Delphi Method and Expert Foresight

Last Updated June 2, 2026

The Delphi Method is a structured foresight technique that uses repeated rounds of expert judgment to explore uncertain futures, clarify assumptions, compare expectations, identify areas of agreement and disagreement, and support long-range decision-making when evidence is incomplete. It is especially useful when the future cannot be estimated by simple trend extrapolation, when expert knowledge is dispersed across fields, and when uncertainty is too complex for a single forecast, model, or committee discussion.

Unlike ordinary expert panels, the Delphi Method does not rely on a single meeting or open debate. It uses iterative questioning, controlled feedback, anonymity or partial anonymity, statistical summaries, qualitative reasoning, and repeated reflection. Experts respond individually, review summarized responses from the group, revise their judgments if they choose, and help build a clearer picture of uncertainty over time.

At its best, Delphi is not a method for manufacturing consensus. It is a disciplined way to make expert judgment more transparent, comparable, and revisable. It can show where experts converge, where they disagree, where assumptions differ, where uncertainty remains high, and which future developments deserve deeper monitoring. This makes it valuable for strategic foresight, public policy, technology assessment, sustainability planning, public health, climate adaptation, governance, research prioritization, and long-range institutional strategy.

The method is particularly important because future-oriented decisions often must be made before complete evidence exists. Governments, institutions, communities, researchers, and organizations may need to prepare for emerging technologies, climate risks, public-health threats, labor changes, infrastructure pressures, geopolitical shifts, or ecological thresholds before definitive data are available. Delphi offers a structured way to use expertise without pretending that expertise eliminates uncertainty.

What Is the Delphi Method?

The Delphi Method is an iterative expert-consultation process designed to gather, compare, refine, and interpret expert judgments about uncertain questions. It is commonly used when reliable evidence is incomplete, when future developments are uncertain, when knowledge is distributed across multiple experts, and when open group discussion may produce bias, dominance, conformity, or premature agreement.

The method usually begins with a research or foresight question. A panel of experts is selected. Experts answer a first-round questionnaire individually. Their responses are summarized by facilitators. The summary is then shared back to the panel, often including statistical distributions, anonymized reasoning, areas of agreement, areas of disagreement, and uncertainty ranges. Experts then respond again in later rounds, revising or defending their judgments in light of group feedback.

The goal may be consensus, but it does not have to be. In many serious foresight settings, the most valuable outcome is not agreement but structured clarity: where experts converge, where they diverge, why they diverge, which assumptions drive disagreement, and what evidence or signals would change their views.

The Delphi Method was developed in the mid-twentieth century in the context of long-range forecasting and strategic analysis, but it has since been adapted across many fields. It has been used in technology forecasting, health research, education, environmental planning, public policy, security studies, sustainability, organizational strategy, and research-priority setting.

Delphi is best understood as a structured learning process for expert judgment under uncertainty.

Back to top ↑

What Is Expert Foresight?

Expert foresight is the use of specialized knowledge, professional judgment, and structured deliberation to explore possible future developments. It recognizes that many future-relevant questions cannot be answered by historical data alone. Some questions involve emerging technologies, slow ecological change, institutional fragility, geopolitical uncertainty, public-health preparedness, social transformation, legal adaptation, or values-based governance choices that have no simple historical precedent.

Expert foresight does not mean that experts can predict the future with certainty. It means that expert judgment can help identify drivers, uncertainties, constraints, thresholds, plausible developments, early indicators, and decision implications. Experts may understand technical feasibility, institutional barriers, scientific uncertainty, legal constraints, operational realities, ethical tensions, and implementation risks that are not visible in aggregate data.

Good expert foresight is structured, plural, and transparent. It does not ask one expert to pronounce the future. It compares multiple expert perspectives, examines assumptions, clarifies uncertainty, documents disagreement, and connects judgment to decision contexts. Delphi is one of the most important methods for doing this because it reduces some of the weaknesses of ordinary expert consultation.

