The Delphi Method

Posted in management by Christopher R. Wirz on Wed Apr 01 2015

The Delphi Method (also called Delphi Technique) is a structured expert judgment process used to achieve consensus on a topic through anonymous, iterative rounds of questionnaires and controlled feedback. Developed by the RAND Corporation in the 1950s for forecasting and decision-making, it's now widely used in project management for estimation, risk assessment, and expert consultation.

Core Characteristics

Key Features:

  • Anonymous participation - Experts don't know who else is participating or what others initially said
  • Iterative rounds - Multiple cycles of questionnaires (typically 2-4 rounds)
  • Controlled feedback - Facilitator summarizes results and shares with group between rounds
  • Statistical aggregation - Results compiled and analyzed quantitatively
  • Consensus building - Process continues until reasonable agreement emerges

The Name: Named after the Oracle of Delphi in ancient Greece, where people sought wisdom and prophecy.

The Process

Standard Delphi Process Flow

Round 1: Initial Input

  1. Facilitator identifies problem/question
  2. Panel of experts selected (typically 10-30 people)
  3. Questionnaire sent to all experts independently
  4. Experts provide estimates, opinions, or forecasts anonymously
  5. Facilitator collects and analyzes responses

Round 2: Controlled Feedback

  1. Facilitator compiles Round 1 results (statistical summary, range, median)
  2. Summary shared with all experts (without identifying who said what)
  3. Experts review group results and rationale for outliers
  4. Experts revise their estimates based on others' input
  5. Experts can explain reasoning for their position
  6. Facilitator collects Round 2 responses

Round 3: Convergence (if needed)

  1. Facilitator shares updated statistical summary
  2. Highlights areas of agreement and disagreement
  3. Provides rationale from experts holding divergent views
  4. Experts make final revisions
  5. Facilitator collects Round 3 responses

Round 4+: Additional Refinement (optional)

  • Continue if consensus hasn't emerged
  • Most Delphi studies reach stability by Round 3
  • Diminishing returns after Round 4

Final Analysis:

  • Facilitator compiles final results
  • Documents level of consensus achieved
  • Reports median, range, and distribution of responses
  • Captures key insights and rationale

Project Management Applications

1. Estimation

Duration Estimation:

  • Ask experts to estimate task durations
  • Achieve consensus on realistic timeframes
  • Reduces individual bias and optimism
  • Produces more reliable estimates than single expert

Cost Estimation:

  • Estimate project or component costs
  • Build budget from expert consensus
  • Identify assumptions affecting estimates
  • Understand range of possible costs

Resource Requirements:

  • Estimate team size needed
  • Determine skill requirements
  • Forecast resource capacity needs

2. Risk Identification and Assessment

Risk Identification:

  • Experts independently identify potential risks
  • Consolidate comprehensive risk list
  • Capture risks individual experts might miss
  • Build complete risk register

Risk Analysis:

  • Experts estimate risk probability and impact
  • Achieve consensus on risk ratings
  • Prioritize risks based on expert judgment
  • Reduce subjective bias in risk assessment

Example Risk Delphi:

  • Round 1: Rate risk probability (1-5 scale)
  • Round 2: Review group median, adjust ratings
  • Round 3: Finalize with rationale for outliers
  • Result: Consensus-based risk priority ranking

3. Requirements Prioritization

Approach:

  • Experts rank or score requirements
  • Identify highest-priority features
  • Build consensus on what's essential
  • Reduce political influence on prioritization

4. Vendor Selection

Process:

  • Experts evaluate vendor proposals
  • Rate vendors on multiple criteria
  • Achieve consensus on best selection
  • Reduce bias from individual preferences

5. Forecasting and Planning

Strategic Planning:

  • Forecast future trends affecting project
  • Predict technology evolution
  • Anticipate market changes
  • Plan for long-term uncertainties

Scenario Planning:

  • Experts propose possible future scenarios
  • Assess likelihood of each scenario
  • Plan responses to different futures

6. Problem-Solving

Complex Decisions:

  • When multiple expert perspectives needed
  • Technical decisions with high uncertainty
  • Novel problems without historical data
  • Controversial issues requiring objectivity

Advantages

1. Eliminates Group Dynamics Issues:

  • No dominant personalities controlling discussion
  • Introverts contribute equally
  • Political hierarchies don't suppress junior experts
  • Reduces groupthink and conformity pressure

2. Reduces Bias:

  • Anonymous participation reduces anchoring
  • Experts can change minds without losing face
  • Peer pressure minimized
  • Focus on merits, not personalities

