What is planning poker?
Planning poker (also called scrum poker or pointing poker) is a consensus-based way to estimate the relative size of user stories or other backlog items. Each person who will do the work holds a deck of cards with values such as 1, 2, 3, 5, 8 and 13. The team discusses a story, everyone chooses a card in private, and all cards are turned over together. If the numbers agree, you record the estimate and move on. If they differ, the outliers explain what they see that others might not, and the team votes again.
James Grenning described the technique in 2002, and Mike Cohn popularized it in Agile Estimating and Planning. It is a lightweight descendant of the Wideband Delphi method: independent estimates first, discussion second. The value is less in the number than in the conversation it forces. A 3 next to a 13 almost always means two people have a different picture of the work, and finding that out during refinement is far cheaper than finding it out halfway through the sprint.
Planning poker usually produces story points, a relative measure of effort, complexity and uncertainty, rather than hours. Teams then use the points with their historical velocity to plan how much fits into a sprint.
Planning poker rules (the short version)
- Only the people doing the work vote. Developers, testers and anyone else who builds the item. The product owner answers questions but normally does not vote.
- One story at a time. Read it aloud, show the acceptance criteria, and take clarifying questions before anyone votes.
- Vote in private. Nobody says a number out loud before the reveal. This is the whole point: it prevents anchoring on the first or most senior opinion.
- Reveal together. All cards are shown at the same moment.
- Outliers speak first. The highest and lowest voters explain their reasoning. Nobody is wrong for voting high or low.
- Re-vote, usually once or twice. If the team still has a wide spread after two rounds, the story needs refinement, a spike, or splitting rather than more debate.
- Estimates are relative. A new story is compared with stories the team has already sized, not converted from hours.
- The ? card is legitimate. It means "I can't estimate this with what I know," which is useful information.
How to run a planning poker session, step by step
1. Prepare the stories and two reference stories
Before the meeting, the product owner picks the stories to size and makes sure each has a title, a short description and acceptance criteria. Pick two stories the team has already finished and agrees on, for example a small 2 and a medium 5, and keep them visible. Every new estimate is a comparison against those anchors. Budget about 5 minutes per story; a 60-minute session realistically covers 8 to 12 stories. See our backlog refinement guide for deciding which stories are ready.
2. Create a room and share the link
With physical cards, hand out a deck to each voter. Online, the facilitator creates a room, picks a deck, and shares the link or room code in the meeting chat. On Alignlee that takes one screen: enter your name, choose a deck, and create the room. Participants open the link and type a name.
3. Read the story and clarify
The product owner reads the story and acceptance criteria. Questions are about scope and behavior ("does this include the admin view?"), not about numbers. Timebox this to two or three minutes.
4. Everyone votes privately
Ask voters to compare the story with the reference stories, not to count hours. If someone is unsure, they should still vote and signal low confidence, or play the ? card.
5. Reveal all cards at once
With physical cards, the facilitator counts down and everyone flips. Online tools hide votes until reveal. In Alignlee, votes stay hidden and the reveal happens automatically once half of the voters have voted, so nobody has to wait on a facilitator click.
6. Discuss the outliers (with a script)
Use the same wording every time so it feels routine rather than confrontational:
"We have a spread from 3 to 13. Sam, you went highest: what are you seeing that makes this bigger? Priya, you went lowest: what are you assuming that makes it smaller? Everyone else, just listen for now. Then we'll vote again."
Keep each explanation to a minute. Typical causes of a wide spread: someone knows about a hidden dependency, someone assumed existing code can be reused, or the acceptance criteria were read differently.
7. Re-vote and settle
Vote again after the discussion. If the cards are now within one step of each other (5 and 8, say), pick the value most people chose or the higher of the two, and move on. If the spread is still wide after the second vote, stop: say "this one needs more refinement" and park it with a named owner. Repeated re-votes produce fatigue and groupthink, not accuracy.
8. Record the estimate and move on
Write the agreed points into your tracker as you go, or export the session and update the tracker afterwards. If you use Jira, our Jira planning poker guide covers a simple workflow.
Worked example: one story, two rounds
Story: "As a customer, I can download my invoices as PDF." Five developers vote on the Fibonacci deck. The reference stories are a 2 (add a field to the profile page) and a 5 (CSV export of orders). Hypothetical numbers:
| Voter | Round 1 | Confidence | Round 2 |
|---|---|---|---|
| Alex | 5 | 90% | 8 |
| Bea | 5 | 80% | 8 |
| Chris | 8 | 40% | 8 |
| Dana | 13 | 20% | 8 |
| Eli | 3 | 70% | 5 |
Round 1 has a spread from 3 to 13. Dana (13) explains that the existing PDF library cannot render the invoice layout and would need replacing; Eli (3) assumed the CSV export code could be reused. After that discussion the team re-votes and lands on 8, with one 5. The facilitator records 8.
