How planning poker works

The whole method exists to defeat one specific failure: anchoring. When a senior engineer says “that’s about a three” before anyone else speaks, the rest of the room converges on three. The number that comes out is not the team’s estimate, it is that one person’s estimate with extra steps.

Voting privately and revealing together removes that path entirely. The disagreement becomes visible, and disagreement is the useful part. A round where one person votes 2 and another votes 13 has surfaced a misunderstanding about scope that would otherwise have shown up halfway through the sprint.

1. Read the story out loud

   The person who wrote the story explains it and answers questions before anyone touches a card.

2. Everyone votes at once

   Each voter privately picks the card matching their estimate of the effort involved.

3. Reveal together

   All cards flip at the same moment so nobody has anchored on somebody else’s number.

4. Discuss the outliers

   The highest and lowest voters explain their reasoning. This is where the real information appears.

5. Re-vote or record

   Vote again if the discussion changed minds. Otherwise record the agreed estimate and move on.

The host runs the session. They reveal the cards, reset a round for a re-vote, move the team to the next story and can change anyone’s role. In SprintsPoker the host can also hand hosting to someone else, which matters when the usual facilitator is away.

Voters pick cards. They can change their mind freely until the reveal, which is worth saying out loud to a team new to the technique. Nobody sees a changed vote.

Spectators watch without voting. This is the right role for a product manager who wants to hear the discussion, or a stakeholder sitting in. Spectators are excluded from the vote count and from the agreement score, so their presence never distorts the result.

Do not average the votes. An average of 2 and 13 is 7.5, a number nobody in the room believes. The spread is a signal that people are estimating different pieces of work.

Ask the lowest and the highest voter to explain, in that order. Low voters usually know a shortcut. High voters usually know a landmine. Once both have spoken, re-vote. If the second round is still split, the story is too big or too vague to estimate, and the right move is to split it rather than force a number.

That advice assumes a majority to converge on, which a pair does not have: two voters always split 50/50 and both are the outlier. Running planning poker for 2 people works, but it needs a tie-break rule agreed before the session rather than a third round of voting.

SprintsPoker shows an agreement percentage on reveal so you can see this at a glance. Above eighty percent, record it and move on. Below half, expect a conversation.

Estimation meetings run long for predictable reasons: stories arrive unrefined, one story eats twenty minutes, or the team re-litigates the point scale every session. Agree the scale once and leave it alone.

A countdown timer helps more than it sounds like it would. Two minutes per story is usually enough, and the deadline converts an open-ended debate into a decision. If you need more than two rounds on a story, that is your signal to split it.

If you are still choosing a scale, the card deck guide compares Fibonacci against T-shirt sizing, and the story point guide covers what the numbers should mean.