Scrum poker is a collaborative estimation technique where team members vote on the effort required for tasks using numbered cards. It improves accuracy in Agile project planning by incorporating diverse perspectives.
The method was developed in 2002 by James Grenning as part of his work on planning poker, a technique for improving software estimation. The most common card sets use the Fibonacci sequence (0, 1, 2, 3, 5, 8, 13, 20, 40, 100) to reflect increasing uncertainty in estimating larger tasks.
Team members independently select cards representing their estimates, then reveal them simultaneously. Outliers discuss their reasoning, and the group repeats until consensus forms. This process reveals hidden assumptions and encourages discussion about task requirements.
For example, a team estimating a feature to implement user authentication might initially show cards ranging. The highest voter explains their estimate, revealing they considered integrating with a third-party service. The team then adjusts their understanding and re-estimates.
| Card Type | Values | Use Case |
|---|---|---|
| Fibonacci Sequence | 0, 1, 2, 3, 5, 8, 13, 20, 40, 100 | Most common for story points |
| T-Shirt Sizes | XS, S, M, L, XL, XXL | Relative sizing without numbers |
| Hours/Days | 0.5, 1, 2, 4, 8 hours or days | Time-based estimation |
Teams use scrum poker during sprint planning to estimate user stories in the product backlog. The technique works because it:
A 2016 study by Laurens van der Starre found teams using planning poker achieved 30% greater estimation accuracy than those using single-expert estimation.
Many teams confuse scrum poker with simple voting or assume it produces exact time estimates. However:
I've seen teams change their planning sessions by adopting scrum poker. One team reduced their estimation time by 40% while improving sprint completion rates from 70% to 95% within three months.
While similar to wideband Delphi or affinity mapping, scrum poker differs by:
For complex projects, some teams combine scrum poker with spike solutions to better understand unknowns before estimation.