Expert foresight is strongest when it includes several kinds of expertise. Scientific expertise matters, but so do technical, institutional, community, professional, legal, ecological, ethical, and lived expertise. A climate adaptation Delphi, for example, should not rely only on modelers. It may also need infrastructure planners, public-health practitioners, emergency managers, community organizers, ecologists, labor representatives, housing experts, and people directly exposed to climate risk.

Expert foresight is not a substitute for evidence. It is a way to reason responsibly where evidence, uncertainty, judgment, and future consequence meet.

Back to top ↑

Why the Delphi Method Matters

The Delphi Method matters because many strategic decisions must be made under deep uncertainty. Institutions often need to act before trends fully mature, before data are complete, before new technologies stabilize, before ecological thresholds are crossed, before public-health risks become crises, or before policy windows close. In such contexts, waiting for perfect evidence can be itself a form of failure.

Delphi helps decision-makers use expert judgment without relying on informal authority or one-time opinion gathering. It provides a structured process for comparing judgments, revising assumptions, and documenting uncertainty. This is especially important when expert views differ sharply. Rather than hiding disagreement behind a consensus statement, Delphi can show where disagreement is stable and why it matters.

In futures thinking, Delphi can support scenario planning, horizon scanning, weak signal analysis, trend interpretation, research prioritization, technology assessment, policy design, and strategic planning. It can identify plausible development timelines, estimate uncertainty ranges, compare risk priorities, assess feasibility, rank emerging issues, evaluate policy options, and test assumptions.

Its value is not that experts are always right. Experts can be biased, overconfident, discipline-bound, or influenced by professional incentives. Delphi matters because it creates a process in which expert judgment becomes more visible, comparable, and open to revision.

Delphi matters because it turns expert judgment from a black box into a structured, revisable, and interpretable foresight process.

Back to top ↑

Delphi vs Expert Panels, Surveys, and Forecasts

The Delphi Method is sometimes confused with surveys, expert panels, workshops, or forecasts. It can include elements of all four, but its distinctive value lies in combining structured expert input with iteration, feedback, and revision. A survey captures responses at one point in time. A panel discussion captures interaction, but it may be shaped by hierarchy and group dynamics. A forecast estimates future outcomes, but it may not expose the assumptions behind expert judgment. Delphi sits between these methods.

The distinction is important because Delphi should not be treated as a generic consultation exercise. A poorly designed Delphi can become merely a repeated survey. A strong Delphi uses iteration to improve reasoning, expose assumptions, clarify disagreement, and support decision-relevant learning.

Delphi is not just expert opinion. It is expert judgment structured through feedback, revision, and uncertainty awareness.

Back to top ↑

The Delphi process is strongest when it treats expert judgment as something to be examined, compared, revised, and interpreted rather than simply collected.

Back to top ↑

Expert Judgment and Uncertainty

Expert judgment is valuable because experts understand patterns, constraints, mechanisms, and failure modes that non-specialists may miss. But expert judgment is not infallible. Experts may be overconfident, discipline-bound, influenced by professional incentives, anchored to familiar models, or slow to recognize disruptive change outside their field. Delphi’s value lies in structuring expert judgment so that these limitations can be reduced, surfaced, or at least documented.

In foresight, uncertainty takes several forms. There may be uncertainty about facts, uncertainty about mechanisms, uncertainty about timing, uncertainty about interaction effects, uncertainty about social response, uncertainty about policy implementation, and uncertainty about values. Delphi can help clarify which kind of uncertainty is present.

For example, experts may agree that a technology is technically feasible but disagree about adoption timing. They may agree that climate risk is increasing but disagree about institutional capacity to adapt. They may agree that a public-health intervention is valuable but disagree about workforce feasibility. They may agree that AI governance is necessary but disagree about whether accountability should be built through regulation, procurement, professional standards, or public oversight.

A strong Delphi study does not erase uncertainty. It classifies uncertainty so that decisions can be made more responsibly.

Back to top ↑

Anonymity, Feedback, and Iteration

Three features distinguish Delphi from many other expert consultation methods: anonymity, controlled feedback, and iteration. These features are designed to reduce some of the weaknesses of group judgment.

