navbar 4Resource papers in action research

Delphi face to face


This is a resource file which supports the regular public program "areol" (action research and evaluation on line) offered twice a year beginning in mid-February and mid-July.  For details email Bob Dick  or  

...  in which a face to face version of the delphi process is described (delphi is a process which uses an expert panel to make complex decisions, and can be used for action research)  1




Delphi is not usually a face-to-face small group technique.  By far its most common use is for future forecasting by a panel of experts.  It typically takes place via several rounds of a mailed survey.

At its simplest, you can think of delphi as asking a question (or several) of the panel, and collating and summarising the replies.  The summary is sent out to start the next cycle.  A panel member may change her estimate in the direction of an emerging consensus.  Alternatively she may leave her original estimate intact and provide information justifying it.

At the heart of the technique you will find a cyclic process and a way of using disagreement as a trigger for deeper analysis.  A panel member who is uncertain of her estimate is probably motivated to modify it in the direction of the emerging consensus.  A panel member who feels her original estimate was justified is motivated to identify and exchange information which may persuade others to her point of view.

If her estimate is an informed one, she may then produce information which other members did not previously have.  In effect, this then educates those panel members.  From round to round, the relevant information is shared.  Decisions are them made on the basis of more complete information.

In most delphi processes, therefore, the amount of consensus increases from round to round.


Delphi has some features which can be used in other information-collection techniques:

  • there is no face-to-face interaction
  • each participant is given time for thought and an equal opportunity to contribute
  • in particular, disagreements are used to generate pooled information and increased understanding.

Instead of discarding an item of information which disagrees with other items, you use it to gather more detailed information.  This is a feature which can be used within cyclic small-group processes, or where two different sets of information are available.

The main elements of delphi can be combined with techniques for group facilitation, and used in a small-group or large-group setting.  The resulting process may be used for forecasting, as with delphi.  It can also be used for other problem-solving and task-oriented activities which depend on information which at the start is dispersed rather than shared.

Other applications suggest themselves.  For instance, these are times when multi-disciplinary teams are becoming more common.  But different specialised languages hinder teamwork.  A process such as delphi can allow a common language to be forged from the different languages.

An example of a delphi-style process is described below.  It is included as an interesting demonstration of a process which makes maximum use of idiosyncratic information, yet does so in the interests of eventual consensual decision making. 


A group version of Delphi

Here is a detailed description of a face-to-face delphi...

  1. Individual work.  Working individually and without discussion, each person responds to a question which defines the required information.  Ideally, this question is specific enough to eliminate most irrelevant information, but otherwise places as few constraints on the information as possible.
  2. Small groups.  Participants collect together in groups of similar people.  They prepare a group list of information arranged in order of importance.  (The use of groups of similar people makes it likely that information important to a particular category of participant will reach the whole group list.)  
  3. Whole group.  Gather the important items from each group on newsprint.  To do this, ask each group in turn to contribute the most important item on their list not already on the whole group list.
  4. Whole group.  A multiple-vote procedure 2 is used to rank the items from most to least important.  A natural cut-off point is chosen between items with high scores, and those with low scores.  Somewhere between six and nine items are appropriate for most topics, but it saves debate if a natural gap occurs between the selected items and the rest.
  5. Individual.  Each person considers what changes she wishes to make to the group list in the light of the whole group list.
  6. Small groups.  Small groups compare the list of top items on the group list to those on the whole group list.  Where the group list differs from the whole group list, the group has two options -- change its list to conform more nearly to the whole group list;  or develop evidence for changing the whole group list more in the direction of the group list.
     This is done as follows.  Add to the group list those items from the whole group list which the group previously omitted, but is prepared to accept.  Prepare a brief report (with a time limit, which may be one minute for example) supporting any of the top items from the group list, and which the group believes should be added to the whole group list.  The purpose of this report is not to persuade others to their point of view, but to present evidence which they think others may have overlooked.
    Each group documents its revised list on one sheet of newsprint, and its evidence (in note form) on another.
  7. Whole group.  Groups report back.  Their revised lists are displayed without comment.  Each group in turn displays its sheet of evidence and speaks briefly to it.  Each group report is followed by a brief session of questions for clarification only.  Strict time limits are enforced.  If these reporting-back sessions become too time consuming, they not only prolong the activity, but may begin to bore some participants.  Time urgency helps arousal.
  8. Whole group.  Return to step 4 to repeat the cycle as far as step 7.  Continue cycling through steps 4 to 7 until consensus emerges.  (Time constraints may require a fixed number of cycles and instructions to hasten consensus.  Consensus can be increased by having two rounds of voting instead of one at step 4.)

