How to run planning poker

A round takes two minutes when the team is prepared and half an hour when it is not. The difference is almost always the same: whether everyone knows what exactly is being estimated. Here is an order that works both in a meeting room and on a call.

What to prepare in advance

A team will take a long time to size an item it is seeing for the first time. Collect the list beforehand and make sure every entry has a sentence or two of description and a clear idea of what “done” means.

A reasonable session covers five to fifteen items. Beyond that, attention runs out halfway through the list and estimates start being cast just to move on.

A list of items, each with a short description.
A reference item whose size the team already agrees on — everything else is compared to it.
An agreement on the deck and on what the result of a round means: median or average.
A facilitator: someone who keeps order, asks the questions and records the result.

The facilitator’s job

The facilitator does not estimate — they keep the round moving. Their job is to name the item, let people ask questions, wait for the cards, reveal them and write down the result.

Their most important duty is to stop the round turning into a design discussion. “How will we build it” is a separate conversation and takes many times longer than sizing. When the team drifts into architecture, the facilitator pulls it back with one question: “is that enough to name a size?”

The round itself

The order is simple, but every step in it earns its place:

Name the item and give the team thirty seconds for questions.
Wait until everybody has picked a card. Do not reveal early: one open card spoils the whole round.
Reveal all cards at the same time.
If the spread is narrow, take the result and move on — there is nothing to argue about.
If the spread is wide, give the floor to the extremes, a minute each.
Run the round again. A third round is almost never needed: if the team has not converged after the second, the item should be split rather than estimated.

What to do about a wide spread

A spread is not a failure of the round — it is its main output. A “one” and an “eight” on the same item mean two people estimated two different items, and it is worth finding out which ones.

Usually the cause is one of three: part of the team knows about hidden complexity (a migration, another service, old code), part of the team does not know where the item ends, or the wording allows two readings. All three are fixed by clarifying, never by averaging.

Never take the arithmetic mean as a way to “close the question”. The average of one and eight describes neither of the two items people had in mind.

How long it should take

An item everybody understands the same way is sized in a minute. An item with a disagreement takes three to five. Once you are ten minutes in, you are discussing the solution rather than the size — take it to a separate meeting and mark the item as needing clarification.

A timer helps not by hurrying people but by making time visible: the team notices on its own that a fifth minute is going into a single card.

Common mistakes

There are only a few, and they repeat from team to team.

Estimating in hours. Hours differ from person to person; points are shared.
Revealing cards one by one — that is voting out loud again, only slower.
Re-estimating an item after it is done. An estimate is what you knew before, not after.
Inviting people who will not do the work: someone else’s card adds nothing to the conversation.
Estimating an item with no description. That item goes back to refinement, not onto the table.