Anonymity helps protect independent judgment. In open meetings, experts may defer to senior figures, dominant personalities, institutional representatives, or perceived consensus. Anonymity reduces the pressure to agree, lowers reputational risk, and can make it easier for experts to express minority views.

Controlled feedback allows experts to learn from the group without being overwhelmed by debate. Instead of a free-form discussion, facilitators summarize the distribution of responses and the reasoning behind them. Experts can see where their views sit relative to the panel, but they are not forced to conform.

Iteration creates time for reflection. Experts may revise their judgments after seeing group feedback, or they may choose to maintain their position and explain why. Either result is useful. Revision may show learning and convergence. Stable dissent may reveal deeper uncertainty, disciplinary differences, or important minority warnings.

Delphi is valuable because it slows down expert judgment enough to make learning possible.

Back to top ↑

Consensus, Dissensus, and Stability

Many Delphi studies aim for consensus, but consensus should not be treated as the only valid outcome. In futures thinking, disagreement may be strategically meaningful. If experts continue to disagree after several rounds, that disagreement may reveal alternative futures, unresolved uncertainty, disciplinary differences, competing values, missing evidence, or genuinely unstable system conditions.

Consensus can be measured in several ways: percentage agreement, interquartile range, standard deviation, stability of ratings across rounds, ranking convergence, or qualitative agreement. But consensus criteria should be defined before analysis, not invented afterward. Otherwise the method may falsely present uncertain judgments as settled findings.

Dissensus is not failure. It may be one of the most important outcomes of a Delphi study. A panel that fails to converge on the timing of climate migration, the governance risks of AI, the feasibility of energy transition, or the likelihood of public-health system failure is telling decision-makers something important: the future is not yet well understood, and strategy should remain adaptive.

Stability is another useful concept. If expert ratings do not converge but also stop changing across rounds, the study may have reached stable disagreement. At that point, additional rounds may not produce value. The appropriate output may be an uncertainty map rather than a consensus statement.

In expert foresight, knowing where experts disagree may be as important as knowing where they agree.

Back to top ↑

Major Variants of the Delphi Method

The Delphi Method has evolved into several variants. Different versions serve different purposes, and the choice of variant should depend on the foresight question, decision context, time available, panel composition, and desired output.

These variants matter because Delphi should not be forced into a consensus-oriented form when the question is political, ethical, or deeply uncertain. A policy Delphi may be more useful than a classical Delphi when public values are contested. A participatory Delphi may be more legitimate when affected communities hold knowledge that formal experts lack. A real-time Delphi may be useful when fast iteration is necessary.

The method should fit the uncertainty. A consensus-seeking Delphi is not always the right tool.

Back to top ↑

Expert Selection, Diversity, and Panel Design

Expert selection is one of the most important decisions in any Delphi study. The panel defines the knowledge base. If the panel is narrow, the results will be narrow. If the panel excludes affected communities, implementation practitioners, or minority perspectives, the study may reproduce institutional blind spots while appearing methodologically rigorous.

Expertise should be defined in relation to the question. A technology foresight Delphi may require engineers, policy experts, ethicists, labor specialists, regulators, procurement officials, civil-rights advocates, and affected users. A climate adaptation Delphi may require climate scientists, infrastructure planners, public-health experts, emergency managers, community organizers, housing specialists, ecologists, and finance experts. A public-health Delphi may require clinicians, epidemiologists, care workers, data specialists, community health workers, disability advocates, and public administrators.

Panel diversity is not merely symbolic. It changes what the study can see. Different experts notice different risks, constraints, values, and future pathways. Homogeneous expert panels may converge quickly, but convergence within a narrow field may be less valuable than structured disagreement across a more diverse one.

A Delphi study is only as strong as the knowledge ecology it creates.

Back to top ↑

Questionnaire Design and Foresight Quality

Questionnaire design determines the quality of Delphi output. Poorly framed questions produce vague, biased, or unusable results. Good questions are clear, decision-relevant, appropriately scoped, and designed to reveal uncertainty rather than conceal it.