3. Leverages Distributed Expertise:

  • Geographic barriers don't matter
  • Time zone differences accommodated
  • Experts participate on their schedule
  • Access to broader expert pool

4. Structured and Documented:

  • Clear process and methodology
  • Traceable reasoning and rationale
  • Documented evidence for decisions
  • Defensible results

5. Convergence Toward Consensus:

  • Typically achieves reasonable agreement
  • Outliers have opportunity to influence group
  • Statistical aggregation reduces individual errors
  • Group wisdom emerges

6. Cost-Effective:

  • No need to gather experts in one location
  • No travel costs or time
  • Asynchronous participation
  • Less time than multiple meetings

Limitations and Challenges

1. Time-Consuming:

  • Multiple rounds take weeks or months
  • Each round requires time for response and analysis
  • Slower than single meeting
  • May not work for urgent decisions

2. Requires Expert Commitment:

  • Participants must complete multiple rounds
  • Attrition between rounds
  • Declining response quality in later rounds
  • Needs motivated, engaged experts

3. Facilitator Skill Critical:

  • Proper summarization essential
  • Bias can enter through framing
  • Must present feedback neutrally
  • Quality depends on facilitator competence

4. May Not Reach Full Consensus:

  • Some issues remain divisive
  • Forcing consensus can be artificial
  • May converge on compromise, not optimal solution
  • Outliers may represent important insights

5. Limited Interaction:

  • Experts can't debate directly
  • Misunderstandings harder to resolve
  • Creative synergy reduced
  • Complex reasoning difficult to convey

6. Quality Depends on Expert Selection:

  • Must identify true experts
  • Need diverse perspectives
  • Too homogeneous = narrow thinking
  • Wrong experts = poor results

7. Anonymity Can Be Double-Edged:

  • Reduces accountability
  • May encourage careless responses
  • Experts might not invest sufficient thought
  • Hard to probe reasoning deeply

Best Practices

Expert Selection:

  • Choose recognized subject matter experts
  • Include diverse perspectives and backgrounds
  • Avoid conflicts of interest
  • Aim for 10-30 participants (optimal range)
  • Ensure expertise relevant to specific question

Question Design:

  • Clear, unambiguous questions
  • Specific and focused
  • Quantifiable when possible
  • Avoid leading or biased framing
  • Pilot test with small group

Facilitation:

  • Maintain strict anonymity
  • Provide neutral, complete feedback
  • Share statistical summaries (median, quartiles, range)
  • Include rationale for outlier positions
  • Don't manipulate toward predetermined answer
  • Set clear deadlines and expectations

Feedback Between Rounds:

  • Share distribution of responses (box plots, histograms)
  • Report median and interquartile range
  • Provide anonymous rationale for extreme positions
  • Highlight areas of agreement and disagreement
  • Don't identify individuals or create peer pressure

Stopping Criteria:

  • Predetermined consensus threshold (e.g., 70% within range)
  • Stability across rounds (responses no longer changing)
  • Maximum number of rounds reached (typically 3-4)
  • Diminishing returns evident
  • Time constraints require conclusion

Managing Outliers:

  • Don't force conformity
  • Request explanation of reasoning
  • Share rationale with group
  • Outliers may have unique valid insight
  • Include in final report with context

Documentation:

  • Record all responses anonymously
  • Document process and methodology
  • Explain how consensus was determined
  • Report level of agreement achieved
  • Capture key insights and rationale
  • Note remaining disagreements

Variations

Modified Delphi:

  • Starts with face-to-face meeting to frame issue
  • Then proceeds with anonymous rounds
  • May conclude with group discussion
  • Balances structure with interaction

Real-Time Delphi:

  • Uses software for immediate feedback
  • Participants see updated statistics as others respond
  • Faster than traditional rounds
  • Loses some iteration benefits

Policy Delphi:

  • Designed to explore policy options, not reach consensus
  • Focuses on identifying diverse perspectives
  • Useful when disagreement is expected and valuable
  • Goal is understanding, not agreement

e-Delphi:

  • Conducted entirely online
  • Uses web-based survey tools
  • Automated analysis and feedback
  • Faster turnaround between rounds

Delphi Conference:

  • Combines Delphi with synchronous elements
  • Uses technology for structured group sessions
  • Maintains some anonymity features
  • Faster than pure Delphi

Example: Project Duration Estimate

Scenario: Estimating duration for software development project

Round 1:

  • Question: "How many weeks will it take to complete the XYZ software module?"
  • 20 experts respond independently
  • Results: Range 8-24 weeks, Median 14 weeks, Mean 15.2 weeks

Round 2:

