To facilitate planning poker, prepare stories before the meeting, then run each one through the same tight loop: read it (1 minute), answer questions (2 minutes), vote silently, reveal at once, hear the highest and lowest voters (2 minutes), and revote. Allow at most two revotes before you park the story. Most well-prepared stories should take 2 to 5 minutes, and a 60-minute session should clear 8 to 12 of them.
The rest of this guide is the detail behind that loop: what to do the day before, a minute-by-minute agenda, the timebox rules, and the exact words to use when a session goes sideways.
The facilitator's job in one paragraph
You are not there to estimate. You are there to protect two things: independent first votes and the team's time. Everything below serves one of those. If you notice yourself saying a number out loud before the reveal, or letting a debate run because it is interesting, you have stopped facilitating and started participating.
Before the session: prep checklist
Most slow sessions are lost before they start. A story that arrives without acceptance criteria will burn ten minutes of the whole team's time discovering what it means. Copy this checklist into your refinement calendar invite or team wiki.
24 to 48 hours before
- Product Owner has picked 10 to 14 candidate stories, ordered by priority (bring more than you expect to finish).
- Each story has a one-sentence goal and testable acceptance criteria. Run them against a user story readiness checklist.
- Known dependencies and open questions are written on the story, not held in someone's head.
- Mockups, API docs, or links are attached where relevant.
- Stories are shared with the team so people can pre-read. Ask for questions in writing ahead of time.
- Anything too big to finish in a sprint is flagged for splitting, not estimating. See how to estimate large epics.
Morning of
- Pick two reference stories the team already sized (for example, a known 2 and a known 8) and have them ready to show.
- Decide the deck in advance. Fibonacci is the default for most teams.
- Open the planning poker room and paste the link in the invite.
- Agree who records final estimates and parked questions (not the facilitator, if possible).
- Have a timer ready for the discussion windows.
Who should be there
Developers and testers vote. The Product Owner answers questions and usually does not vote. Managers and stakeholders can watch, but should not vote or talk numbers. If your tool supports a spectator role, use it for them.
The 60-minute session agenda
This agenda assumes stories were pre-read. If they were not, expect to finish 6 to 8 instead of 10 to 12.
| Time | What happens | Facilitator notes |
|---|---|---|
| 0:00 to 0:03 | Open, state the goal | "We're aiming for 10 stories. Two revotes max, then we park." |
| 0:03 to 0:05 | Show the two reference stories | Re-anchors the team on what a 2 and an 8 look like for you |
| 0:05 to 0:25 | Stories 1 to 5 (hardest first) | Start with the stories that need fresh brains |
| 0:25 to 0:27 | Two-minute stand-up break | Skip only if the session is 45 minutes or less |
| 0:27 to 0:50 | Stories 6 to 10 | Watch for fatigue signs (see below) |
| 0:50 to 0:55 | Batch round for similar small stories | "Same as the last one? Any objection to a 2?" |
| 0:55 to 1:00 | Recap | Read back estimates, parked stories, who owns each parked question |
Two things about this agenda are deliberate. Hard stories go first because energy is highest at the start. And the session ends at 60 minutes even if stories are left. A second 30-minute session tomorrow produces better estimates than minutes 60 to 90 today. If your sessions regularly need more time, the fix is upstream, covered in how to shorten grooming sessions.
The per-story loop and its timeboxes
| Step | Timebox | Rule |
|---|---|---|
| 1. Read the story | 1 min | PO reads the goal and acceptance criteria. No implementation talk yet. |
| 2. Clarifying questions | 2 min | Questions about what, not how. "Does this include admin users?" is fine. "We should use Redis" is not. |
| 3. Silent vote | 30 to 60 sec | No numbers spoken, typed, or hinted. |
| 4. Reveal | instant | All cards at once. |
| 5. Outliers explain | 2 min total | Highest and lowest voter, about a minute each. Questions only from others. |
| 6. Revote | 30 sec | Back to silent voting. |
| 7. Decide | 30 sec | Accept, take the tiebreak, or park. |
How long to discuss each story? Plan on 2 minutes of discussion after the reveal, and a hard cap of about 8 minutes for the whole story. As a rough guide:
| Story type | Total time | What it looks like |
|---|---|---|
| Clear and familiar | 2 to 3 min | One vote, cards within one step, done |
| Standard | 4 to 6 min | One outlier round, one revote |
| Unclear | Stop at 8 min | Park it with a named question and owner |
Revote rules
- Votes agree or sit on two adjacent cards (say 3s and 5s): accept. If it is a genuine split, take the higher card and move on.
