Why smart people don’t make smart teams

The hiring assumption breaks under team work

Most organisations recruit as if team performance is the sum of individual capability. The logic is understandable. Hire people with strong technical knowledge, sound judgement and a record of delivery. Put enough of them in one room. Expect the result to improve. That logic works better for work done alone than for work that depends on coordination, disagreement, handoffs and decisions made under uncertainty.

Complex delivery does not ask only whether one person can find an answer. It asks whether a group can notice the right problem, bring forward partial information, test competing explanations and act before the cost of delay grows. A strong individual can contribute to that work. They cannot perform the group’s conversations, boundaries and decision habits for everybody else.

This is why a team can be full of impressive CVs yet still wait for permission, repeat the same argument, discover dependencies late or optimise a local task while the wider outcome slips. Those are not signs of low intelligence. They are signs that the team has no reliable way to use the intelligence already present.

MIT tested the difference

MIT researchers, led by Anita Williams Woolley and colleagues, studied groups completing a range of tasks. Their 2010 Science paper reported a general collective-intelligence factor: groups that performed well on one kind of task tended to perform well on others. The important finding for leaders was equally clear. The average or maximum individual intelligence of group members did not explain group performance in the way the hiring assumption predicts.[1]

That does not make individual ability irrelevant. A team needs relevant knowledge and skill. The finding changes the order of attention. Once people are capable enough to do the work, the team’s ability to combine capability becomes the constraint worth measuring.

The group is the unit of performance

Collective intelligence is not a flattering label for harmony, morale or a workshop. It is a measurable property of a group. It describes how well a team performs across varied tasks that require reasoning, coordination and judgement. The team, not the person, is the unit that receives the score.

MIT’s work also pointed to observable conditions associated with stronger collective intelligence: social sensitivity and a more equal distribution of conversational turn-taking. In plain terms, teams do better when members can read one another and when the most confident voice does not own every important conversation. These are working conditions. They can be observed, practised and improved.

Google looked inside effective teams

Google’s Project Aristotle ran team-effectiveness research from 2011 to 2015. Its question was practical: what makes one team effective when another team, in the same organisation and with access to similar talent, is not? The research did not find a single perfect mix of personalities. It identified team dynamics that shape whether people can do useful work together.

Google’s published account puts psychological safety at the top of its list. It also identifies dependability, structure and clarity, meaning, and impact. These are not separate from delivery. Psychological safety determines whether a warning is voiced. Dependability determines whether a promise can be trusted. Structure and clarity determine whether people know the target, the role and the boundary. Meaning and impact connect daily work to a result worth pursuing.[2]

Project Aristotle therefore adds a useful operating view to the MIT finding. If collective intelligence is the team-level property, psychological safety, clarity, dependable follow-through and visible impact are conditions that help a team use it. They make information available in time and give people a reason and a route to act on it.

Collective intelligence is not group agreement

A team with high collective intelligence is not a team that agrees quickly or avoids tension. Fast agreement can be a warning sign if it comes before evidence has been heard. The stronger test is whether disagreement improves the decision. Can a junior specialist question an assumption? Can an expert say “I do not know”? Can the team separate a claim from the evidence behind it? Can it make a choice, record its reasons and return to the choice when the facts change?

These questions matter because organisations often mistake activity for intelligence. More meetings, more status updates and more senior approvals can create the appearance of control while slowing the point at which information becomes action. A collective-intelligence measure asks a different question: did the team turn its combined knowledge into a better outcome?

That makes the property measurable. Start with a real outcome, a visible baseline and a review date. Observe how the team defines evidence, shares uncertainty, assigns outcome ownership, tests risk and routes decisions. Then review whether the outcome, risk exposure or quality guardrail moved. The measure is not whether people enjoyed the meeting. It is whether the team got better at thinking together in a way that changes work.

The range of team performance makes the choice clear

In the dataset behind a 2025 study of the video game industry, revenue generated per team member ranged from 1.33 to 15,158. The authors log the measure, because the distribution is that skewed. Teams doing comparable work, drawing on the same talent pool, in the same industry, produced outcomes spanning four orders of magnitude.[3]

Those are money figures for team output, not scores, so they are not comparable to any range of individual measurements, and we are not going to present them as if they were. The honest reading is narrower than it first looks, and it is still decisive: the variation that matters sits between teams rather than inside them. MIT's own study found a group-level factor accounting for more than 43% of the variance in how groups performed.[1]

That is a warning against spending all improvement effort on individual capability while leaving the team system unexamined.

The same warning sits behind Agile failure figures. InnerCircle reports that 32% of Agile programmes fail, with an estimated £37bn wasted in the UK alone. Scrum Inc. reports that 47% of Agile transformations fail. Frameworks can help. But no framework can make a team share evidence, own an outcome or make a timely decision when those habits are absent.[4]

You cannot hire your way out of a team-level problem

Replacing people may sometimes be necessary. It is not a general answer to a collective-intelligence problem. A new hire enters the same unclear target, the same guarded meeting, the same unrecorded decision and the same approval queue. Over time, even an excellent person learns the operating pattern around them.

Nor is the answer to place all responsibility on one leader. Leaders set conditions, but a team needs practices that still work when the leader is busy, absent or wrong. The work has to live in the team: a target people can explain, evidence standards that make assumptions visible, clear outcome contracts, early learning on the greatest risk, and decisions made where information is richest.

That is why the first practical move is a measurement, not a recruitment campaign. Measure the team across the conditions that produce or suppress collective intelligence. Find the weakest constraint. Change one practice. Re-measure against a real outcome. A score matters only when it leads to a better decision and a visible result.

Raise the intelligence the team already has

The value of the research is not a new label. It is a more useful management choice. Stop treating the team as a container for individual talent. Treat it as a performance system with its own strengths, risks and evidence. That moves the conversation from “Who is the star?” to “How well can this group see, think, decide and learn together?”

MIT tells us that the property exists. Google shows that its operating conditions can be made visible. The performance range shows why it deserves attention. You cannot hire your way out of a collective-intelligence problem. You have to measure it, raise it and keep checking whether it changes the outcome.

Sources

1. Woolley, A. W. et al. (2010), “Evidence for a Collective Intelligence Factor in the Performance of Human Groups”, Science. Read the paper.
2. Google re:Work, “The five keys to a successful Google team”. Read Google’s account.
3. SAGE, 2024, published individual and team performance ranges. Read the published data.
4. InnerCircle, “UK wasting £37bn a year on failed Agile IT projects”; Scrum Inc. reports 47% of Agile transformations fail. Read the InnerCircle report.