Delphi questions may ask experts to rate likelihood, importance, feasibility, impact, urgency, desirability, uncertainty, or priority. They may ask for expected timelines, conditions for success, barriers, early indicators, policy options, research needs, scenario drivers, or criteria for monitoring. They may also invite qualitative explanations, assumptions, dissenting arguments, and conditional reasoning.

The best Delphi questionnaires do not rely only on numerical rating scales. Numbers without reasoning can create false precision. Qualitative explanations are essential because they reveal why experts hold different judgments. Two experts may rate an issue as equally important for entirely different reasons. Another expert may give an outlier rating because they see a risk that others missed.

A Delphi questionnaire should be designed to reveal judgment, uncertainty, and reasoning—not merely collect numbers.

Back to top ↑

Power, Legitimacy, and Whose Expertise Counts

Delphi studies are often presented as technical exercises, but they are also shaped by power. Who is invited as an expert? Which forms of knowledge count? Who frames the questions? Which uncertainties are considered legitimate? Which futures are included or excluded? These choices influence the outcome before the first round begins.

Formal expertise is important, but it can be incomplete. In many future-oriented problems, people closest to risk have knowledge that formal experts may lack. Workers understand how technologies alter labor processes. Tenants understand housing insecurity before market reports fully describe it. Indigenous communities may observe ecological change across generations. Disabled people may detect digital exclusion before technology governance frameworks recognize it. Youth may understand intergenerational stakes that current institutions discount.

A more legitimate Delphi process does not abandon expertise. It broadens the definition of expertise to include relevant knowledge wherever it resides. This is especially important for public-interest foresight, sustainability transitions, climate adaptation, health equity, technology governance, and decisions affecting marginalized communities.

Power also shapes consensus. A Delphi panel may converge around dominant assumptions if the panel is drawn from dominant institutions. Apparent consensus may reflect shared blind spots rather than reliable judgment. For this reason, Delphi reports should document panel composition and interpret consensus critically.

A Delphi study is more trustworthy when it treats expertise as plural, situated, and accountable.

Back to top ↑

Applications of Delphi and Expert Foresight

The Delphi Method is used across many fields because structured expert judgment is useful wherever uncertainty is high, evidence is incomplete, and decisions carry long-term consequences. Its value depends on the quality of the question, panel, facilitation, feedback, and interpretation.

In each domain, Delphi works best when it is connected to action. A study that produces expert rankings but no decision pathway may become another report. A study that links expert judgment to monitoring, scenario planning, strategy, policy, or implementation can become a practical foresight tool.

Back to top ↑

Strengths and Limitations

The Delphi Method has several strengths. It enables structured expert judgment under uncertainty. It reduces some of the distortions of open group discussion. It supports reflection across rounds. It can combine quantitative ratings with qualitative reasoning. It can show both agreement and disagreement. It is adaptable across fields and useful when historical data are insufficient.

But Delphi also has limitations. Results depend heavily on panel selection, question design, facilitator interpretation, response rates, consensus criteria, and the quality of feedback. Delphi can create false authority if expert judgments are presented without uncertainty or limitations. It can also produce artificial consensus if experts feel pressured to align with group summaries. If the panel is narrow, the method may reproduce dominant assumptions while appearing rigorous.

Delphi is strongest when it is transparent about uncertainty and modest about authority.

Back to top ↑

A Practical Delphi Workflow

A practical Delphi workflow should move from problem framing to panel design, questionnaire construction, iterative rounds, feedback, analysis, and decision translation. The process should be designed as a structured learning system rather than a ritualized search for consensus.

A practical Delphi should not end with a table of ratings. It should explain what the ratings mean, why experts differed, what assumptions matter, what evidence would change judgments, and how decision-makers should use the findings.

The purpose of Delphi is not to replace judgment with procedure. It is to make judgment more disciplined, transparent, and useful.

Back to top ↑

Mathematical Lens: Expert Judgment, Convergence, and Uncertainty

A Delphi panel can be represented as a set of expert judgments across rounds:

A simple consensus measure can use dispersion around the median:

Change across rounds can be represented as:

Expert uncertainty can be represented using a confidence interval or rating range:

A foresight priority score can combine likelihood, impact, uncertainty, and urgency:

These equations do not make Delphi objective in a mechanical sense. They clarify how judgments, dispersion, revision, uncertainty, and priority can be tracked across rounds.

