By the time a project is publicly in trouble, the warning signs have usually been sitting in the room for months, felt by two or three people who never quite said them out loud. Bias mitigation is the unglamorous craft of getting those people to speak before the decision is locked, not after the post-mortem. Three tools do most of the work: the pre-mortem, red-teaming, and the devil's advocate. Each has real research behind it, and each has a way of failing if you run it badly.

The quick version

  • Pre-mortem: before you commit, imagine the plan has already failed, then list why. Imagining a fixed outcome makes people far better at naming the causes.
  • Red-teaming: assign a separate group whose only job is to attack the plan, assumptions, evidence, blind spots, as a competitor or critic would.
  • Devil's advocate: appoint someone to argue against the consensus. Useful, but a known weaker cousin of the other two, people discount an objection they know is staged.
  • The shared point: good decisions need dissent designed in. None of these works as a checkbox; the structure has to make disagreement safe and expected.

The idea in depth

All three tools exist to solve one problem that the research keeps rediscovering: groups converge on a shared story too quickly, then defend it. The fixes differ in how they pry that story back open.

Why agreement is the danger, not the disagreement

The foundational case is Irving Janis's Victims of Groupthink (1972), his study of US foreign-policy fiascoes, most famously the 1961 Bay of Pigs invasion. Janis argued that cohesive, high-status groups under pressure suppress doubt to preserve harmony, producing an illusion of unanimity in which silence is read as consent. His prescriptions read like a charter for everything below: assign a member to be a devil's advocate, invite outside critics, and have the leader withhold their own view until others have spoken (Britannica overview).

Daniel Kahneman gives the individual-level account in Thinking, Fast and Slow (2011). His coinage WYSIATI, "what you see is all there is", describes how confidence is built on the coherence of the story we can tell, not the completeness of the evidence. A neat, vivid plan feels right precisely because it leaves out what we can't see. The practical move: stop asking "are we confident?", you always will be, and manufacture a reason to go looking at the half of the picture that's missing.

The pre-mortem: borrowing the clarity of hindsight

The pre-mortem, popularised by psychologist Gary Klein in Harvard Business Review in 2007, is the most efficient way to do exactly that. As Klein puts it, "team members assume that the project they are planning has just failed" and then write the history of that failure (Performing a Project Premortem). The mechanism is a finding from Deborah Mitchell, Jay Russo and Nancy Pennington, published in the Journal of Behavioral Decision Making in 1989: prospective hindsight, imagining an event has already happened, increased people's ability to correctly identify reasons for a future outcome by roughly 30% ("Back to the future," Mitchell et al., 1989).

That number is doing something specific. "What could go wrong?" invites a vague list nobody owns. "It's a year from now and this failed, what happened?" treats the failure as certain, which licenses people to be concrete and stops the optimist in the room from batting each risk away. So in practice: run a ten-minute pre-mortem on any decision you'd hate to get wrong, and have people write their answers down silently before anyone hears the senior person's view. That last bit is Kahneman's own advice, decorrelate the judgments before they contaminate each other.

flowchart TD
  A(["Plan looks solid, team aligned"]) --> B("Pause before committing")
  B --> C(["Assume it's a year later
and the plan failed"])
  C --> D("Each person writes the causes
silently, in private")
  D --> E("Share, cluster, surface
the risks nobody owned")
  E --> F(["Strengthen the plan
or change the decision"])
The pre-mortem turns a hypothetical failure into a fixed fact, which is what unlocks honest causes. Leaders Loop

Red-teaming and the devil's advocate: structured opposition

Where a pre-mortem reframes the question, red-teaming reassigns the people. A red team is a separate group tasked with attacking a plan from the outside, its assumptions, its evidence, the competitor's likely response. The modern discipline was formalised by the US Army's University of Foreign Military and Cultural Studies (founded 2004), whose Applied Critical Thinking Handbook packages structured techniques for challenging assumptions and mitigating groupthink. Bryce Hoffman's Red Teaming (2017) translates it for business, the pitch is to make critical and contrarian thinking a built-in part of how any organisation plans, rather than a thing that happens by accident when someone's brave. The edge a red team has over a lone dissenter is mandate. Opposition is the job, so nobody pays a social cost for doing it.