  • Facilitator shares: "Group median is 14 weeks, range 8-24"
  • Includes rationale from extremes:
    • 8 weeks: "Assumes we reuse 60% from existing module"
    • 24 weeks: "Includes integration testing often underestimated"
  • Experts reconsider and resubmit
  • Results: Range 10-18 weeks, Median 13 weeks, Mean 13.8 weeks

Round 3:

  • Facilitator shares: "Median now 13 weeks, range 10-18"
  • Notes: "Integration testing concern widely shared, reuse assumption questioned"
  • Final estimates collected
  • Results: Range 12-16 weeks, Median 14 weeks, Mean 14.1 weeks

Outcome:

  • Consensus estimate: 14 weeks
  • Contingency planning for 12-16 week range
  • Key assumptions documented
  • Risk of integration testing delays noted

When to Use Delphi vs. Other Methods

Use Delphi When:

  • Expert judgment needed but can't meet in person
  • Risk of dominant personalities skewing results
  • Controversial or political issues
  • Time permits multiple rounds (weeks available)
  • Need documented, defensible consensus
  • High uncertainty with no historical data
  • Geographic distribution of experts

Use Alternative Methods When:

  • Urgent decision needed (use expert interviews or single meeting)
  • Group synergy and debate important (use facilitated workshop)
  • Factual data available (use analogous or parametric estimation)
  • Simple question with clear answer (use quick consultation)
  • Team building important (use collaborative planning)

Alternatives:

Method When to Use Instead
Brainstorming Need creative ideas quickly, group synergy valuable
Nominal Group Technique Need consensus in single session, face-to-face possible
Expert Interviews Time-critical, need depth from few experts
Facilitated Workshop Complex problem requiring real-time discussion
Wideband Delphi Estimation with team discussion between rounds
Planning Poker Agile estimation with team in same location

Measuring Consensus

Quantitative Measures:

Interquartile Range (IQR):

  • Difference between 75th and 25th percentile
  • Smaller IQR = greater consensus
  • Target: IQR < 20% of median

Coefficient of Variation:

  • Standard deviation / mean
  • Lower percentage = more agreement
  • Target: < 0.5 (50%)

Stability:

  • Percentage of respondents changing estimates < 10%
  • Indicates convergence achieved

Consensus Threshold:

  • Percentage within predetermined range
  • Example: 70% of responses within ±15% of median

Qualitative Indicators:

  • Rationale for positions aligning
  • Fewer outliers in later rounds
  • Decreasing standard deviation
  • Narrowing range

Integration with Other PM Techniques

Three-Point Estimation:

  • Use Delphi to generate optimistic, most likely, pessimistic estimates
  • Combine with PERT calculations
  • Produces statistically robust estimates

Risk Management:

  • Delphi for qualitative risk assessment
  • Follow with quantitative methods (Monte Carlo)
  • Build comprehensive risk profile

Requirements Prioritization:

  • Use Delphi to score requirements
  • Combine with MoSCoW or Kano model
  • Develop consensus-based priority list

Vendor Evaluation:

  • Delphi for multi-criteria scoring
  • Objective selection based on expert consensus
  • Reduces procurement disputes

Key Concepts

Delphi Method - Structured expert judgment technique achieving consensus through anonymous, iterative questionnaires with controlled feedback.

Facilitator - Person managing Delphi process, distributing questionnaires, analyzing results, and providing feedback without influencing content.

Expert Panel - Group of subject matter experts participating in Delphi study, typically 10-30 individuals.

Anonymity - Experts don't know others' identities or initial responses, reducing bias and groupthink.

Iterative Rounds - Multiple cycles of questionnaires (typically 2-4) allowing experts to reconsider based on group feedback.

Controlled Feedback - Facilitator's summary of group responses shared between rounds to inform experts without revealing identities.

Consensus - Reasonable agreement among experts, measured quantitatively or by stability of responses.

Convergence - Process of expert estimates moving closer together across rounds.

Outlier - Response significantly different from group median or outside typical range.

Stability - Point where expert responses stop changing significantly between rounds.

Median - Middle value when responses arranged in order; preferred measure for Delphi consensus.

Interquartile Range (IQR) - Difference between 75th and 25th percentile; measures consensus level.

Groupthink - Phenomenon where group pressure leads to conformity and poor decisions; Delphi designed to prevent this.

Wideband Delphi - Variation including group discussion between estimation rounds; common in agile for planning poker.

Modified Delphi - Variation starting with face-to-face meeting before anonymous rounds.

Real-Time Delphi - Software-enabled variation providing immediate statistical feedback as experts respond.

Policy Delphi - Variation exploring diverse perspectives rather than forcing consensus.