Back to top ↑

Computational Modeling for Delphi Foresight

Computational tools can support Delphi studies by managing panel data, tracking changes across rounds, calculating consensus metrics, summarizing qualitative themes, scoring priorities, and producing transparent outputs. They should not replace facilitation, careful question design, expert reasoning, or ethical judgment. Their purpose is to make the process auditable and easier to interpret.

A useful computational Delphi workflow may include:

• Panel register: anonymized expert IDs, expertise categories, sector representation, and recruitment criteria.
• Round response data: ratings, rankings, probability estimates, timeline estimates, confidence ranges, and qualitative rationales.
• Consensus metrics: medians, interquartile ranges, standard deviations, agreement percentages, and stability across rounds.
• Dissensus analysis: identification of persistent disagreement, polarized clusters, and outlier rationales.
• Uncertainty tracking: comparison of confidence ranges and judgment revision across rounds.
• Theme coding: structured qualitative analysis of explanations, assumptions, barriers, and early indicators.
• Priority scoring: decision-relevant ranking of risks, opportunities, policies, capabilities, or research needs.
• Foresight translation: conversion of Delphi results into scenarios, monitoring indicators, policy options, or strategic recommendations.

Computational Delphi analysis should document assumptions carefully. It should identify the number of participants per round, attrition, missing data, consensus criteria, scoring formulas, and limitations. A chart or score without methodological transparency can create false confidence.

The goal is not to automate expert foresight. The goal is to make expert judgment traceable, comparable, and accountable.

Back to top ↑

Advanced R Workflow: Delphi Consensus and Disagreement Profiles

The R workflow below creates a stylized Delphi dataset across three rounds. It compares median judgments, interquartile ranges, stability, and priority scores. It is designed as an evergreen example of how Delphi analysis can distinguish convergence from persistent disagreement.

# ------------------------------------------------------------
# R Workflow: Delphi Consensus and Disagreement Profiles
# Purpose:
#   Analyze stylized Delphi ratings across rounds using
#   median, interquartile range, stability, and priority scoring.
#
# Optional dependency:
#   install.packages(c("tidyverse"))
# ------------------------------------------------------------

library(tidyverse)

responses <- tibble(
  expert_id = rep(paste0("E", 1:12), times = 3),
  round = rep(1:3, each = 12),
  issue = "Public AI Accountability Before Deployment",
  likelihood = c(
    0.62, 0.68, 0.55, 0.71, 0.48, 0.73, 0.60, 0.66, 0.52, 0.70, 0.58, 0.64,
    0.66, 0.70, 0.60, 0.74, 0.54, 0.76, 0.64, 0.68, 0.58, 0.72, 0.62, 0.66,
    0.68, 0.71, 0.63, 0.75, 0.56, 0.77, 0.66, 0.69, 0.60, 0.73, 0.64, 0.67
  ),
  impact = c(
    0.84, 0.88, 0.78, 0.90, 0.72, 0.91, 0.82, 0.86, 0.76, 0.89, 0.80, 0.85,
    0.86, 0.89, 0.80, 0.91, 0.76, 0.92, 0.84, 0.87, 0.78, 0.90, 0.82, 0.86,
    0.87, 0.90, 0.82, 0.92, 0.78, 0.93, 0.85, 0.88, 0.80, 0.91, 0.83, 0.87
  ),
  urgency = c(
    0.74, 0.82, 0.68, 0.86, 0.60, 0.88, 0.72, 0.80, 0.64, 0.84, 0.70, 0.76,
    0.78, 0.84, 0.72, 0.88, 0.66, 0.90, 0.76, 0.82, 0.68, 0.86, 0.74, 0.80,
    0.80, 0.85, 0.74, 0.89, 0.68, 0.91, 0.78, 0.83, 0.70, 0.87, 0.76, 0.81
  )
)

