What is the Delphi Method?

The Delphi method is a structured technique for gathering and refining the judgment of a panel of experts through repeated rounds of anonymous questionnaires with controlled feedback. In each round, panellists answer independently; the facilitator aggregates the responses (statistics, ranges, the reasoning behind outliers) and feeds the summary back; panellists then reconsider and answer again, seeing where the group sits without knowing who said what. The process continues until responses stabilise, which may mean consensus, or a clarified, well-reasoned disagreement, which is a legitimate result. The method was developed at the RAND Corporation in the 1950s (named, with a wink, for the oracle at Delphi) to forecast technological and military developments for which no data existed, and its design targets a specific problem: group discussion is dominated by status, volume, and anchoring, so the method removes the group and keeps the collective.

Why Anonymity and Iteration Matter

The two design choices are the method. Anonymity strips away the social pressures that distort face-to-face judgment: deference to seniority, reluctance to contradict, the bandwagon that forms once an opinion is voiced, every focus-group pathology at once. Panellists change their minds in Delphi because of arguments, not because of who made them. Iteration with feedback lets the group learn from itself: an outlier's reasoning, shared anonymously, can move the majority, and a majority's stability can persuade an outlier, without either being cornered. The result is closer to what the panel collectively knows than either a poll (one round, no learning) or a meeting (learning, but corrupted by dynamics), and it is a working application of the independent-judgments-before-group-judgments principle.

Running One

1. Frame the questions precisely.
Delphi works on specific, answerable judgments (likelihood, timing, priority, impact), not on open exploration. Vague questions produce vague convergence.

2. Select the panel for expertise and range.
Ten to thirty panellists is typical; the selection is a judgment sample and should be documented as one, with deliberate diversity of perspective so that convergence means something.

3. Run a first round that is partly open.
Often a qualitative round to surface the issues, then structured rounds that quantify them.

4. Feed back honestly and completely.
Distributions, not just averages; the reasoning of dissenters, not just the mainstream. The facilitator's neutrality is the method's integrity.

5. Stop at stability, not at agreement.
Two to four rounds is common; the criterion is that answers stop moving. A stable split is a finding: the panel disagrees, and here is why.

Delphi in Product and Research Work

The method suits questions where evidence is unavailable and expertise exists: forecasting which technologies or regulations will reshape a market, prioritising a roadmap across stakeholders whose politics would otherwise dominate, converging a research team on severity ratings, or building expert-derived criteria for an evaluation (a Delphi round is a rigorous way to decide what a heuristic set should contain for a domain). It is also used to structure internal decision-making: anonymous, iterated estimates from a product team beat a planning meeting's anchored consensus on nearly every dimension except speed. Modern asynchronous survey tooling has made the mechanics trivial; the discipline of the facilitator remains the hard part.

The Limitations

Delphi produces expert opinion, refined; it does not produce evidence, and convergence can be convergence on a shared error. Panel selection determines the outcome more than any other factor, and a homogeneous panel converges quickly on its blind spots. The process is slow (weeks, across rounds), attrition between rounds is common, and facilitator bias (in framing, in summarising, in what dissent gets emphasised) can steer the group without anyone noticing. It is a tool for structured judgment under uncertainty, and it should be reported as exactly that.

The Takeaway

The Delphi method collects expert judgment without the room: anonymous answers, honest feedback, repeated rounds, stability as the stopping rule. Frame questions precisely, choose the panel for expertise and range, feed back distributions and dissent, and accept a well-reasoned split as a result. It replaces the loudest voice with the panel's best collective guess, which is the most a group of experts can honestly offer when the data doesn't exist yet.