Why smart teams still make poor decisions

Intelligence is not a defence against groupthink. Research consistently shows that cohesive, high-performing teams with talented members produce some of the worst collective decisions precisely because their cohesion makes them effective at defending poor decisions as well as good ones. And project delivery teams are, by their nature, exactly the kind of group most vulnerable to this dynamic.

How groupthink works in project environments

Irving Janis identified eight symptoms of groupthink: the illusion of invulnerability, collective rationalisation, the belief in the group’s inherent morality, pressure on dissenters, self-censorship, the illusion of unanimity and the appointment of unofficial guardians who filter uncomfortable evidence from the group.

Project environments have structural features that intensify every one of these. Delivery pressure makes challenge feel disruptive. Senior sponsor visibility creates authority gradients that suppress dissent. Milestone deadlines create commitment escalation, where reversing course feels like failure rather than good governance. Groupthink is most dangerous in project environments not because it is rare, but because it is mistaken for alignment. A team that has stopped asking difficult questions feels, from the inside, like a high-performing team.

How AI makes this worse

AI intensifies the groupthink dynamic in a specific and largely unrecognised way. When a board receives an AI-generated delivery confidence score of 89%, the cost of challenging that figure rises significantly. The AI has managed to produce a number with apparent precision and authority and questioning it feels like questioning data rather than questioning judgement, a higher-stakes social act.

This is why automation bias and groupthink compound each other: the AI confidence becomes a social object the group organises around, rather than a proposition the group challenges. The more authoritative the AI output appears, the harder it is to challenge, and the less likely challenge becomes.

Cognitive diversity as risk management

One of the most underused protections against groupthink is cognitive diversity: team members who process information and reach conclusions through genuinely different cognitive pathways. Research has consistently found that cognitively diverse teams solve unfamiliar problems faster than homogeneous ones.

Neurodivergent professionals, including those with autism, ADHD, dyslexia and dyspraxia, frequently bring pattern recognition, lateral thinking and hyper-focused technical capability that conventional team structures routinely undervalue. The same characteristics that appear as friction in standard meeting cultures, asking unexpected questions, resisting social consensus, thinking non-linearly, are precisely the cognitive styles most likely to surface the risks and assumptions that groupthink conceals. Including cognitive diversity in project governance is a risk management strategy, not a diversity initiative.

Psychological safety: The structural requirement

Amy Edmondson’s research on psychological safety shows that teams where it is safe to raise concerns, acknowledge uncertainty and challenge authority significantly outperform those where those behaviours are socially costly. Not because they are more comfortable, but because they have access to more accurate information. 

For project governance, psychological safety is not an aspiration. It is a structural requirement for the quality of risk information that reaches decision-makers. A project professional who does not create the conditions for honest challenge is not demonstrating governance competence. They are performing delivery theatre. 

Practice checklist

• Name groupthink symptoms explicitly: Make Janis’s eight symptoms part of team governance induction. Ask regularly: what are we collectively not saying? Who has stopped pushing back?
• Mandate pre-mortems: Before major decisions, assign someone to argue that the plan has already failed. Not to be contrarian, but to surface what shared optimism is concealing.
• Build cognitive diversity into governance: Include neurodivergent professionals and non-standard thinkers in governance roles. Different thinking models are a measurable risk management asset.
• Separate information from advocacy: Distinguish board papers that present evidence from those that advocate conclusions. AI-generated outputs should be clearly labelled, not embedded as neutral data.
• Ask the standing question: Build ‘What are we not being told?’ into board culture as a standing agenda prompt, not an exceptional one. 

 

You may also be interested in:

• Visit APM bookstore: Evolving project leadership
• APM podcast: Why psychological safety is not a nice-to-have
• APM research: Dynamic Conditions for Project Success