round_summary <- responses %>%
  group_by(round, issue) %>%
  summarise(
    median_likelihood = median(likelihood),
    iqr_likelihood = IQR(likelihood),
    median_impact = median(impact),
    iqr_impact = IQR(impact),
    median_urgency = median(urgency),
    iqr_urgency = IQR(urgency),
    priority_score = 0.30 * median_likelihood +
      0.40 * median_impact +
      0.30 * median_urgency,
    .groups = "drop"
  )

stability <- round_summary %>%
  arrange(round) %>%
  mutate(
    likelihood_shift = abs(median_likelihood - lag(median_likelihood)),
    impact_shift = abs(median_impact - lag(median_impact)),
    urgency_shift = abs(median_urgency - lag(median_urgency)),
    priority_shift = abs(priority_score - lag(priority_score))
  )

print(round_summary)
print(stability)

responses_long <- responses %>%
  pivot_longer(
    cols = c(likelihood, impact, urgency),
    names_to = "dimension",
    values_to = "rating"
  )

ggplot(responses_long, aes(x = factor(round), y = rating)) +
  geom_boxplot() +
  facet_wrap(~ dimension) +
  labs(
    title = "Delphi Rating Distributions Across Rounds",
    x = "Round",
    y = "Expert Rating"
  ) +
  theme_minimal(base_size = 12)

ggplot(round_summary, aes(x = round, y = priority_score)) +
  geom_line(linewidth = 1.2) +
  geom_point(size = 3) +
  labs(
    title = "Delphi Priority Score Across Rounds",
    x = "Round",
    y = "Priority Score"
  ) +
  theme_minimal(base_size = 12)

dir.create("outputs", showWarnings = FALSE)

write_csv(responses, "outputs/delphi_round_responses.csv")
write_csv(round_summary, "outputs/delphi_round_summary.csv")
write_csv(stability, "outputs/delphi_stability_summary.csv")

This workflow shows how a Delphi process can be analyzed without reducing it to a single consensus number. Dispersion, stability, and priority shifts all matter.

Back to top ↑

Advanced Python Workflow: Simulating Iterative Delphi Rounds

The Python workflow below simulates expert judgments across Delphi rounds. It models partial convergence, persistent uncertainty, and expert-specific resistance to revision. It is useful for showing why Delphi should be interpreted as a learning process rather than a mechanical consensus machine.

# ------------------------------------------------------------
# Python Workflow: Simulating Iterative Delphi Rounds
# Purpose:
#   Simulate expert judgments across rounds using controlled
#   feedback, partial convergence, uncertainty, and resistance
#   to revision.
#
# Optional dependencies:
#   pip install pandas numpy matplotlib
# ------------------------------------------------------------

from pathlib import Path

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

OUTPUT_DIR = Path("outputs")
OUTPUT_DIR.mkdir(exist_ok=True)

rng = np.random.default_rng(42)

experts = [f"E{i:02d}" for i in range(1, 21)]
rounds = [1, 2, 3, 4]

initial_judgments = rng.normal(loc=0.62, scale=0.16, size=len(experts))
initial_judgments = np.clip(initial_judgments, 0.05, 0.95)

resistance = rng.uniform(0.25, 0.75, size=len(experts))
confidence = rng.uniform(0.45, 0.90, size=len(experts))

rows = []

current = initial_judgments.copy()

for round_number in rounds:
    median_feedback = float(np.median(current))
    iqr_feedback = float(np.quantile(current, 0.75) - np.quantile(current, 0.25))

    for index, expert in enumerate(experts):
        rows.append({
            "expert_id": expert,
            "round": round_number,
            "judgment": current[index],
            "confidence": confidence[index],
            "resistance_to_revision": resistance[index],
            "round_median_feedback": median_feedback,
            "round_iqr_feedback": iqr_feedback
        })

    if round_number != rounds[-1]:
        adjustment = (median_feedback - current) * (1 - resistance) * 0.55
        noise = rng.normal(loc=0, scale=0.025, size=len(experts))
        current = np.clip(current + adjustment + noise, 0.05, 0.95)

df = pd.DataFrame(rows)

