Form
Delphi study report with question, panel description (anonymous), method per round, statistical aggregation, range and consensus measure, qualitative rationale, outlier positions, limitations, and implications for strategy or forecast.
RunsheetDecision Making
Delphi Method
ComplexityHigh
Time1-4 Wochen
Participants6-30 Experten
FormatAsync
MaturityCanonical
01
Preparation
What needs to be ready before start
Materials
Survey tool (Google Forms, Typeform, LimeSurvey); expert data (list with contact information and expertise); aggregation tool (spreadsheet, statistics package); communication channel (e-mail, dedicated portal); round plan.
People / roles
Study owner (method owner for Delphi); 6-30 experts as panel; one method specialist with statistical experience; for large studies a documentation team.
Pre-read
Clear research question or question set; definition of expert criteria; anonymization commitment; planned number of rounds (typically 2-4); scales or response formats; time window per round (typically 1-2 weeks).
Time needed
1-4 weeks
Setup
Round template prepared. Panel list with individual codes for anonymity. Aggregation format defined (median, quartile, consensus measure). Communication plan between rounds.
02
Core question
The one question this method answers
What range or which consensus emerges when experts give iterative, anonymous estimates on an uncertain future question?
03
Flow
Marker: Phase
| Step | Duration | Action | Hint |
|---|---|---|---|
1Phase 1: Define panel and question | 1 week | Define expert criteria (background, experience). Recruit panel of 6-30 people. Formulate the question precisely with scales or structured answer format. | A heterogeneous panel with experts from different perspectives yields more robust results than a homogeneous one. 30+ experts increase drop-out risk without much extra value. |
2Phase 2: Round 1 — first estimates | 1-2 weeks | Send survey to the panel, answer anonymously. For qualitative questions, optional open rationale. Send reminder after 7 days. | Anonymity is the core of the method. If identity leaks through, it destroys the effect. Choose tools and process to help experts feel safe. |
3Phase 3: Aggregation | 3-7 days | Aggregate responses statistically (median, quartiles, distribution). For qualitative answers, cluster themes. Collect rationales anonymously. | Aggregation must stay neutral. If favorite answers are reinforced at this step, the panel is influenced. The method specialist should validate aggregation, not the owner alone. |
4Phase 4: Round 2 (and possibly more) | 1-2 weeks per round | Return aggregated results to the panel with rationale excerpts. Experts review and either revise or defend their answers. | Outliers are explicitly asked to justify their position. Consensus often appears in rounds 2-3. More than 4 rounds increases fatigue without adding insight. |
5Phase 5: Close and report | 1-2 weeks | Evaluate consensus measure (for example interquartile range, agreement rate). Final report with ranges, rationales, outlier arguments, method description. | If consensus is not reached, the result can be an honest range, which is often more valuable than forced consensus. Reflect on how the study related to panel experience. |
04
Artifact
What comes out at the end
Versioning / ownership
Own report per study with date and question in title. Follow-up studies (for example annual) should link prior results for trend comparison.
Tool alternatives
- LimeSurvey or SurveyMonkey for surveying
- Welphi or Delphi2 as specialized platforms
- Google Forms + Sheets for lightweight studies
- Excel or R scripts for aggregation
delphi-method-working-template.md
Compact working template for the Delphi Method with anonymous rounds, feedback, convergence, and recommendation.
Delphi Method Working Template
Goal
Build a shared assessment through anonymous expert input across repeated rounds.
Context
What question should be answered, and which kind of expertise is needed?
Input
- Research question:
- Panel members:
- Source material:
- Round schedule:
Working area
- Round 1 statements:
- Round 2 feedback:
- Convergence signals:
- Remaining disagreements:
- Recommendation draft:
Output artifacts
- Round summaries:
- Consensus statement:
- Recommendation:
Open questions
- ...
Decision / next step
Owner, date, and success signal.
05
Example output
Concrete filled scenario, fictional example
delphi-method-beispiel.md
Concrete filled scenario, fictional example
Delphi Study — Adoption of autonomous AI agents in DACH tax advisory 2026-2029 (completed 12.06.2026)
Question: What share of DACH tax firms will use autonomous AI agents for document capture in production by Q4/2028?
Panel: 14 experts (8 tax professionals with technical affinity, 3 software vendors, 3 trade associations). Anonymous with codes.
Round 1 (n=14): Median 18%, quartiles 8%-32%. High spread.
Round 2 (n=13): Median 22%, quartiles 14%-30%. One outlier at 5% cited regulatory constraints. One outlier at 45% cited generational leadership change.
Round 3 (n=12): Median 24%, quartiles 18%-28%. Consensus measure: interquartile range 10 percentage points (before: 24).
Result: Consensus at 18-28% by Q4/2028, with two documented drivers: legal reform and leadership transitions. Outlier positions are retained as worst- and best-case scenarios.
Implications for Sabine, solo tax advisor: Market share in 2027 remains below mainstream; AI integration should create differentiation rather than follow baseline standards.
06
Pitfalls
Recognize symptoms and steer against them
Trap
Anonymity is leaky
Symptom
Responses allow direct identification of a person, so experts self-censor.
What to do
Use tools with true anonymization. Ensure aggregation uses minimum bucket size. Anonymize rationales before publication.
Trap
Panel too homogeneous
Symptom
All experts think similarly, range is narrow and blind spots remain.
What to do
Recruit panel deliberately as a heterogeneous mix. At least three perspectives (implementer, provider, observer).
Trap
Too many rounds
Symptom
After round 3 there is no movement, panel fatigue increases, drop-out rises.
What to do
Limit to 3-4 rounds. If no consensus after round 3, this is the result. Forced consensus weakens the study.
Trap
Owner influences aggregation
Symptom
Owner's preferred position is reinforced, outliers are filtered.
What to do
A method specialist or external reviewer checks aggregation. Keep answer transparency in the report.
Trap
Question too broad
Symptom
Responses spread too widely, aggregation remains arbitrary.
What to do
Keep the question to a scale or structured answer space. Use open answers only as rationale, not as main data.
07
Stop criteria
Done signals checkable in under a minute
Question requires quick answer and Delphi cycle is too long.
Panel cannot be recruited, fewer than 5 experts available.
Anonymity cannot be ensured and responses are filtered.
Question is factually answerable with existing data and Delphi effort is disproportionate.
Owner cannot aggregate neutrally, method specialist is missing.
Panel expects compensation outside budget or timing is not feasible.
Finished the runsheet?
Go to the profile for purpose, similar methods, and sources or continue to the next method in the catalog.