The Delphi technique is a structured expert consensus method that project managers use to gather reliable estimates, risk assessments, and forecasts from panels of subject matter experts when direct data is unavailable or insufficient. Developed by the RAND Corporation in the 1950s and 1960s for strategic forecasting, the Delphi technique has been adopted across project management, risk management, technology forecasting, and policy development as a rigorous alternative to informal expert opinion gathering. For project managers facing estimation challenges where historical data is sparse — new technology implementations, novel business models, early-stage feasibility studies — the Delphi technique provides a systematic process for distilling expert knowledge into defensible, consensus-based estimates.
Why the Delphi Technique Exists
Expert opinion is one of the most valuable inputs to project planning, but gathering it poorly produces results that are worse than no expert input at all. The problems with informal expert opinion gathering are well-documented: the most senior or most confident expert dominates the discussion regardless of their actual knowledge relevance; group dynamics create pressure to conform to early positions (anchoring); experts who disagree with the majority suppress their dissent to avoid conflict; and the combination of expert views in group settings produces regression toward a mean that eliminates the informational value of outlier positions.
The Delphi technique addresses all of these problems through two mechanisms: anonymity (experts provide opinions without knowing others’ identities, eliminating status-based dominance) and iteration with controlled feedback (experts revise their estimates in light of the aggregate group position while retaining the right to maintain their dissenting view with justification). The result is a structured convergence toward consensus that preserves minority expert perspectives and produces statistically defensible final estimates.
The Seven-Step Delphi Process
Step 1: Define the Question
Delphi works best for questions that are genuinely uncertain but where expert knowledge can meaningfully reduce that uncertainty. The question must be precisely formulated — ambiguous questions produce divergent expert responses that reflect different question interpretations rather than different underlying knowledge. For project management applications, suitable Delphi questions include: “What is the realistic duration range for implementing this technology in our organisational context?”, “What is the probability that this regulatory requirement will change within the next 18 months?”, or “What are the five most likely technical risks in this architecture?”
Step 2: Select the Expert Panel
Panel selection is the most consequential decision in a Delphi exercise. Panel members should have genuine expertise relevant to the specific question — domain knowledge, practical experience, and access to current information. Panels of 8–15 experts typically balance diversity of perspective with manageability. Including experts with different organisational perspectives (technical, business, regulatory), different career stages, and different institutional affiliations produces richer, more robust convergence than homogeneous panels. The facilitator should maintain panel anonymity throughout the process — experts should not know who their co-panelists are.
Step 3: Round 1 — Open Questionnaire
The first round questionnaire is typically open-ended, asking experts to provide their estimates or assessments in their own words with supporting rationale. This unrestricted initial round prevents premature anchoring and ensures the full range of expert perspectives is captured before any convergence pressure is applied. Quantitative estimates should include a point estimate plus a confidence interval or range.
Step 4: Compile and Summarise
The facilitator compiles all Round 1 responses, calculates statistical summaries (median, interquartile range for quantitative estimates; thematic grouping for qualitative responses), identifies areas of agreement and disagreement, and prepares a structured summary for distribution to the panel. The summary presents the group’s aggregate position without revealing individual identities. Outlier positions — responses significantly outside the central tendency — are noted with their supporting rationale (anonymised) to give the full panel visibility of the range of informed expert views.
Step 5: Round 2 — Revised Estimates
Each expert receives the Round 1 summary and is asked to revise their estimate in light of the aggregate group position. Experts whose revised estimate remains outside the IQR (interquartile range) of the group are asked to provide written justification for their continued outlier position. This iteration structure creates gentle convergence pressure while explicitly protecting the right to maintain an evidence-based dissenting position — a critical design feature that distinguishes Delphi from groupthink.
“The Delphi technique’s power is not in finding the average expert opinion — it is in creating a structured process that reveals where genuine expert uncertainty lies and where well-informed consensus exists.” — Harold Linstone, co-author of The Delphi Method
Steps 6–7: Convergence and Finalisation
Rounds are repeated until a convergence criterion is met — typically when the IQR stops decreasing significantly between rounds (usually two to four rounds). The final expert consensus is documented with: the central tendency estimate, the confidence interval, the degree of remaining expert disagreement, and the key reasoning that drove convergence. For project management applications, the Delphi output feeds directly into the relevant planning artefact — schedule estimates, risk register probabilities, budget contingency calculations — as a documented, methodologically sound input.
Delphi vs Other Expert Estimation Techniques
Project managers should understand when Delphi is the right choice relative to other expert estimation approaches. Planning Poker — the Agile estimation technique using Fibonacci-sequence cards — is a faster, synchronous Delphi variant appropriate for sprint-level user story estimation with small, co-located teams. It preserves anonymity through simultaneous card reveal and iterates through discussion to convergence, but lacks the structured rounds and statistical rigor of formal Delphi. The Nominal Group Technique (NGT) is a structured group process for generating and prioritising ideas — it shares Delphi’s structure but involves in-person group interaction rather than anonymous remote rounds. Delphi is superior to both for high-stakes, complex estimates where full expert anonymity and multiple iteration rounds are warranted.
Delphi Technique Quick Reference
| Parameter | Recommended Specification | Rationale |
|---|---|---|
| Panel size | 8–15 experts | Balance diversity with manageability |
| Number of rounds | 2–4 rounds | Stop when IQR stops decreasing |
| Time per round | 5–10 business days | Allow thoughtful reflection |
| Convergence criterion | IQR < 25% of median or <15% IQR change | Objective consensus threshold |
| Anonymity | Full throughout all rounds | Prevents status-based dominance |
Key Takeaways
- The Delphi technique provides structured, anonymous, iterative expert consensus — solving the group dynamics problems (anchoring, conformity pressure, status dominance) that undermine informal expert opinion gathering.
- Panel selection is the most consequential Delphi design decision — panel members must have genuine, relevant expertise and diverse organisational perspectives.
- Anonymity throughout all rounds is a non-negotiable design requirement — removing it reverts the process to a conventional expert meeting with all the associated biases.
- Experts who maintain outlier positions in Round 2+ must provide written justification — this preserves minority expert knowledge that group convergence pressure would otherwise suppress.
- Stop when the IQR stops decreasing — typically two to four rounds — and document not just the consensus estimate but also the remaining degree of expert disagreement as a transparency input to planning decisions.
- Use Delphi for high-stakes, novel estimation challenges where historical data is sparse; use Planning Poker for sprint-level Agile estimation where speed and team familiarity are priorities.