summary = (
    df.groupby("round")["judgment"]
    .agg(
        median_judgment="median",
        mean_judgment="mean",
        lower_quartile=lambda x: np.quantile(x, 0.25),
        upper_quartile=lambda x: np.quantile(x, 0.75),
        standard_deviation="std"
    )
    .reset_index()
)

summary["interquartile_range"] = summary["upper_quartile"] - summary["lower_quartile"]
summary["median_shift"] = summary["median_judgment"].diff().abs()

print("\nDelphi round summary:")
print(summary)

df.to_csv(OUTPUT_DIR / "simulated_delphi_responses.csv", index=False)
summary.to_csv(OUTPUT_DIR / "simulated_delphi_round_summary.csv", index=False)

plt.figure(figsize=(10, 6))
for expert in experts:
    subset = df[df["expert_id"] == expert]
    plt.plot(
        subset["round"],
        subset["judgment"],
        marker="o",
        linewidth=1,
        alpha=0.45
    )

plt.xlabel("Round")
plt.ylabel("Expert Judgment")
plt.title("Simulated Delphi Expert Judgments Across Rounds")
plt.tight_layout()
plt.savefig(OUTPUT_DIR / "delphi_judgment_paths.png", dpi=150)
plt.close()

plt.figure(figsize=(10, 6))
plt.plot(summary["round"], summary["interquartile_range"], marker="o", linewidth=1.8)
plt.xlabel("Round")
plt.ylabel("Interquartile Range")
plt.title("Delphi Dispersion Across Rounds")
plt.tight_layout()
plt.savefig(OUTPUT_DIR / "delphi_dispersion_by_round.png", dpi=150)
plt.close()

plt.figure(figsize=(10, 6))
plt.plot(summary["round"], summary["median_judgment"], marker="o", linewidth=1.8)
plt.xlabel("Round")
plt.ylabel("Median Judgment")
plt.title("Delphi Median Judgment Across Rounds")
plt.tight_layout()
plt.savefig(OUTPUT_DIR / "delphi_median_by_round.png", dpi=150)
plt.close()

This workflow demonstrates a central lesson of Delphi foresight: convergence can occur, but it is not guaranteed; disagreement can remain meaningful; and stability should be interpreted alongside reasoning, panel composition, and uncertainty.

Back to top ↑

GitHub Repository

The companion repository for this article contains computational examples for Delphi panel design, iterative expert judgment, consensus metrics, dissensus analysis, uncertainty ranges, priority scoring, expert-foresight reporting, and scenario/policy translation.

Back to top ↑

Why This Matters

The Delphi Method matters because future-oriented decisions often require structured judgment before certainty is available. Climate adaptation, AI governance, public-health preparedness, infrastructure resilience, sustainability transitions, research priorities, education systems, and institutional strategy all involve uncertain developments that cannot be reduced to historical extrapolation alone.

Expert foresight is necessary, but it must be handled carefully. Expertise can clarify uncertainty, but it can also reproduce blind spots. Consensus can guide action, but it can also conceal power. Disagreement can be frustrating, but it may reveal where strategy should remain adaptive. Delphi provides a disciplined method for working through these tensions.

At its best, Delphi does not pretend that experts can see the future. It helps institutions learn from expert judgment while documenting uncertainty, dissent, assumptions, and the limits of knowledge. It creates a structured space between ignorance and overconfidence.

The Delphi Method is therefore not a tool for eliminating uncertainty. It is a tool for making expert judgment more transparent, plural, and useful in the face of uncertainty.