- Votes spread wider: highest and lowest explain, then revote.
- Still spread after the first revote: one more revote, but only if the discussion surfaced new information. If nobody learned anything new, a third vote just measures who got tired first.
- Still spread after the second revote: park it. Write down the specific question that would settle it and who will answer it.
The detailed version of these rules, including when to take the higher estimate versus splitting the story, is in our guide to consensus fatigue and how many rounds to vote.
What to do when the timer runs out mid-discussion
Do not ask for "one more minute." Say what you heard and call the vote: "Time. I'm hearing the disagreement is about whether we need a migration. Let's vote with that on the table." If the vote is still spread, it gets parked. Teams adapt quickly once they learn the timebox is real, and they start putting their most important question first.
What to say: facilitator phrases for common moments
Exact wording matters because your tone tells the team whether disagreement is welcome. These are phrases that work.
Opening the vote
"Cards down, nobody say a number. If you'd vote a question mark, vote it. That tells us something."
When the PO or a senior person pre-anchors ("this should be quick")
"Let's hold that thought until after the reveal. I want everyone's independent read first."
Anchoring has its own article: how to stop anchor bias in planning poker.
Asking an outlier to explain
"Priya, you voted 13 and most people are around 5. You might be seeing something the rest of us missed. What's in your 13?"
Then to the low voter:
"Sam, you went with 3. What makes this simpler than it looks?"
Notice the framing: the outlier is a source of information, not someone who needs correcting. Never say "Why so high?" It sounds like an accusation.
When people start debating instead of explaining
"Let's keep this to questions for Priya and Sam for now. Arguments go into the revote."
Stalled debate (same two people, same point, third lap)
"We've heard both sides twice. Seems like the disagreement is really about [X]. Can anyone in the room answer that right now? If not, we park it and [name] finds out by Thursday."
Implementation rabbit hole
"Good design conversation, but it's a how question. Can we agree it's somewhere between 5 and 8 and take the design to the kickoff?"
Silent member (has not spoken in several stories)
Do not put them on the spot with "Jordan, you've been quiet." Ask a specific, low-risk question tied to their expertise:
"Jordan, you worked in the billing module last sprint. Anything in there that would surprise us?"
Or give everyone a turn so nobody is singled out:
"Quick round before we revote: one sentence each on your biggest risk. Jordan, start us off?"
Dominant voice
"Thanks, Alex, that's useful. I want to hear from someone who hasn't spoken on this one."
Votes cluster with no discussion five stories in a row
"We've agreed on everything for the last five. Either we're very calibrated or we're tired. Anyone want to argue the other side on this one?"
Parking a story
"We're parking this. Open question: does the partner API support batch calls? Owner: Dana. We'll re-estimate next session."
Closing
"We sized nine. Two parked, owners are Dana and Lee. Thanks for keeping it tight."
Running a fast session without cutting corners
Speed comes from removing dead time, not from rushing the thinking.
- Pre-reads. If people have seen the story, the PO's reading becomes a 30-second reminder.
- Batch similar stories. For three stories like "add field X / Y / Z to the form," estimate the first one properly, then ask "Same size for the other two? Any objection?"
- Do a quick T-shirt pass on big backlogs. Sort 20 stories into S/M/L in ten minutes, then only point the ones heading into the next two sprints.
- Round up on adjacent splits. A 3 vs 5 argument rarely changes sprint capacity enough to justify five minutes of debate.
- Move simple stories async. Stories nobody expects to discuss can be voted before the meeting, so the live session only covers the interesting ones.
A useful way to get buy-in for timeboxing is to do the arithmetic out loud. Hypothetical example: 8 people spending 15 minutes on each of 12 stories is 3 hours of meeting, or 24 person-hours. At 5 minutes per story it is 1 hour, or 8 person-hours. The meeting cost calculator will put a currency figure on that if your manager responds to money.
