Every six months at Microsoft, the managers of one group met for a few days in a single conference room, as many as 30 of them. The blinds were drawn and the doors closed. A grid of possible rankings went up on a whiteboard or a poster board, and everyone broke out the Post-it notes. Each note carried the name of someone on their team. One by one, the managers placed their people into the boxes.
The numbers usually didn’t work on the first round. Every unit had to declare a set share of its people top performers, then good, average, below average and poor. “There are some pretty impassioned debates and the Post-it notes end up being shuffled around for days so that we can meet the bell curve,” one manager told Kurt Eichenwald for Vanity Fair in 2012. He added: “I myself have had to give rankings to people that they didn’t deserve because of this forced curve.”
Microsoft hadn’t invented the curve. As Bloomberg later put it, the practice of pitting employees against one another “to reward the excellent and weed out the weak” gained widespread popularity in the 1980s, after Jack Welch brought it to General Electric. Welch had a case for it.
The case for the curve
In GE’s letter to share owners for 2000, from Jack Welch and his top executives: “In every evaluation and reward system, we break our population down into three categories: the top 20%, the high-performance middle 70% and the bottom 10%.”
The top 20% “must be loved, nurtured and rewarded in the soul and wallet.” The middle 70% were to be energized to move up. The bottom 10%, the letter said, tend to stay there, and a company that bets on its people “must remove that lower 10%, and keep removing it every year.” Keeping them was “false kindness,” a form of cruelty, because a new leader would eventually take them out, sometimes midway through a career.
The appeal, in the letter’s words, is “rigorous discipline in evaluating, and total candor in dealing with, everyone.” A manager who would rather call everyone fine has to say who is weakest. In 2005, Steven Scullen, Paul Bergey and Lynda Aiman-Smith ran a computer simulation of a system that fires a fixed share every year. It suggested such a system could lead to “noticeable improvement” in the workforce’s potential, mostly in the first several years, with more gain from firing more people and from fewer leaving on their own. It modeled hiring and firing; it measured no real company. Its inputs, as the abstract lists them, include how reliable ratings are and how well hiring works, not how people behave once they know they’re ranked against each other.
What the curve does to the room
At Microsoft, Eichenwald found that behavior. Even meeting every goal didn’t guarantee a good rating, because a colleague might beat theirs by more. So people tried to do good work and also, he wrote, to make sure their colleagues didn’t. “One of the most valuable things I learned,” one engineer said, “was to give the appearance of being courteous while withholding just enough information from colleagues to ensure they didn’t get ahead of me on the rankings.” Executives told him many stars avoided teams full of strong developers, fearing for their ranking.
A forced curve makes managers name their weakest. It can also turn teammates into rivals.
Grading on a curve can sort a lecture hall of strangers. On a team, it can make a colleague’s good week a threat to your review.
Lab experiments by Linda Loberg, Stephan Nüesch and Johann Nils Foege, published in 2021, found the same split in miniature. Under a forced distribution, people working alone on a card-sequencing task went faster. People working in teams went slower and shared less knowledge, and the system was seen as unfair when people had to collaborate.
There is a second problem: the curve sorts whatever ratings it is given. Deloitte cited a 2000 study by Michael Mount, Steven Scullen and Maynard Goff, in which 4,492 managers were each rated by two bosses, two peers and two subordinates. Of the variance in ratings, 62% came from the raters’ own peculiarities of perception, and 21% from actual performance. A forced shape makes ratings spread out. It cannot make them measure better.
After the curve
On November 12, 2013, Microsoft’s head of human resources, Lisa Brummel, sent the company a memo. “No more curve,” it said: there would no longer be “a pre-determined targeted distribution,” and managers would hand out rewards within their budgets as they judged best. Presented to managers that morning, GeekWire reported, it ended in a standing ovation.
Deloitte’s redesign, described in the Harvard Business Review in 2015 by Marcus Buckingham and Ashley Goodall, went at the ratings themselves. It had counted close to 2 million hours a year spent on forms, meetings and ratings. Now, at the end of each project, team leaders respond to four statements about what they themselves would do with each person, such as “I would always want him or her on my team.” Deloitte still wanted “differentiated performance data” and pay that recognized performance. It changed the question.
The sixth person
Say your company asks every manager to rate at least one person per team “needs improvement,” and you run a team of six who all shipped what they promised. Somebody has to go in the box.
You could pick whoever had the quietest year. The team may notice, and next quarter some might help each other a little less. Or you could refuse the quota and rate everyone fine, and then nobody on your team hears that the one shaky project was shaky. Welch’s letter is a warning about the second road. Eichenwald’s reporting is a warning about the first.
Either way, the form goes back with a name in the box, or with your reasons for leaving it empty.