Group Judgment and Pre-Mortems

A group can combine diverse knowledge and detect individual blind spots, but discussion also creates anchoring, conformity, status pressure, and diffusion of responsibility. Better group judgment comes from designing how information enters, disagreement is protected, and authority is exercised.

“Everyone agreed” is weak evidence if the most senior person spoke first, nobody recorded a private estimate, and disagreement carried social cost. Conversely, disagreement is not automatically wisdom; it must connect to evidence, values, assumptions, or consequences.

Design the sequence before the meeting. Once a room has publicly converged, asking “Any objections?” rarely recovers the information that silence buried.

Decision rights specify who recommends, who supplies input, who must approve, who executes, and who makes the final choice. Ambiguity creates two failures: endless consensus-seeking or authority that ignores relevant knowledge.

Use five roles:

Role	Responsibility	Not the same as
Owner	Makes the final call and explains it	Doing all analysis
Recommender	Integrates evidence and proposes an option	Owning the final decision
Contributor	Supplies expertise, stakeholder effects, or data	Holding a veto
Approver	Confirms a defined legal, safety, budget, or policy boundary	Re-evaluating every preference
Executor	Implements and reports operational feedback	Automatically owning strategy

One person can hold several roles, but the final owner should be singular for a bounded decision. “The committee owns it” often means no individual is accountable for resolving conflict.

Approvals need scope. Security may approve whether controls meet policy, not whether the product strategy is attractive. Finance may confirm funding, not select a technical architecture. Write each approver’s boundary to prevent hidden vetoes.

An independent judgment is an estimate or ranking formed before exposure to other participants’ answers. Independence preserves information that would otherwise collapse around the first confident speaker.

Before discussion, ask each participant to submit:

• preferred option;
• probability of the primary success outcome;
• largest downside;
• assumption most likely to be false;
• confidence range;
• one fact that would change the recommendation.

Display the distribution, not only the average. Estimates of 30%, 35%, and 90% average to about 52%, but the disagreement is the valuable signal. Ask the 30% and 90% estimators which evidence, base rate, or definition differs.

After discussion, collect a second private round. Movement shows learning; persistent spread reveals a real unresolved assumption or value conflict that a verbal consensus might hide.

Constructive dissent challenges a proposal by naming a specific assumption, consequence, value, or piece of evidence. It differs from vague negativity and from personal opposition.

Use this structure:

I disagree with [claim or option] because [evidence or value] suggests [consequence]. The disagreement would be resolved or reduced by [test, threshold, or new evidence].

Weak dissent	Strong dissent
“This feels too risky.”	“The adverse case exceeds our £200,000 loss limit, and no tested rollback exists.”
“Customers will hate it.”	“In the pilot, renewal fell 8 points among the affected segment; scale only if a revised flow recovers at least 6.”
“Finance never understands product.”	“The model omits two quarters of migration labour, which changes the cost ranking.”
“We should be bolder.”	“The staged option closes the market window; a bounded full launch may protect the growth objective better.”

The owner must reward useful challenge visibly: summarize it fairly, request the relevant test, and record whether it changed the decision. Punishing a well-founded objection teaches the whole group to withhold the next one.

Do not require the dissenter to solve the entire problem. Identifying a serious hidden failure is valuable even before a replacement design exists.

A pre-mortem asks a team to imagine that a plan has failed and explain why. Gary Klein’s method uses prospective hindsight—placing people mentally after the failure—to make concerns easier to express and expand the set of threats considered.

Run it in 30–45 minutes:

1. State the plan and time horizon: “It is eighteen months later; the expansion failed.”
2. Give everyone three quiet minutes to write reasons independently.
3. Collect one reason per person in rounds until lists are exhausted.
4. Clarify mechanisms without defending the plan.
5. Rank threats by impact, likelihood, and detectability.
6. Assign mitigations, indicators, owners, and deadlines.
7. Reconsider the plan if a severe threat cannot be reduced.

Good failure stories are concrete: “Supplier certification slipped six months because we never reserved testing capacity,” not “execution was poor.” Specific mechanisms can produce early signals and safeguards.

A pre-mortem is not a ritual to validate the plan. If no identified risk can change scope, sequence, budget, stop rules, or the decision itself, the exercise has no authority.

A red team is a person or group explicitly authorized to challenge a proposal, model, system, or plan from an adversarial perspective. It is useful when the decision is consequential, the proposal has strong sponsorship, or failure modes are difficult to surface in ordinary discussion.

