The loudest person in a planning meeting can make an estimate sound settled before quieter specialists have explained their assumptions. A Delphi process changes the order: participants respond individually, receive structured feedback and have another opportunity to judge the question.
RAND’s Delphi manual describes anonymity, repeated data collection, feedback and a statistical group response as central characteristics. The method needs design and facilitation; sending the same survey twice is not enough on its own.
Start with a question people can answer on the same terms
Imagine a team estimating when a service could be ready for a pilot. One specialist interprets “ready” as technically deployed; another includes staff training and customer support. Averaging their dates would conceal a disagreement about the milestone, not resolve uncertainty about the work.
Define the milestone, assumptions and response format first. Ask for reasons and dependencies alongside the estimate. In a later round, provide an anonymised summary of the range and the relevant reasoning, then let participants reconsider their own answer.
Do not edit the feedback until every response appears to agree. A minority view may identify a dependency that the majority overlooked. Record meaningful disagreement and the conditions under which a different answer would become plausible.
Agree on how the process ends
Set the number of rounds or a stopping rule in advance, along with how the results will be reported. Otherwise the facilitator can keep asking until the panel produces a preferred answer. Decide who belongs on the panel based on the question, not merely seniority or availability.
A practical output might contain a central estimate, a range, key assumptions and unresolved concerns. It should also identify the date of the exercise, because new evidence can change a forecast.
Delphi can make dispersed expertise easier to hear and compare. It cannot turn shared assumptions into facts or guarantee that the future will cooperate. Its value lies in making the reasoning and remaining uncertainty more visible before a decision is taken.