How to run backlog refinement with planning poker
Backlog refinement can be a short working conversation: give people enough context to compare an item, surface assumptions with private votes, then decide what needs to happen next. Teams can adapt this flow to their own planning rhythm; it is not a required ceremony or a rule about team scale.
Use a light agenda
- Name the item and intended outcome. State the problem, who benefits, and what a usable result should do.
- Share the relevant context. Cover known constraints, dependencies, and open questions without solving every implementation detail.
- Vote privately. Each participant chooses a card before numbers are discussed.
- Reveal and discuss meaningful outliers. Ask what led the low and high voters to their choices.
- Choose the next move. Revote after clarification, split the item if it is too broad, or leave it Uncertain while someone learns more.
Refinement Pointer supports the voting part of the conversation. Choose either the 1, 2, 3, 5, 8, 13 deck or XS, S, M, L, XL deck when you create a Room, share the Room link, then use reveal and reset for each item.
Discuss the gap, not every card
A narrow spread may be enough to move on. When the cards are far apart, invite the outliers to explain their assumptions. A low estimate may rely on an existing component that others did not know about. A high estimate may expose a compatibility concern, research task, or dependency. Keep the question concrete: what would make this item more complex, more uncertain, or more effortful than the reference work?
Fictional example
For a proposed notification setting, most of a fictional product team votes M, while one person chooses XL because the setting must work for accounts with different permissions. The group learns that permissions are owned by another service. They split the research and integration work, then revote the smaller delivery item.
Do not use the cards as a timer or a verdict on anyone's ability. They are relative team-local estimates, informed by complexity, uncertainty, and effort. They do not convert to hours and do not measure productivity.
Make the next action clear
End each item with a useful decision: it is ready to be planned at the chosen relative size, it needs a smaller slice, or it needs information before estimating. Then reset the Round for the next conversation. Participants can change a choice before reveal, and the Organizer can end the Session when the team is done. The Room link and recorded history remain available for the next Session. Capture any planning note in your usual system.
For more on choosing and interpreting the cards, see how to use story points in planning poker.