Set rules:

• The red team receives the same evidence and enough time to investigate.
• Its mandate names what to challenge: assumptions, security, economics, operations, stakeholder harm, or all of these.
• Claims require a mechanism or evidence, not rhetorical aggression.
• The proposal team must restate the strongest criticism before replying.
• A neutral owner or reviewer resolves factual disputes and assigns tests.
• The red team’s work ends; it does not become a permanent shadow authority.

Rotate roles when possible. Permanent critics may become identified with opposition, while advocates stop listening. A person can defend the proposal in one decision and challenge another.

Use red teams selectively. Routine reversible choices need lighter challenge; making every small decision adversarial slows work and exhausts trust.

A decision record preserves the frame, options, evidence, disagreements, owner, and commitment. It allows action after debate while keeping minority information available for review.

Include:

Field	What to record
Decision	Exact choice, owner, and date
Outcomes and constraints	What the choice is trying to protect
Options	Serious alternatives considered
Evidence	Base rates, estimates, tests, and known gaps
Recommendation	Leading option and tradeoff
Disagreement	Strongest unresolved objections in the dissenter’s own logic
Resolution	Owner’s reason for choosing despite disagreement
Conditions	Mitigations, stop rules, review date, and reopen triggers

Do not rewrite dissent as “some concerns were raised.” Record: “Operations estimates only 40% probability of migration by September because two required integrations have no owner.” Precision protects the signal.

After resolution, distinguish disagree and commit from silence. A contributor may state disagreement, ensure it is recorded, and then support execution unless the action violates law, safety, ethics, or a defined approval boundary.

The owner must not later pretend the objection never existed. Review should compare the disputed assumption with what happened.

A worked example shows the methods in sequence. A retailer is considering opening five stores in a new region. The chief executive favors speed; finance worries about cash; operations worries about staffing; commercial leaders expect strong demand.

Rights: the chief executive owns the expansion decision. Finance approves compliance with the minimum cash-reserve policy. The commercial director recommends. Operations, property, people, and local market specialists contribute.

Independent round: success estimates range from 35% to 80%. The commercial team uses national demand; the low estimates use local footfall and recruitment data. The group discovers it has been using two meanings of success: opening on time and reaching break-even within twelve months.

Pre-mortem: imagined failure reasons include delayed permits, inability to recruit managers, a competitor’s price response, construction overruns, and weak repeat visits after opening promotions. Staffing and repeat demand rank highest because they combine impact, uncertainty, and late detection.

Red-team challenge: a separate analyst rebuilds the demand model using three comparable regions and finds the forecast ignores cannibalization of online sales. A two-store staged option, previously dismissed as timid, now preserves learning and cash.

Resolution: open two stores first; reserve leases on three more with expiry dates. Scale only if repeat visits, contribution margin, and manager retention meet thresholds for two quarters. The commercial director records disagreement that the market window may close, and the owner accepts that opportunity cost explicitly.

The process did not make everyone agree. It converted status conflict into estimates, mechanisms, tests, and an accountable choice.

A group-decision protocol is a repeatable agenda that protects independent information and still reaches action. Use it for choices important enough to require several perspectives but bounded enough for one owner.

Stage	Time	Action
Prepare	Before	Send frame, options, evidence, rights, and decision date
Independent round	10 min	Submit option, probability, downside, and key assumption privately
Reveal	10 min	Display distribution and identify largest disagreements
Evidence	20 min	Examine reasons, base rates, and missing facts
Challenge	20 min	Run focused dissent or pre-mortem
Integrate	15 min	Revise options, mitigations, thresholds, and recommendation
Decide	10 min	Owner resolves or names one bounded evidence gap
Record	After	Publish rationale, dissent, owners, and review triggers

Final checklist:

• [ ] One person owns the final bounded decision.
• [ ] Contributor, approver, recommender, and executor roles are explicit.
• [ ] Participants estimate independently before discussion.
• [ ] The distribution and reasons are examined, not hidden by an average.
• [ ] Dissent names a claim, consequence, and possible test.
• [ ] A pre-mortem produces owned changes or explicit risk acceptance.
• [ ] Red-team scope and adjudication are defined for high-stakes cases.
• [ ] The strongest unresolved disagreement is recorded precisely.
• [ ] Resolution includes mitigations, stop rules, and reopen triggers.
• [ ] Commitment does not override legal, safety, or ethical boundaries.

Strong group judgment does not eliminate authority or disagreement. It protects diverse evidence long enough for a named owner to make a clear choice and remain accountable to what the group knew.