Learning objective
By the end of this lesson, you can prepare controlled round feedback that supports revision without pressuring experts toward the median.
Why it matters
Convergence can look like quality even when experts still rely on different assumptions. A Delphi study is useful because it reveals and refines judgement—not because every participant ends with the same number.
Core concept: feedback without authority pressure
Round feedback should include:
- the distribution, not only the average;
- recurring assumptions and evidence;
- anonymised reasons across the range;
- material minority or safety-critical arguments;
- an invitation to revise or retain with explanation.
Do not reveal identities, rank experts by seniority or tell participants that movement toward consensus is desirable.
Visual explanation
Text alternative: independent responses are summarised as a distribution and reasons, then returned for justified revision until the stopping rule is met.
Worked example
Experts estimate autonomous claims-processing potential between 15% and 80%. The range reflects three different assumptions: document completeness, permitted decision scope and appeal requirements.
Round two reports those assumptions and the distribution. It does not claim that the median is correct. Experts revise conditionally, producing a narrower estimate for document-complete claims while retaining disagreement about contestability.
Common mistake
Mistake: using AI to produce a smooth “panel consensus” paragraph.
The summary may erase the minority rationale that identifies the most consequential risk. AI-assisted coding must retain links to original responses and receive human review.
Quick check
Which feedback is strongest?
A. “The senior expert chose 70%.”
B. “Please move closer to the group average.”
C. “Here is the distribution and anonymised assumptions; revise or retain with reasons.”
D. “Outliers were removed.”
Answer: C. It supports reflection without status pressure.
Practical prompt
Write one expert-estimation question with outcome, population and horizon. Define the round-one response fields, summary statistics and stopping rule.
Summary
- Ask a specific shared question.
- Collect initial judgement independently.
- Show distributions and assumptions.
- Preserve outliers and minority reasons.
- Stop under a pre-defined rule.
Next lesson
Continue with Make Participation Rights Explicit, where the team must distinguish advice, recommendation and accountable authority.