The Delphi Method in Foresight - and How We Use It
Originally developed at the RAND Corporation in the 1950s, the Delphi method is a structured way to gather and synthesize expert opinion on questions about the future – particularly where data is scarce and uncertainty is high. It is a multi-stage, survey-based approach used in foresight, typically designed to arrive at a stable assessment through controlled feedback across several rounds, while avoiding the influence of dominant individuals.
In foresight, Delphi is less about asking open questions and more about testing sharply formulated hypotheses about how the future might unfold. It allows us to systematically confront these hypotheses with diverse expert perspectives – across hierarchies, geographies, and sectors – and to understand not only where views converge, but where they meaningfully diverge. The goal is not a single prediction, but a clearer map of plausible futures, underlying assumptions, and critical uncertainties.
At the Futurewise Company, we have adapted the method to fit strategic work with executives. Instead of anonymous multi-round surveys, we run Delphi as semi-structured 1:1 interviews based on hypotheses derived from earlier foresight tools such as Futures Wheels or scenario matrices. This approach preserves comparability across interviews while allowing for depth, nuance, and unexpected insights. We then synthesize these expert perspectives to refine scenarios, explore future growth opportunities, and build decision-relevant views on strategic risks and opportunities.
Starting With Hypotheses
Before we ever talk to an expert, we need to know what we are actually trying to learn. That foundation comes from earlier foresight work — typically a Futures Wheel, a 2×2 scenario matrix, or both.
These tools generate something useful: a map of the drivers, dependencies, and fields that matter for our research question. From that map, we extract two things. First, the topical territory we need to cover. Second, the conditions that are simultaneously high impact and high uncertainty — the places where the future is genuinely unsettled and where expert input will move our thinking the most.
This, for how we conduct Delphi interviews at the Futurewise Company, is the raw material for hypothesis formulation.
A hypothesis like “manufacturing will change in Germany” is useless. Nobody will disagree with it, and nobody will say anything interesting in response. Compare it to: “In 2036, most manufacturing and production will be outsourced. What remains in Germany is R&D only.”
That version is specific. It commits to a timeframe, a geography, and a structural claim. It is also slightly provocative — and that is intentional. Vague hypotheses get vague answers. Provocative hypotheses force experts to either defend or dismantle the claim, and either response gives you signal.
The rule we follow: be specific, and lean toward the provocative end. If your hypothesis could be the headline of a contrarian opinion piece, you are probably in the right zone.
Choosing the Right Experts
Once the hypotheses are sharp, we work backwards from them to figure out who needs to weigh in. A good Delphi panel is not just “smart people we can reach.” It is a deliberately structured group designed to surface different angles on the same question.
We think about diversity along several axes:
- Hierarchy. Strategic voices (C-suite, policy leads) see the system. Operational voices (engineers, plant managers, frontline analysts) see the friction. Both are necessary, and they often disagree in productive ways.
- Geography. A question about German manufacturing looks different from Munich, Shenzhen, and Detroit.
- Organizational background. We aim for a mix across academia, government, media, industry, civil society, think tanks, and consultancies. Each sector carries its own incentives and blind spots, and the contrast is where insight lives.
The goal is not balance for its own sake. It is to make sure that when consensus does emerge, it is meaningful – and when disagreement emerges, we know exactly which fault lines are doing the work.
Finding and Reaching Experts
The hardest part is often just getting people to talk to you. A few channels work consistently for us:
LinkedIn is the obvious starting point, but the higher-yield move is browsing speaker lists from relevant conferences. People who agree to speak publicly on a topic are, by definition, experts in their field and willing to talk about it — a rarer combination than it sounds.
Internal networks are underrated. Your CEO, your shareholders, and your sales team often have networks that took decades to build. Use them. A warm introduction from them beats a cold message every time.
Beyond that: scientists who have published recently in the area, VCs investing in the space (they have done their own diligence and are usually happy to share a slice of it), and journalists covering the beat. Journalists in particular tend to know a lot of people and have a strong sense of who is worth listening to versus who is just loud.
Designing the Questions
Generic questions produce generic answers. The framing techniques below are designed to get experts out of their default talking points and into more useful territory.
Post-mortem questions project the panelist into a future where something has already happened, and ask them to explain it. “It is 2036, and Germany has lost its position as a manufacturing hub. What were the three decisions in the 2020s that caused this?” This bypasses the natural tendency to hedge and forces a causal narrative.
Best-case scenario questions do the inverse. “Imagine the optimistic version of 2036 for German industry. What does it look like, and what had to go right?” Useful for surfacing what experts secretly hope for, which is often a better signal than what they predict.
Lived-reality questions ground the abstract in the concrete. “You wake up in 2036 and walk into a production site. What has changed since 2026?” Sensory, specific framing pulls people away from buzzwords and toward detail.
Counterfactual futures stress-test dependencies. “If Person X had not remained in office after 2028, what would have shifted in the years that followed?” This exposes which assumptions in an expert’s worldview are actually load-bearing — and which are just decoration.
Running the Interviews
In our case, we run the panel as semi-structured 1:1 interviews. Fully structured interviews are too rigid; you miss the digressions where the real insight often lives. Fully unstructured ones make cross-comparison impossible. Semi-structured is the sweet spot.
We build a question guideline organized around the hypotheses, clustered by theme. Not every expert gets every question — a regulatory scholar and a factory floor manager have different things to offer, and forcing them through identical scripts wastes everyone’s time. But within each cluster, we keep enough overlap across panelists that we can compare answers meaningfully later.
Every interview is recorded and transcribed. This is non-negotiable. Memory distorts, and the analysis phase depends on having the actual words.
Analyzing What You Heard
The analysis is essentially a search for two patterns: similarity and difference.
Where do experts converge? Convergence across a diverse panel — different sectors, hierarchies, geographies — is a strong signal. It does not mean they are right, but it means a particular view of the future is widely held by people who have thought hard about it. That is worth knowing, and worth challenging.
Where do they diverge? This is usually the more interesting finding. Sharp disagreement often maps onto the high-uncertainty zones we identified at the start, and the shape of the disagreement matters. Are the academics aligned against the industry voices? Are the strategists split from the operators? Is it a regional pattern? Each of these tells you something different about what the real fault lines in your topic actually are.
Our recommendation: Read between the lines: sometimes the most important things get mentioned in passing – a throwaway aside, a half-finished thought – and AI-driven analysis tends to miss exactly those moments, so during the interview take notes, especially on the minor mentions that feel quietly major.
The output of a Delphi round is rarely a clean answer. It is a more textured map: clearer on which futures have broad expert backing, clearer on which ones are genuinely contested, and clearer on what would have to change for one trajectory to win out over another.
Why Delphi?
Delphi works because it does something neither pure data analysis nor pure intuition can do on its own. It systematically harvests tacit knowledge — the kind that lives in the heads of people who have spent careers in a field — and structures it well enough that you can compare, contrast, and act on it.