Notice the confidence column. The plain average of round 1 is 6.8, pulled up by Dana's low-confidence 13. Alignlee's confidence-weighted estimate for the same round is about 5.9, because votes cast with low confidence count for less. Either way, the spread is the signal: the team learned about the PDF library before committing to the sprint, which is what planning poker is for. More on this in our guide to confidence voting.
The three planning poker card decks
Most teams use one of three decks. Alignlee offers all three, each with a ? card. Custom decks are not supported.
| Deck | Values | Use it when |
|---|---|---|
| Fibonacci | 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, ? | You size stories in story points and want the gaps to widen as items get bigger and less certain. |
| Modified Fibonacci | 0, ½, 1, 2, 3, 5, 8, 13, 20, 40, 100, ? | You need a 0 or ½ for trivial changes and rounder numbers (20, 40, 100) for epics. This is the deck Mike Cohn's cards popularized. |
| T-shirt sizes | XS, S, M, L, XL, XXL, ? | Early roadmap sizing, non-software teams, or stakeholders who find numbers misleading. |
Why non-linear values? Estimation error grows with size, and the widening gaps stop people from arguing over 6 versus 7. Our guide to Fibonacci estimation goes deeper, and Fibonacci vs T-shirt sizing helps you pick.
Common planning poker mistakes and how to fix them
| Mistake | What it looks like | Fix |
|---|---|---|
| Anchoring before the vote | "This is probably a 3, right?" before anyone votes. | No numbers out loud until the reveal. Facilitator interrupts politely every time. |
| Converting hours to points | "Two days, so that's a 5." | Compare with the reference stories instead. Points are relative. |
| Endless re-votes | Fourth round on the same story. | Two rounds maximum, then park the story for refinement or a spike. |
| Averaging away disagreement | 3 and 13 become an 8 without discussion. | Always hear from the outliers before settling. |
| Managers or stakeholders voting | Estimates drift down when the person asking for the work votes. | Let them join as spectators and answer questions. |
| Stories not ready | Half the session is spent writing acceptance criteria. | Use a readiness check during refinement, before the poker session. |
For more depth, read how to facilitate planning poker and how to stop anchor bias.
Alignlee's free online planning poker tool
You can run everything above with paper cards. If your team is remote or hybrid, an online tool handles the hidden votes and simultaneous reveal for you. Here is exactly what Alignlee does, and what it does not.
No signup, no login
The host enters a name, picks a deck, and creates a room. Everyone else opens the link or types the 6-character room code and enters a name. There are no accounts, emails, or passwords, which means external contractors and client stakeholders can join without IT tickets or a "yet another account" conversation. It is free, with no paid tiers. A tip for clear discussion: ask people to use recognizable names such as "Sarah (backend)" rather than "Guest", because votes are shown with names after the reveal.
Unlimited participants, guests and stakeholders
There is no participant cap, so you never have to choose which specialist gets the last seat. That said, large voting groups slow discussion down. With more than about ten voters, hear only from the single highest and lowest voter, keep a strict two-round limit, and consider letting people who only need context join as spectators instead of voters.
Spectator mode
Tick "Join as spectator" when joining (the host can also create a room as a spectator). Spectators see the story, voting progress and results, but cannot vote, and they are excluded from the vote math. It suits product owners who want to answer questions without influencing the number, managers who want visibility, and new hires who should watch a couple of sessions before voting. A simple flow for spectators: questions during the story reading, silence during voting and the outlier discussion, and a chance to add context before the team moves on.
Confidence voting and the weighted estimate
Each vote carries a 0 to 100% confidence slider. The result is a confidence-weighted estimate: a vote at 50% counts normally, low-confidence votes count less, and high-confidence votes count more. A story with lots of low-confidence votes is a refinement candidate even if the numbers agree. For a sprint-level check, Confidence Pulse mode lets the team rate sprint confidence from 1 to 5 and see the average, in the style of a fist-of-five vote.
Auto-reveal at 50% and auto-advance
Votes are hidden until the reveal, which triggers automatically once half of the voters have voted. Votes are then shown with names; there is no anonymous mode, which keeps the outlier conversation straightforward. Two minutes after a reveal the room advances to the next round automatically, and anyone who needs more discussion time can cancel the countdown. There is no per-round timer, so timebox discussions yourself.
Rejoin within 24 hours
Rooms stay open for 24 hours. If someone's laptop sleeps or a tab is closed, opening the link again in the same browser rejoins the room automatically. The host can end the meeting for everyone at any time. Rooms are not permanent: create a fresh room for each session.