Back to top ↑

Related Articles

• Futures Thinking
• What Is Futures Thinking?
• Forecasting, Foresight, and Futures Studies
• Futures Literacy and Anticipatory Capacity
• Possible, Plausible, Probable, and Preferable Futures
• Scenario Planning
• Strategic Foresight Methods
• Trend Analysis and Megatrends
• Horizon Scanning
• Weak Signals and Early Indicators
• Backcasting and Strategic Planning
• Systems Modeling
• Resilience Thinking

Further Reading

• Dalkey, N. and Helmer, O. (1963) ‘An experimental application of the Delphi method to the use of experts’, Management Science, 9(3), pp. 458–467.
• Gordon, T.J. (1994) The Delphi Method. Washington, DC: United Nations University Millennium Project. Available at: Millennium Project.
• Häder, M. (2023) Delphi Method. Cham: Springer.
• Helmer, O. (1967) Analysis of the Future: The Delphi Method. Santa Monica: RAND Corporation. Available at: RAND Corporation.
• Hines, A. and Bishop, P. (2015) Thinking About the Future: Guidelines for Strategic Foresight. 2nd edn. Houston: Hinesight.
• Landeta, J. (2006) ‘Current validity of the Delphi method in social sciences’, Technological Forecasting and Social Change, 73(5), pp. 467–482.
• Linstone, H.A. and Turoff, M. (eds) (2002) The Delphi Method: Techniques and Applications. Newark: New Jersey Institute of Technology. Available at: NJIT.
• Okoli, C. and Pawlowski, S.D. (2004) ‘The Delphi method as a research tool: an example, design considerations and applications’, Information & Management, 42(1), pp. 15–29.
• Rowe, G. and Wright, G. (1999) ‘The Delphi technique as a forecasting tool: issues and analysis’, International Journal of Forecasting, 15(4), pp. 353–375.
• Voros, J. (2003) ‘A generic foresight process framework’, Foresight, 5(3), pp. 10–21. Available at: Emerald.

Back to top ↑

References

• Dalkey, N. and Helmer, O. (1963) ‘An experimental application of the Delphi method to the use of experts’, Management Science, 9(3), pp. 458–467.
• Gordon, T.J. (1994) The Delphi Method. Washington, DC: United Nations University Millennium Project. Available at: Millennium Project.
• Government Office for Science (2024) The Futures Toolkit: Tools for Futures Thinking and Foresight Across UK Government. London: Government Office for Science. Available at: UK Government.
• Government Office for Science (2025) A Brief Guide to Futures Thinking and Foresight. London: Government Office for Science. Available at: UK Government.
• Häder, M. (2023) Delphi Method. Cham: Springer.
• Helmer, O. (1967) Analysis of the Future: The Delphi Method. Santa Monica: RAND Corporation. Available at: RAND Corporation.
• Landeta, J. (2006) ‘Current validity of the Delphi method in social sciences’, Technological Forecasting and Social Change, 73(5), pp. 467–482.
• Linstone, H.A. and Turoff, M. (eds) (2002) The Delphi Method: Techniques and Applications. Newark: New Jersey Institute of Technology. Available at: NJIT.
• Okoli, C. and Pawlowski, S.D. (2004) ‘The Delphi method as a research tool: an example, design considerations and applications’, Information & Management, 42(1), pp. 15–29.
• Organisation for Economic Co-operation and Development (OECD) (no date) Strategic Foresight. Available at: OECD.
• Organisation for Economic Co-operation and Development Observatory of Public Sector Innovation (OECD OPSI) (no date) Futures & Foresight. Available at: OECD OPSI.
• Organisation for Economic Co-operation and Development (OECD) (2021) Strategic Foresight for Better Policies: Building Effective Governance in the Face of Uncertain Futures. Paris: OECD Publishing. Available at: OECD.
• Organisation for Economic Co-operation and Development (OECD) (2025) Foresight Toolkit for Resilient Public Policy. Paris: OECD Publishing. Available at: OECD.
• Rowe, G. and Wright, G. (1999) ‘The Delphi technique as a forecasting tool: issues and analysis’, International Journal of Forecasting, 15(4), pp. 353–375.
• United Nations Development Programme (UNDP) (2018) Foresight Manual: Empowered Futures for the 2030 Agenda. New York: UNDP. Available at: UNDP.
• United Nations Educational, Scientific and Cultural Organization (UNESCO) (no date) Futures Literacy & Foresight. Available at: UNESCO.
• Voros, J. (2003) ‘A generic foresight process framework’, Foresight, 5(3), pp. 10–21. Available at: Emerald.

Back to top ↑