The devil's advocate is the lightest-weight version, one person assigned to argue the contrary case. It's better than nothing, and it's what most teams can actually do on a Tuesday. But here is the honest limitation, and it's a sharp one.

"In general, no role playing technique stimulates divergent thinking as does authentic dissent."

That's the conclusion of Charlan Nemeth's experimental work (summarised across her studies and her 2018 book In Defense of Troublemakers): a staged contrarian tends to make a group bolster its original view rather than genuinely rethink it, because everyone knows the objection isn't real. Authentic dissent, someone who actually believes the opposing case, produces more original thinking and more real attitude change (Nemeth et al., 2001). What follows from that is awkward but worth doing: prefer real disagreement to the performed kind. Instead of appointing a token sceptic, find the person who actually has doubts and protect them while they say so. And if you have no choice but to assign the role, at least rotate it, and give the advocate enough time and evidence to build a case they can half-believe themselves.

flowchart LR
  A(["A decision worth de-biasing"]) --> B{"How much
opposition can
you resource?"}
  B -->|"10 minutes"| C(["Pre-mortem
reframe the question"])
  B -->|"A meeting"| D(["Authentic dissent
protect a real sceptic"])
  B -->|"A team"| E(["Red team
mandate the attack"])
  C --> F(["Surface hidden risks,
then decide"])
  D --> F
  E --> F
Pick the tool by how much opposition you can afford to resource, and prefer real dissent to a staged role. Leaders Loop

A worked example

A 40-person software company is about to sign a 12-month deal to rebuild its billing system on a new vendor platform. The leadership team is aligned, the demo was impressive, and the head of engineering is keen. This is exactly the moment the research warns about: cohesion, time pressure, a vivid story, and a senior champion.

Before signing, the COO runs a 15-minute pre-mortem. The prompt isn't "any concerns?", it's "it's next June, the migration has been abandoned, and we've lost a quarter of work. Write down what killed it." Each of the eight people writes privately for four minutes, then reads out. Three independently name the same thing: the vendor's API can't handle their refund logic, and nobody has tested it. (These figures are illustrative.) That risk had been invisible because the demo never touched refunds, a textbook WYSIATI gap.

That single surfaced assumption changes the decision. The team doesn't cancel; they make signing conditional on a two-week paid pilot that tests refunds first. The pilot is cheap insurance bought with ten minutes of structured pessimism, and notice they didn't need a standing red team or a staged devil's advocate to get there. The pre-mortem did the work because it was small enough to actually run.

Frequently asked questions

Isn't this just negativity that kills momentum?

The opposite, when it's timed right. Klein's point is that a pre-mortem is run before commitment, and it often increases buy-in: people who've had their doubts heard commit harder to the plan that survives. Unstructured negativity drifts; a bounded ten-minute exercise with a clear end does not.

Pre-mortem, red team, devil's advocate, which one do I actually use?

Match the tool to what the decision is worth and what you can resource. Default to a pre-mortem; it's the cheapest and, dollar for minute, the most effective. Stand up a red team for high-stakes, hard-to-reverse bets. Use a devil's advocate when you have nothing else, but know it's the weakest of the three.

Why does the devil's advocate underperform?

Nemeth's research found that when people know an objection is staged, they treat it as theatre and use it to reinforce their existing view. Real dissent, a colleague who genuinely disagrees, is what actually broadens a group's thinking. Protect the real sceptic before you appoint a fake one.

Does the leader's behaviour matter?

It's decisive. Janis found groups defaulted to the leader's stated preference. The cheapest single intervention is for the most senior person to speak last and to ask for the written, independent view first, so the room isn't anchored before it thinks.

How often should we do this?

Reserve it for decisions you'd genuinely hate to get wrong and that are costly to reverse. Run a pre-mortem on every one of those. Running them on trivial, easily-undone choices just trains people to tune the exercise out.

Related in the Toolkit

Where to go next