Planning Poker

What planning poker is

Planning poker is a consensus estimation technique. The team takes a backlog item, everyone privately picks a card representing how big they think it is, and all the cards are revealed at the same time. If the estimates agree, the item is done in twenty seconds. If they do not, the people at the extremes explain what they were thinking — and that conversation is the entire value of the exercise.

The technique was described by James Grenning in 2002 and popularised by Mike Cohn. It exists because the two obvious alternatives both fail: asking one expert produces a number nobody else understands, and going round the table out loud produces a number everyone has anchored to whoever spoke first.

Why teams use it

• It surfaces misunderstandings before the work starts. A one and a thirteen on the same item almost always means two people are describing different pieces of work. Finding that out during estimation costs five minutes; finding out mid-sprint costs the sprint.
• It removes the anchoring effect. Simultaneous reveal is not a ritual detail. Without it, the first number spoken sets the range for everyone else.
• It gives quiet people the same weight. Everyone commits to an answer before anyone can be talked out of it.
• It builds a shared reference scale. After a few sprints, "this is a five" carries real meaning within the team, and forecasting gets better without anyone estimating in hours.

It is not a scheduling tool. Story points do not convert to days, and a team's velocity is a property of that team, not a rate you can compare between teams or negotiate upward.

The estimation scales

QRetro ships four decks. Pick one and keep it — switching scales mid-project throws away the reference points the team has built.

Why the gaps widen is covered in detail on the Fibonacci scale page.

How to run a session

1. Prepare the list beforehand. Eight to twelve items is a realistic hour. Anything nobody can describe in two sentences is not ready to estimate.
2. Agree a reference item. Pick something the team has already finished and call it a three. Every later estimate is relative to it, and without a reference the first few numbers are arbitrary.
3. Read the item and take questions. Two minutes, no estimates spoken aloud.
4. Everyone votes at once. Cards stay hidden until all votes are in.
5. Reveal. If the spread is one step, take the higher value and move on — do not negotiate over a single step.
6. If the spread is wide, ask the extremes to explain. Highest and lowest only. Usually one of them knows something the rest do not.
7. Re-vote once. If the second round still disagrees, the item is not understood well enough. Send it back to refinement rather than averaging the difference.

Common mistakes

• Averaging the votes instead of discussing them. The average of a 2 and a 13 is not an estimate, it is a way of not having the conversation that would have told you why.
• Letting the loudest person speak first. The reveal is simultaneous for a reason; the discussion afterwards should still start with the outliers, not the senior.
• Converting points to hours. The moment a five means "two days", the team stops estimating relative size and starts negotiating deadlines, and the technique is dead.
• Estimating items nobody has refined. Planning poker measures understanding as much as size. An unclear item produces a wide spread every time.
• Re-voting until everyone agrees. Two rounds is the limit. Persistent disagreement is information, not an obstacle.
• Comparing velocity between teams. Points are calibrated per team. A cross-team comparison rewards whoever inflates their scale.

Connecting your tracker

If Jira or Linear is connected to the team, tasks can be imported straight into a game instead of being retyped, and the agreed estimate can be written back — Jira story points, Linear estimate — when the integration was connected with write permissions. Details on the Jira page.

Running it in QRetro

1. Open a game and pick an estimation scale.
2. Add tasks manually, or import them from Jira or Linear.
3. Share the game link. Participants join from the browser without creating an account.
4. Select a task, let everyone vote, then reveal all cards at once.
5. Discuss the outliers, re-vote if needed, and sync the agreed estimate back to the tracker.

Planning poker in QRetro is free: no per-seat fee, no game quota, and estimation lives alongside retrospectives on the same accounts.