What Google learned about great teams
They spent two years and studied 180 teams to find the formula. The answer had nothing to do with who was on the team.
Almost everything that matters at work gets done by a team.
Teams are where ideas get built and tested. They are where most people actually experience their job. They are also where things quietly fall apart: mismatched skills, fuzzy goals, friction nobody wants to name out loud.
So in 2012 Google decided to find out what separates the teams that fly from the ones that grind. They had form here. Project Oxygen had already cracked what makes a great manager. This time the People Analytics team went after teams, and gave it a codename: Project Aristotle, a nod to his line about the whole being greater than the sum of its parts.
Researcher Julia Rozovsky led it. The expectation going in was fairly obvious. Put your best people together, give them an experienced manager and enough resources, and you get a great team.
That turned out to be wrong.
A group of people is not the same as a team
Before you can study something you have to draw a line around it. Google split two things that often get treated as one.
Held together by the org chart
Low interdependence. They meet now and then to share information and hear updates. People can do their jobs perfectly well without each other.
Held together by the work
Highly interdependent. They plan, solve problems, decide and review progress together on a shared piece of work. Members genuinely need each other to get anything done.
Reporting lines only tell part of the story, so the researchers let teams define themselves. The ones they ended up studying ranged from three people to fifty, with a median of nine.
The obvious metrics all fell apart
The first instinct was to count things. Lines of code written. Bugs fixed. Customer satisfaction scores. Every one of them cracked under pressure. More lines of code is not automatically better code. More bugs fixed just means more bugs got created in the first place.
So they mixed hard numbers with human judgment, and asked three different groups to rate the same teams. The interesting part was what each group talked about when asked to explain their rating.
Results
Sales numbers. Products shipped. Did the team deliver what it promised?
Culture
What it actually feels like to work here. Was this a team worth being part of?
Ownership, vision, goals
Sitting between the two, they cared about both the big picture and the people in it.
Nobody was wrong. They were each looking at a different face of the same object. So effectiveness got defined by all four measures together, including sales against quarterly quota.
Who is on the team barely mattered
They tested team composition against team dynamics. Personality traits. Skill sets. Emotional intelligence, using questions drawn from the Toronto Empathy Questionnaire. Tenure, seniority, location, team size, the ratio of introverts to extroverts.
Almost none of it predicted anything.
What mattered was less about who was on the team, and much more about how the team worked together.
Five behaviors came through. They held up across engineering and sales, across different countries, and across every way the team measured effectiveness. And they are not equally weighted. The first one does most of the lifting, and to a large extent it makes the other four possible.
What effective teams actually do
Ranked by how much each one predicts. The bar under each heading shows roughly how much weight it carries.
Psychological safety
People feel able to take an interpersonal risk in front of each other. They can admit a mistake, ask an obvious question or float a half-formed idea without anyone deciding they are ignorant, incompetent or difficult.
- Opinions, questions and concerns get raised in the room while they still matter
- Mistakes are treated as information, not as a character flaw
- People rely on each other, and disagree with each other, in the same meeting
- Feedback happens regularly and in both directions
- Somebody makes sure the quiet people get heard
Reliability
People do what they said they would do, to a standard the team is proud of, by the time they said they would do it.
- Agreed goals get hit, and the bar for quality is real
- Members can count on each other to carry their share
- Expectations get stated, and whether they were met gets discussed
- Communication rather than control
- There are rules for how the team works together, and people follow them
Structure and clarity
Everyone knows what is expected, how to deliver it and what happens as a result. Goals are specific, stretching and actually achievable.
- Clear roles and clear responsibilities, with no quiet overlaps
- Shared values underneath the way the team works
- Decisions get made a defined way, so people know how to influence them
- Everyone has the information they need to make good calls in their own patch
- Progress is visible, and updates are regular rather than requested
Individual sense of purpose
The work matters to people personally. What that means is different for everyone: financial security, supporting a family, helping the team win, or simply doing something you are good at.
- People can point to why their piece of the work is worth doing
- There is room to grow rather than just to deliver
- Individual strengths get used, not just tolerated
- A real sense of achievement, not only a closed ticket
- Enough slack in time and budget to try something new occasionally
Significance and influence
People believe the work makes a difference. Not just that it is finished, but that it changed something for somebody.
- People can trace a line from their work to the organization succeeding
- Value shows up for a customer, internal or external
- Individuals get visibility for what they contributed
- The team is trusted to organize itself, and given the room to do it
Psychological safety is not being nice
This is where the idea usually gets mangled. Safety gets heard as comfort. Avoid conflict. Do not upset anyone. Soften the feedback until it disappears.
Amy Edmondson, the Harvard researcher who named the concept, pairs it with something else entirely: accountability. High standards. Where you land depends on how much of each you have.
This is where our own mindset asks something specific of us. Think big calls for stepping outside comfort zones and taking calculated risks. Nobody does that on a team where the last person who tried got punished for it.
Safe teams argue more, not less. They just argue about the work instead of around it. The research since has been consistent on this: safety without a learning orientation gets you a very agreeable team that goes nowhere.
And it is built one interaction at a time. The single biggest lever is how a leader reacts the first time somebody brings bad news. That one reaction sets the price of honesty for months.
Hybrid work removed the free version
Most of what used to build safety happened by accident. The corridor conversation. Reading someone's face across the desk. The five minutes after a difficult meeting where somebody says what they really thought.
On a distributed team none of that is automatic. Video calls compress the signals people use to judge whether it is safe to speak. Silence on a call reads as agreement when it usually is not.
So the work that used to be free now has to be deliberate. That is the practical shift since 2015, and it is the reason a decade-old study still gets quoted every week.
Four things a leader can do this week
Go first
Say what you got wrong before you ask anyone else to. Naming your own mistake costs you almost nothing and tells the team exactly what is allowed.
Ask, then wait
Ask a real question and sit through the silence. Most people stop after four seconds and answer it themselves. Count to ten instead.
Watch how you take bad news
The first time somebody brings you a problem early, your reaction is the whole policy. Thank them for the timing before you deal with the content.
Be specific about who owns what
Ambiguity feels flexible from the top and feels like a trap from below. Write down who decides, who delivers and what good looks like.