Method 08 · Condense and decide
Delphi Survey
The Delphi method surveys a group of experts in several rounds, in writing and anonymously; after each round everyone receives the condensed group view and may revise their estimate. Anonymity takes status and rhetoric out of play, the feedback lets outliers state their reasons — and often it is precisely the dissenters' reasoning that is most valuable. Delphi delivers no truth, but a robust, traceable condensation of what informed people consider probable, together with the places where they remain divided.
Procedure
- Formulate the questions. Precisely, quantifiably where possible: "When will technology X reach TRL 8?" instead of "How do you see the future of X?"
- Assemble the panel. Diversity over size: different disciplines, industries, standpoints. Ten to thirty people usually suffice.
- Round 1. Anonymous, in writing; estimate plus brief reasoning.
- Feedback. Distribution of answers and the reasons of the dissenters, to everyone.
- Round 2 (and possibly 3). Revision permitted; stop when the distribution stabilises — not only at consensus.
Strengths
- Removes hierarchy and volume from the formation of judgement.
- Shows not only the median but the spread — and thus the real uncertainty.
- The dissenters' reasons are often the actual insight.
Limits
- Costly and slow; response rates fall over the rounds.
- Experts err collectively — Delphi does not average errors away.
- Quality stands and falls with the wording of the questions.
How we use it in our roadmaps
We use Delphi selectively where scanning and bibliometrics yield no timeline — for breakthroughs whose timing only specialists can estimate: error-corrected quantum computers, commercial fusion, humanoid robots in routine operation. The Delphi results enter the roadmap timeline as milestones with spread; the spread becomes the uncertainty band.
In the source apparatus we document which assessment stems from a Delphi round, how large the panel was and how wide the distribution. A Delphi milestone with a tight distribution is a different statement from one where the experts disagree by ten years — both are valuable, but different.
References
- Linstone, H. A. & Turoff, M. (eds.) (1975): The Delphi Method. Techniques and Applications. Addison-Wesley, Reading, MA.
- Häder, M. (2014): Delphi-Befragungen. Ein Arbeitsbuch. 3rd ed., Springer VS, Wiesbaden.
- Rowe, G. & Wright, G. (1999): The Delphi Technique as a Forecasting Tool: Issues and Analysis. International Journal of Forecasting, 15(4), 353–375.