Works in mobile browsers
The interface is responsive and works in phone and tablet browsers, with nothing to install. That matters most in hybrid meetings, where people in the room vote from their phones. There is no native app and no offline mode; you need a connection to vote.
Privacy: what is stored
A room holds the room code, the display names people type, and their votes and confidence levels. There are no accounts, so no email addresses or passwords are collected. Rooms expire after 24 hours or when the host ends the meeting. The site uses Google Analytics and Vercel Analytics, which is disclosed in the privacy policy. Even so, keep confidential customer names or unreleased product details out of story titles if your policies require it.
Meeting minutes export (.md)
Before ending the meeting, export the meeting minutes as a Markdown (.md) file with the rounds and results. When the host ends a session, a summary shows the number of rounds, how many reached consensus, and the average time per round. Alignlee does not call any AI service, but you can paste the .md file into an assistant such as Claude yourself; see analyzing meeting minutes with Claude.
Icebreakers and light gamification
An optional "would you rather" icebreaker can open the session, and there are a few light touches such as achievements and a slot machine on the waiting screen. If you want a warm-up question outside the room, use the random icebreaker generator.
What Alignlee does not do
- No Jira or other tracker sync. Paste issue keys into the round title and type the agreed points back into your tracker.
- No Slack app or bot. Share the room link in any chat instead.
- No custom decks, CSV export, offline mode, or installable app.
- No persistent team rooms or long-term history; rooms last 24 hours.
- English only.
If any of those are must-haves, our planning poker alternatives page compares other tools honestly.
Running planning poker in a hybrid or in-room meeting
Hybrid estimation fails when the people in the room hold physical cards and the remote people watch a laptop camera pointed at the ceiling. A few habits fix most of it:
- Everyone votes on their own device, even in the room. No physical cards that remote people cannot see. A phone browser is enough.
- Put the room on the big screen. One display shows the voting room; if you have a second, use it for the remote participants' faces so they are not hidden behind shared content.
- Fix audio before video. A speakerphone in the middle of the table beats a laptop microphone that only hears the person next to it. Have someone at the far end of the table speak at normal volume and ask a remote person to confirm.
- Frame faces, not the table. A camera at eye level at the end of the room shows who is talking; a laptop camera shows two people and a wall.
- Call on remote outliers first. When a remote voter is the high or low card, invite them to speak before the room starts talking among itself.
- Run a two-minute check before you start. Can remote people hear the far seat? Can the room see remote faces? Has everyone joined the poker room and cast a test vote?
For fully distributed teams, see remote planning poker, and for teams spread over many time zones, async estimation across time zones.
Planning poker FAQ
What is planning poker in simple terms?
Planning poker is a way for a team to estimate work together. Everyone picks a card with their estimate in private, all cards are revealed at once, and the people with the highest and lowest cards explain their thinking before the team votes again. The simultaneous reveal stops early or senior voices from anchoring everyone else.
What are the rules of planning poker?
Only the people doing the work vote. Each voter picks one card per story without showing it, cards are revealed together, the high and low voters explain first, and the team re-votes until estimates converge or the story is parked for more refinement. Estimates are relative: a 5 should feel roughly like other 5s, not like a fixed number of hours.
Is planning poker free?
The technique itself is free; you can run it with paper cards. Alignlee's online planning poker is also free, with no paid tiers, no trial, and no feature limits.
Do participants need an account?
No. On Alignlee the host enters a name and creates a room. Everyone else opens the link or types the 6-character room code and enters a name. There is no email, password, or login.
How many people can join a planning poker room?
Alignlee has no participant cap. In practice, keep the voting group to the people who will build the work; a group of roughly 3 to 9 voters keeps discussion manageable. Everyone else can join as a spectator.
Can stakeholders watch without voting?
Yes. Tick 'Join as spectator' when joining. Spectators see the story, the voting progress, and the revealed results, but they cannot vote and are excluded from the vote math, so they never skew the estimate.
What happens if someone drops off the call or closes the tab?
Rooms stay open for 24 hours, and the same browser rejoins the room automatically when the person opens the link again. The host can end the meeting for everyone at any time.
Can I save the results of a planning poker session?
Yes. Export the meeting minutes as a Markdown (.md) file before you end the meeting. It lists the rounds and results, so you can paste the final points into Jira or another tracker, or hand the file to an AI assistant to summarize.
Related guides
- Sprint planning guide: using estimates to decide what fits in a sprint.
- Relative sizing with planning poker: building reference stories.
- Calibrating story points: keeping a 5 meaning the same thing sprint after sprint.
- Sprint velocity calculator: turn points into a forecast.