The balance between individual information and group consensus can be changed by giving more time to some sessions, and by altering the instructions to emphasise consensus or individual information. 


Structured discussion

A skilled facilitator can use structured discussion in a way which approximates this cyclic procedure.  Voting can be used to exchange information.  Individuals can then be invited to identify the ways in which their requirements are not met by the vote.  The group can be asked to suggest modifications which preserve the advantages for the group, while relieving the problems for individuals.  You can often achieve a greater level of consensus than you might imagine. 


Types of process

I have talked above as if there were two types of process, consensual and non-consensual.  Before leaving the topic of microprocesses I think it is desirable that I extend this discussion a little.  There are two comments I wish to make.  The first is about the nature of consensus.  The second concerns the differences between the process used in delphi, and some of the simpler processes which are used for reaching consensus.

Consider, first, the processes most often used.  Different people put their point of view.  A decision is then made to accept one or other of the views offered.  Debate is the typical process for exchanging information.

Decisions are most often taken by an arbitrator or leader or by majority vote.  To oversimplify the issue somewhat, one of the expressed views prevails.

This is often referred to as an adversarial model -- "win/lose".  It is commonly found in many of our institutions, such as courts of law, or in politics.  In case you were wondering, I don't recommend it.  (On occasion it has its advantages:  it's probably better than false consensus.)

In distinction to this, consensus does not pit one view against another.  It seeks to reach a decision which will be acceptable by more than just a majority (or more than just those with power).

Many group processes pursue consensus in a simple minded way.  In brief, the processes identify those issues on which there is already substantial agreement.  If the different positions overlap a great deal then this is likely to be both satisfying and reasonably efficient in its use of time.

On other occasions, however, there may be too little overlap.  A decision which pleases some may grossly displease others.  For this, more complex processes may be needed.

The usual approach is some form of negotiation.  This is still a version of the adversarial approach.  The difference is that each person gives up something for something of greater value.  By some canny horse trading, each person may achieve a partial improvement over the prior situation.  You could call it "partial win / partial win".

A different alternative is to strive to develop a better solution.  Instead of accepting one position, or trading within established positions, you go as far as possible towards satisfying the requirements of all.  The result is likely to be an outcome which none of the parties might have predicted at the outset.

This has been called a dialectical process.  I think of it as a tough form of win/win.  It works by expanding the information base each person has.  If better decisions are available, they may then be identified.  Delphi is a dialectical process.

The purpose of both consensual and dialectical methods is similar:  to produce decisions which are satisfying and effective enough to gain the commitment of all those affected.  Simple consensual methods are likely to be appropriate when the areas of agreement are substantial.  They identify existing agreement.

Dialectical methods may work even when there is initial disagreement.  They generate agreement.

I would add a further point.  Dialectical methods are in general not as robust as the simpler approaches.  If you are a relative novice you may wish to avoid issues when there are strong differences of opinion about them.

For those reluctant to use dialectical methods, consensual methods may still be worth using.  When people strive to understand one another, much of the heat usually disappears from interaction.  Relationships improve.  People can, and do, agree in good will to differ.  Using simple consensual methods may lead to a better understanding even where agreement on decisions is minimal.

In any event, simple consensual methods are easily strengthened.  You can use consensual methods to decide process and criteria before moving on to content.  Agreement can then often be found in the face of what first appeared to be strong conflict.  The use of preparatory activities (goal setting, climate setting and team building) increase the likelihood that people will desire consensus rather than aggressive debate. 



  1. A much-edited extract from Bob Dick (1991) Helping groups to be effective:  skills, processes and concepts for group facilitation.  Chapel Hill, Queensland:  Interchange.    [ back ]
  2. There are a number of ways in which participants can be given multiple votes as a way of gaining a greater spread of votes, and more agreement on the favoured items.  For instance, you can give each participant a certain number of votes to spread over the items as they wish.  Or you can ask them to choose the "x" most important items.  An elaboration of the second of these might ask them to choose the x most important items (for two votes each) and the x next most important, for one vote each.
    If this doesn't give you enough agreement, you can repeat the vote until it does.    [ 
    back ]



Copyright (c) Bob Dick 1991-2000.  This document may be copied if it is not included in documents sold at a profit, and this and the following notice are included.

This document may be cited as follows:

Dick, B.  (2000) Delphi face to face  [On line].  Available at




navbar 4  


Maintained by Bob Dick;  this version 3.07w last revised 20050815

A text version is available at URL