relative performance, people have an incentive not only to do well but to make others do worse. In practice this seldom means active undermining. It means withholding help, and not warning a colleague about a mistake.

Avoiding strong teams. Being average in an excellent team is more dangerous than being best in a weak one. Talented people spread themselves out instead of working together.

Hiring down. Existing team members have a reason to prefer recruits who will rank beneath them.

Managing upward. Since rankings are settled in meetings among managers, visibility to other managers matters as much as the work.

The damage falls on exactly the behaviour that modern work depends on. Most valuable output is produced by teams. A system that penalises cooperation taxes the main source of performance. Its effect on willingness to speak up and admit error is described in our guide to psychological safety.

What happened at Microsoft?

Microsoft used stack ranking for years. A 2012 Vanity Fair article by Kurt Eichenwald, based on interviews with current and former employees, reported that many of them named the system as the most destructive process inside the company. Engineers described avoiding working alongside top colleagues, concentrating on being visible to managers who would sit in the ranking meeting, and competing with one another more than with other companies.

In November 2013 Microsoft announced that it was ending the practice: no more curve and no more ratings, with greater emphasis on teamwork and employee growth. The change is often cited as one element of the cultural shift that followed under Satya Nadella, who became chief executive in 2014. Our Microsoft stack ranking case lets you work through the decision.

GE itself moved away from formal forced ranking and, around 2015, from annual ratings, toward more frequent feedback conversations. Adobe, Accenture and others made similar moves. Not every company followed. Some, particularly where leaders believe standards have slipped, continue to use target distributions or have returned to them.

Does forced ranking ever work?

There is a case for it in limited circumstances.

So a one-off or short-run application, in an organisation that has tolerated poor performance for years and has reliable measures, may produce an improvement. The simulation did not model the behavioural costs described above. It assumed ratings measure performance well, which, as the research on rater effects shows, they often do not. And the benefit is front-loaded while the costs accumulate.

Are there legal risks?

Yes. When a fixed proportion of people must be rated low, and those ratings rest on subjective judgment, the pattern of who ends up at the bottom is examined closely. Several large companies using forced ranking faced discrimination lawsuits in the early 2000s, alleging that older workers, women or minorities were over-represented in the lowest band; a number were settled for substantial sums. Where a low ranking leads to dismissal, employment protection law in many countries will require evidence of actual underperformance against a standard, which relative ranking alone does not supply.

How else can you tackle rating inflation?

The underlying problem is real, so abandoning the curve is not enough. The alternatives keep absolute standards and add scrutiny.

Absolute standards. Rate people against a description of the job done well. If all nine meet it, all nine meet it.

Guidelines with explanation. Expected ranges for large groups, with managers free to depart from them if they can show why.

Calibration. Peer review of ratings, run so as to require evidence. It addresses leniency without imposing a quota.

Separate pools for reward. If the bonus budget is limited, say so and distribute it by contribution, without pretending that those who receive less have failed.

Direct management of poor performance. The curve became popular in part because managers avoided difficult conversations. The better remedy is to help them have those conversations, as described in our guide to difficult conversations and the psychology of a PIP.

What does it do to those in the middle?

This is the least discussed effect. In a 20-70-10 system, seven in ten people are told they are in the middle. Most of them believe they are above average, a belief examined in our guide to self-assessment. So the majority of the workforce receives a rating below their self-image every year. Equity theory, described in pay raise psychology, predicts the result: reduced effort or a search for an employer who appreciates them.

The top 20% are not safe from side effects. Told repeatedly that they are stars, some come to expect continuous advancement, and to attribute their success entirely to themselves.

The engineering firm dropped its forced distribution after two cycles. It kept a guideline for the firm as a whole, introduced behavioural standards for each grade and required written evidence for every rating at either end of the scale. The share of “exceeds” ratings settled at around 30%, well below the old 80%. The mechanical design team resumed reviewing one another’s calculations, which, in an engineering firm, was the point.

Frequently asked questions