Wideband Delphi: The Accurate Method Behind Planning Poker
Planning poker is based on a process known as Wideband Delphi. It’s time to rediscover the underlying method, developed in stages first at RAND in the 50s and then by the late, great Barry Boehm and John Farquhar. This iterative, consensus-driven approach is the most accurate estimating method available.
Start by insuring everyone is well-informed about what’s being estimated. Be careful that the person explaining the project doesn’t bias everyone. This can be mitigated by sharing this responsibility.
Involve people with good knowledge of the domain. You can start with as few as two, although estimates will improve as more folk are added until about 5.
Select a diverse array of people to mitigate selection bias. While domain knowledge is important, try to include people who are off the team and even socially distant. With a sufficient number of people, you might eliminate those in authority from the exercise, as these persons tend to influence others.
Process
If the range of estimates isn’t narrowing as you iterate, have everyone reveal the rationale behind their estimates, not just the outliers.
Document all rationales throughout the process, as these may inform your final report to stakeholders.
If it emerges that information is missing or asked for, stop to find it, and then reconvene. When you reconvene, first replay the last round and read out the last revealed rationales for estimates.
Three iterations are often all that’s needed for convergence, with estimates narrowed but unlikely to further converge.
Range
When you are ready to derive an estimate from this exercise, you may use the median or average of individual estimates. If you want to respect an outlier opinion, choose averaging. If there’s a clump of agreement you respect more, choose the median.
The remaining dispersion in estimates can be used to establish an uncertainty around your final estimate, so instead of “20 months” you might report “15 to 30 months with 20 most likely”.
Conclusion
While Wideband Delphi can be done efficiently, its iterations are probably overkill for sprint planning. However, much of what happens in development spans multiple sprints. Next time you have an epic to scope out or a major feature to commit to in a road map, give it a try.
Bonus
My friend Ricardo Valerdi took a deep dive on Wideband Delphi.
