What is Wideband Delphi?
A group estimation technique.
Where is it effective?
For making top-down estimates in situations where there are a lot of unknowns or various kinds of domain knowledge required. Especially useful for early estimates of large, not-yet-well-understood projects — estimates upon which we base our go/no-go decisions and early expectation-setting for upper management and customers.
How does it work?
This variation of Wideband Delphi has been taught or used hundreds of times at Microsoft and elsewhere. It is slightly different from the original described in 1981 by Boehm.
- Schedule the estimation meeting for a particular project or set of projects. The estimators should be team members (and possibly other stakeholders) who are very knowledgeable about the project, and from whom you need buy-in for the schedule of the project. 3-5 estimators is the sweet spot, although much larger groups can work, by breaking them into 3-5 person sub-groups and then combining those estimates.
- Describe what the group is estimating. What part of the project, goals, or outcome are we estimating? What types of resources are we going to include? What units (ideal man-days or man-months, story points, etc.) will we use?
- Ask everyone to estimate individually and privately using their best, instinctual judgement on the estimate. Give them enough time (5-20 minutes) to do this work quietly. Once people are familiar, you may be able to ask them to do this before the estimation meeting. Ask them also to take note of the assumptions and major pieces of work that led to the estimate — they’ll use these later in the group discussion. The private/anonymous aspect of this first round of estimates is important: we’re specifically empowering each person to express their gut instinct, and not be influenced by group think yet.
- Show the results on a spreadsheet or whiteboard, and discuss. The group can see how divergent or convergent their estimates are. Ask each person (but especially the high/low or other interesting estimators) to explain how they got to their estimate — major assumptions, things included, etc. People will be reminded of things they forgot to include in their estimates. They’ll also be forced to confront different assumptions (which you may be able to quickly decide upon). All of which will improve future rounds of estimates.
- Repeat steps 3&4 for a total of 2-3 rounds. After round one, anonymity is dropped and discussions can be short (the moderator should make decisions on behalf of group, for expediency and just for the purposes of the estimate). Often only 2 rounds total are needed to see estimates converge somewhat. A common effect is estimates converge together and up as people realize from the discussions the things they missed from their initial estimates.
Spread the word
Feel free to modify and use.
If you make use of these templates, please help spread the word. And if you can make your sheets public, please put links here in the comments so others can see examples in use. Thanks!