Planning Poker

Trade estimates out loud in a meeting and the first number spoken becomes the anchor. Someone says three, and the person who was thinking eight says five instead. They didn't change their judgment; calling out a number more than twice as large as the one already in the room was uncomfortable. When the senior engineer speaks first, the effect gets stronger, and nobody in the room experiences it as pressure.

It happened in a Pagedeck sprint planning session — the meeting where the team decides what goes into the cycle. On the item for reworking the document sharing screen, Theo said three should cover it. Nobody objected, so three got written down. That work ran past the cycle and into the following week.

The cause wasn't estimation skill; it was how information was distributed. Another developer knew the permission handling on that screen had five exception branches and was privately thinking eight, and by the time three had been said the conversation had already settled. Planning poker inverts that order. Everyone commits a number before anyone speaks, which turns the fact that estimates diverge into an input to the meeting rather than something that quietly disappears.

Reveal simultaneously, explain the spread, and reveal again. Six steps.

1. Fix one reference item. Pick something everyone understands and declare it a three — building the login screen, say. Without that anchor the same five means a different size to each person, and matching cards in that state isn't agreement about anything.
2. Fix the card values. Use a sequence with widening gaps, like 1, 2, 3, 5, 8, 13. The gaps are deliberate, so nobody is invited to believe they can tell a four from a five. Many teams add a question mark card for "I don't know" and a zero for "almost nothing." A table full of question marks is its own useful signal.
3. Explain the item. Read out what gets built and what counts as finished, then take questions. The questions here are frequently worth more than the estimate, because they expose what hasn't been decided yet.
4. Reveal at the same time. Everyone holds their card and turns it over together. Order creates the problem the method exists to remove, so avoid the situation where three people reveal and one reveals after.
5. Have the spread speak. Whoever called highest and whoever called lowest each explain why. Not everyone in turn — the information you need sits at the two ends, and the middle rarely has anything to add.
6. Reveal again. After hearing the explanations, one more round. Two or three rounds usually converge. Still split between three and thirteen on the third round, and you stop matching numbers and split the item instead. A spread that wide isn't an estimation difference; it means people are picturing different work.

Timing belongs in the procedure too. Keep an item under three minutes, and pull anything that runs longer out for two people to sort out after the meeting. Spend fifteen minutes on one item and the rest get waved through, which lowers the quality of the whole set of estimates rather than raising it.

It suits work several people build together, and most of all work where the information is distributed. An item where the designer knows the screen, the developer knows the data structure, and sales knows the customer conditions is where this method earns the most. It also pays off in a newly formed team, because aligning each other's baselines is needed before estimating is worth anything, and the cards force that alignment in public.

Four limits.

1. Too many items — running forty through this takes three hours. The common shape is sorting size bands quickly with T-shirt sizing first and dealing cards only for the items actually going into this cycle.
2. Work only one person knows — if one person handles the server configuration alone, nobody else has a number to reveal. What comes out isn't a consensus; it's that person's estimate with everyone else's signature under it.
3. Remote meetings — with no mechanism for simultaneous reveal, the first number posted in chat becomes the anchor again. Hide values in a tool, or count to three and submit together.
4. Organizations that use the numbers for evaluation — where estimates feed performance reviews, everyone drifts upward. Then the cards stop diverging, and cards that don't diverge make this a procedure that only consumes time.

Not mistaking the purpose is the condition for using it. The artifact is one story point number — the abstract unit for the relative size of work — but the real output is what the team learned about the item. If the number is the goal, one person can write it down without any cards at all.

"Let's flip the cards together. Speaking first drags everyone to that number."

The order of speaking distorts the estimate.

"We have a three and a thirteen. Just those two, tell us why."

Hear the reasoning at the two ends first.

"Three rounds and it's still split. I think we should break this item up."

Split the item rather than keep matching numbers.

Pagedeck has used cards in planning since that week. Mia reads the item, takes questions, counts to three, and everyone reveals. Estimating the same permission handling item again produced a three, an eight, and a thirteen, and the person who called thirteen explained the five exception branches.

The second round came back with everyone at eight or above. Right there the item got split, one half into the current cycle and the other into the next. What Mia recorded wasn't the number eight but the list of five exception branches, and that list went into the requirements document.

The most common belief is that this is a way to get accurate estimates. Pooling several judgments sounds more accurate than one. But a group still doesn't know the size of something nobody has built before. What the method reduces isn't error but one person's bias and the influence of speaking order. Expect accuracy and the first missed estimate becomes the argument for dropping the procedure, at which point the ordering problem walks right back in.

The other common belief is that you keep going until everyone agrees, since the numbers converging looks like the goal. But past the third round, the remaining gap isn't missing information — it means the item itself is vague. Keep rolling and people agree to whatever number ends the meeting, and that number is one nobody believes. Stopping and splitting is the better move, and it costs less than the meeting you'd otherwise extend.