In 2009, Julia Rozovsky was accepted by the Yale School of Management. On campus she was put in a study group the school had set up to build tight bonds. There were five of them. Everyone was smart and curious. They had gone to similar colleges and worked at similar firms, and Rozovsky hoped all that would make it easy.
It didn’t. “I always felt like I had to prove myself,” she told the journalist Charles Duhigg. In meetings, people sometimes jockeyed to lead or picked at each other’s ideas. “People would try to show authority by speaking louder or talking over each other.”
So she found another group: a team for case competitions, where students pitch solutions to real business problems for trophies and cash. Its members had been an Army officer, a think-tank researcher, the director of a health-education nonprofit and a consultant to a refugee program. They emailed each other dumb jokes and usually spent the first ten minutes of a meeting chatting. “We all felt like we could say anything to each other,” she said. They won one competition with a plan for a tiny gym on campus.
The study group broke up in her second semester. The case team stayed together for her two years at Yale. What puzzled her was that the study group’s members were friendly and warm one on one. Only together did things turn fraught.
“I couldn’t figure out why things had turned out so different.”
The roster that didn’t matter
Google hired her after Yale and put her on Project Aristotle, its study of what makes teams work, named for the line about the whole being greater than the sum of its parts. Starting in 2012, the researchers looked at 180 teams, 115 in engineering and 65 in sales, and tested both who was on them and how they worked. The other case on this page tells how the roster kept failing to explain anything. Google’s guide states the result: what mattered “was less about who is on the team, and more about how the team worked together.”
At Google, how a team worked together mattered more than who was on it.
The researchers named five dynamics, in order of importance: psychological safety, dependability, structure and clarity, meaning, and impact. Psychological safety came first. On those teams, in Google’s words, people “feel confident that no one on the team will embarrass or punish anyone else for admitting a mistake, asking a question, or offering a new idea.” The survey item Google later gave teams to check it reads: “If I make a mistake on our team, it is not held against me.”
The idea wasn’t Google’s. Amy Edmondson of Harvard had defined psychological safety in 1999, and Google’s guide credits her. In the interviews, Googlers kept talking about how their teams felt. “That made a lot of sense to me,” Rozovsky told Duhigg, “maybe because of my experiences at Yale.”
Same people, different rules
What the researchers kept running into were norms: what Duhigg describes as the traditions, behavioral standards and unwritten rules of a group. People may act one way alone, he writes, but when they gather, the group’s norms typically override their individual leanings.
Norms are like a city’s driving habits. The same driver waits his turn in one town and leans on the horn in another. In Duhigg’s reading, Rozovsky’s study group drained her because of its norms, the fights over leadership and the habit of critique, which kept her on guard. Her case team’s norms ran the other way: enthusiasm for each other’s ideas, and joking around.
Google’s guide gives a small example of what that risk looks like. Asking “what’s the goal of this project?” can make you sound out of the loop, so it may feel easier to carry on without asking. On a team where a mistake is held against you, staying quiet looks even more sensible.
One company’s data
This is one company’s study of its own teams. Effectiveness meant ratings from executives, team leads and team members, plus sales against quota. Google published the work as a how-to guide, which describes the method in outline but doesn’t include the data or the models.
The guide is careful about its own limits. Team size, seniority, workload and even the individual performance of team members did not show up as significant at Google, but that “doesn’t mean they’re not important elsewhere.” Other research, it notes, has found that smaller teams tend to do better. And whatever makes teams work in your organization, the guide adds, “it may be different from what the Google researchers found.”
Duhigg pointed to a paradox: all of Google’s number crunching had led it to “the same conclusions that good managers have always known.”
Reading the score
Say you lead a team of six whose numbers are fine, and you hand out Google’s five statements anonymously. Four come back strong. One comes back low: “If I make a mistake on our team, it is not held against me.”
The easy move is to explain it away, or to take it personally. Google’s process went the other way: the team lead got the anonymized scores to share with the team, often with a facilitator in the room. If you put the low number on the screen and ask what would raise it, the first minutes may be quiet. That quiet is part of the answer.
Midway through Project Aristotle, Rozovsky made a mistake in a meeting and sent a note afterward explaining how she’d fix it. A teammate replied with one word: “Ouch.”
“It was like a punch to the gut,” she said. She wrote back: “Nothing like a good ‘Ouch!’ to destroy psych safety in the morning.”
He answered: “Just testing your resilience.”
“With one 30-second interaction,” she said, “we defused the tension.”