Decisions
10 min

When a group is worse than one person

Name the conditions under which group decision-making degrades, so that involving people is a choice rather than a default.

Time cost
10 min
Output
A decision about whether and how to involve the group.
Steps
6

Use when

  • Before convening anyone. This determines whether the meeting should happen.
  • A group has produced a decision worse than any individual member would have made.
  • Meetings consistently converge fast and turn out wrong.

Do not use when

  • Acceptance by the group is required for execution. Then involvement is necessary even where accuracy suffers.

Inputs required

  • The proposed group
  • The decision
  • Honesty about the group’s dynamics

Procedure

  1. 01

    Check for a dominant voice

    If one person“s view reliably becomes the group”s view, the group is one person with witnesses — and slower. Use silent-first methods or do not convene.

  2. 02

    Check for shared information bias

    Groups discuss what everyone already knows and skip what only one person knows. The unique information is the reason to have a group at all, and it is the part that gets dropped.

  3. 03

    Check for correlated error

    If everyone shares a background, they share blind spots. A group of similar people is a less reliable estimator than it appears, because the errors do not cancel.

  4. 04

    Check for polarisation

    Groups move toward the extreme of their initial lean after discussion. A group that starts mildly in favour ends strongly in favour, with no new evidence.

  5. 05

    Check for diffused accountability

    When everyone decides, nobody owns the outcome and nobody reviews it. Assign a single owner even for a group decision.

  6. 06

    Decide the structure, then convene

    If two or more conditions hold, either fix the structure — silent writing, anonymous input, assigned dissent — or have one informed person decide and consult individually.

Characteristic failure mode

Convening anyway, because a meeting feels more legitimate than a decision. The group adds confidence without adding accuracy, and the confidence is the dangerous part.

Worked example

Eight colleagues with similar backgrounds meet to evaluate a proposal.

  1. 01Correlated error: all eight trained in the same discipline.
  2. 02Shared information bias likely: only two have read the underlying data.
  3. 03Structure changed: those two write a summary first, everyone estimates privately before discussion.

Result

The private estimates span a range that the meeting, run normally, would have collapsed in the first ten minutes.

Where to go next