Decisions
3–10 days

Delphi method

Collect anonymous estimates, share the distribution without attribution, and repeat — so the group converges on evidence rather than on whoever spoke first.

Time cost
3–10 days
Output
A final distribution with the surviving rationales.
Steps
5

Use when

  • You need a group estimate and status differences would distort a discussion.
  • The experts are distributed and cannot meet.
  • A previous meeting produced instant convergence, which usually means anchoring rather than agreement.

Do not use when

  • Speed matters. Two or three rounds takes days at best.
  • The group needs to build shared understanding, not just a number. Delphi produces the number and none of the understanding.

Inputs required

  • A question with a numeric or rankable answer
  • Five to twenty participants
  • A facilitator who does not participate

Procedure

  1. 01

    Write one precise question

    Ambiguity destroys this method, because participants answer different questions and the spread looks like disagreement.

  2. 02

    Collect round one privately

    Estimate plus a short written rationale, submitted to the facilitator only. No one sees anyone else’s.

  3. 03

    Share the distribution anonymously

    Median, range, and the rationales stripped of attribution. Include the outlying arguments — they carry most of the information.

  4. 04

    Re-estimate

    Participants revise or keep their estimate. Anyone holding an outlying view is asked to say why, still anonymously.

  5. 05

    Stop when it stabilises

    Usually two or three rounds. Report the final distribution, not just the median — residual spread is a real finding about the question.

Characteristic failure mode

Forced convergence. Running rounds until the spread is narrow produces agreement through fatigue, and a false impression of certainty about something the group genuinely does not know.

Worked example

A professional body estimates how long a new process will take to adopt.

  1. 01Round one: estimates from 4 months to 30 months, median 12.
  2. 02Rationales shared. The 30-month estimate cites a dependency nobody else had considered.
  3. 03Round two: median moves to 18, spread narrows to 9–30.

Result

The median moved by six months because of one anonymous rationale that would not have survived being said aloud by its author in a meeting.

Where to go next