What needs to be finished first
A written description of the estimation object with boundary, assumptions, and desired unit exists.
Experts understand optimistic/most-likely/pessimistic estimation or are introduced before round 1.
What needs to be ready before start
Shared estimation sheet with anonymous cells per expert; estimation-object document; assumptions list; whiteboard or synthesis tool; timer per round.
One moderator; 4-12 experts from relevant disciplines; one scribe for assumptions log; optional sponsor who does not estimate.
Estimation-object document distributed at least 48 h before round 1; known constraints; reference data from similar initiatives; list of expert disciplines.
1-4 h plus asynchronous rounds
Prepare estimation sheet with anonymous columns. First round asynchronous or silent in room, nobody sees others' values. Fix clear unit.
The one question this method answers
Which range and expected value describe the estimation object most honestly under explicit expert assumptions?
Marker: Phase
| Step | Duration | Action | Hint |
|---|---|---|---|
1Phase 1: Kickoff and scope sync | 30-45 min | Moderator presents estimation object, assumptions, and unit. Experts ask clarifying questions. Scope corrections are logged. | If scope is still contested after 45 min, postpone. Estimation without scope consensus wastes time. |
2Phase 2: Round 1 - anonymous estimate | 20-30 min | Every expert enters O/M/P values plus assumptions into own column. No discussion during input. | Anyone stating values before reveal anchors the group. Moderator ensures simultaneous reveal. |
3Phase 3: Analyze spread | 30-45 min | Reveal range per value: min, median, max. Identify largest deviations. Experts with extreme values explain assumptions without debate. | Assumption differences are the valuable information. |
4Phase 4: Round 2 - estimate with assumptions knowledge | 20-30 min | Experts revise values based on others' assumptions. Changed values document what was learned. | Convergence is not the goal. If spread remains, uncertainty is real. |
5Phase 5: Document result | 15-30 min | Capture final range, expected value, and conflicted assumptions. For high remaining spread, propose spikes. | Point value without range is fake precision. |
What comes out at the end
Estimation document with estimation-object description, assumptions, anonymized values per round, final range with expected value, and open risks/spikes.
Snapshot per estimation round with date and round number. On scope change, create new estimate, do not overwrite old one.
Compact working template for Wideband Delphi with context, input, output artifacts, and next step.
Structured expert estimation with anonymous individual values, discussion, and iterative convergence.
When and for what do we use this method?
Which data, observations, decisions, or materials are available?
Short notes along the runsheet.
Owner, date, and success signal.
Concrete filled scenario, fictional example
Concrete filled scenario, fictional example
Estimation object: Migration of all existing client data from MariaDB to PostgreSQL incl. data-model adjustment, test migrations, and cutover weekend.
Unit: person-weeks
Experts: 6 (2 backend, 1 DBA, 1 SRE, 1 QA, 1 product lead)
Round 1
Assumption conflict: DBA assumed full schema refactor, backend assumed 1:1 migration. Clarification: minimal refactor, optimization in following quarter.
Round 2
Expected value (PERT): 12.1 person-weeks, range 5-21.
Recognize symptoms and steer against them
Values are named openly or experts hint tendencies before reveal.
Tool or moderator ensures simultaneous reveal. Discard violated round.
Extreme values move to middle without new insight.
Ask which new assumption changed the value. Without assumption update, keep value.
All experts come from same discipline.
Check disciplines before workshop: architecture, engineering, operations, domain, QA.
Sponsor reads only expected value, range disappears.
Always communicate range plus expected value.
Values are in table, rationale missing.
Assumption field required per expert. No assumption, no number.
Done signals checkable in under a minute
Go to the profile for purpose, similar methods, and sources or continue to the next method in the catalog.