Planning Poker: How Teams Estimate Together
Planning poker is how a lot of agile teams put a size on work before they commit to it. Everyone estimates the same story at the same time, in private, and then the whole team shows their cards together. A round takes a minute or two, and it gets the quiet people's opinions on the table along with the loud ones.
Where it comes from
James Grenning described the technique in 2002, and Mike Cohn made it widely known in his book Agile Estimating and Planning. The name comes from the cards: each person holds a hand of numbered cards, and a vote is a card played face down.
The numbers are story points, a relative measure of effort, risk and uncertainty. A 5 means bigger than the 3 you pointed last week and smaller than the 8, whatever that turns out to be in hours.
Why everyone reveals at once
If the lead developer says "that's a 3" before anyone else speaks, most of the room will drift toward 3. That's anchoring, and it hides exactly what estimation is supposed to surface: the tester who knows the edge cases, or the new hire who can't see how the change would work at all. When everyone votes in private and reveals together, every number on the table was picked independently.
How to run a session
- Pick a host. The host reads out each story, keeps time, and decides when to reveal.
- Read the story and take questions. Keep it short: the goal is a shared picture of the work, not a design review.
- Everyone picks a card in private. Nobody says a number out loud.
- Reveal all the cards at once.
- If the cards agree, write the number down and move on.
- If they don't, the lowest and highest voters explain their reasoning, and the team votes again.
Most stories settle in one or two rounds. If a story is still split after the second vote, it usually isn't clear enough yet. Split it or take it offline instead of arguing a number into existence.
Reading a split vote
A spread like 2 and 13 is where planning poker earns its keep. Usually one person knows something the others don't, like a legacy module nobody wants to touch, or a library that already does half the job. Let the outliers talk first, because they're the ones holding that information.
The special cards
Most decks add a few cards that aren't numbers. In Fib, ? means you don't know enough to estimate yet, ☕ asks for a break, and 🃏 says the story is off the top of the scale. If two or three people play ?, the story isn't ready to estimate.
Running it remotely
Physical cards don't work on a video call, so remote teams use an online tool. What matters is that votes stay hidden until the reveal and that joining takes seconds. In Fib, the host creates a room and shares the link, and teammates join with just a name. There's no account and nothing to install. After the reveal, Fib shows the spread, marks the lowest and highest votes, and says whether the team agreed or split. A few habits keep the rounds quick:
- Set the topic before anyone votes, so everyone is sizing the same thing.
- Use a round timer for long backlogs. A visible countdown stops one story from eating the whole hour.
- Write down the reasons behind a split, not just the final number. They're often the start of the acceptance criteria.
Planning poker questions
- How long should a planning poker session take?
- Plan for one to three minutes per story. If a story takes much longer, it probably needs more refinement before the team can estimate it.
- Who should vote?
- Everyone who will do the work: developers, testers, designers. The product owner usually answers questions but doesn't vote, since they won't be building it.
- Should we average the votes?
- No. Averaging a 3 and a 13 gives an 8 that nobody believes. Talk through the gap and vote again until the team agrees, or until the difference no longer matters.
- Do story points translate into hours?
- Not directly. Points compare stories with each other. Over a few sprints, a team learns how many points it usually finishes, and that velocity is what turns points into a forecast.