Keeping the team engaged without gimmicks
People disengage when their vote doesn't seem to matter, when they wait with nothing to do, or when two people debate while everyone else watches. Fix those causes directly.
| Cause of disengagement | Fix |
|---|---|
| "My vote doesn't change anything" | Always hear outliers, even on close votes. A lone 8 among 5s can carry the context everyone missed. |
| Waiting for slow voters | Pre-reads, and a short visible voting window ("30 seconds, then we reveal") |
| Two-person debates | Questions-only rule after outliers explain; move the design talk to kickoff |
| Long PO readings | Share stories ahead; read only the goal and acceptance criteria in session |
| Sessions that drag | Hard 60-minute cap, hardest stories first, a short break at the midpoint |
| Same person always facilitates | Rotate the facilitator every session, or every five stories |
A two-minute warm-up question can help a remote team that rarely talks outside meetings; icebreakers that aren't awkward has options. Keep it short and optional.
Things that backfire: leaderboards for fastest voter (rewards not thinking), "last to vote buys coffee" (punishes careful people), forced enthusiasm, and stacking every engagement trick at once. Pick two fixes from the table that match your actual problem.
Fatigue signals to watch: cameras going off, "I'm fine with whatever," votes converging instantly on the middle card, discussion time collapsing on non-trivial stories. When you see two of these, finish the current story and stop.
A worked example: one hour, ten stories
Hypothetical team: six developers, one tester, a PO. Stories were shared two days earlier.
| Story | Vote 1 | Action | Final | Time |
|---|---|---|---|---|
| Payment retry logic | 5, 8, 8, 13, 8, 5, 8 | Outliers (5 and 13) explain, revote | 8 | 6 min |
| Export audit log | 3, 3, 5, 3, 3, 3, 3 | Hear the 5: "Large tenants time out." Revote | 5 | 4 min |
| Profile avatar upload | 5, 5, 5, 3, 5, 5, 5 | Accept | 5 | 2 min |
| SSO for partner portal | 3, 13, ?, 8, 5, 13, ? | Two "?" votes: nobody knows the partner's IdP. Park | Parked | 4 min |
| Add three fields to signup | 2, 2, 2, 2, 1, 2, 2 | Accept, then batch the other two field stories | 2, 2, 2 | 3 min |
Notice the SSO story. Spending eight more minutes on it would not have produced a better number, because the missing fact was outside the room. Parking it with an owner was the right call.
Common facilitation mistakes
| Mistake | Fix |
|---|---|
| Facilitator says "I'm thinking 5, thoughts?" | Never voice a number before reveal |
| "Close enough, call it 5" without hearing the 8 | Always ask the lone outlier, even on tight votes. It takes 20 seconds. |
| Averaging 3 and 13 to get 8 | A wide spread means a different understanding of the story. Discuss or park. |
| Letting "one more minute" become ten | Timebox is the rule. Summarize, vote, or park. |
| Estimating stories without acceptance criteria | Send them back. Unready stories make every session slow. |
| Running 90+ minute sessions | Cap at 60. Schedule a second short session instead. |
| Facilitator also records everything | Assign a scribe, or use a tool that exports the session |
Where a tool helps
Any tool that hides votes until a simultaneous reveal will do. In Alignlee, votes stay hidden until reveal, and the round auto-advances 2 minutes after reveal unless you cancel it, which works as a built-in nudge for the 2-minute discussion window. There is no round timer for the voting phase, so keep a phone timer for that. At the end you get a summary with the number of rounds, consensus rounds, and average time per round, which tells you whether your timeboxes are holding. For remote-specific setup, see remote planning poker.
FAQ
How do you facilitate planning poker for the first time? Prepare 8 to 10 stories with clear acceptance criteria, pick two reference stories the team agrees on, and explain the rules at the start: silent vote, simultaneous reveal, outliers explain, two revotes maximum. Keep the first session to 45 or 60 minutes and expect to finish fewer stories than planned.
How long should you discuss each story in planning poker? About 2 minutes of discussion after the reveal, with the whole story taking 2 to 5 minutes in most cases. Set a hard cap of around 8 minutes. If a story isn't settled by then, it usually needs information nobody in the room has, so park it with a named owner.
What is a good planning poker timebox for a whole session? 60 minutes, with a short break around the 25 to 30 minute mark. Teams that run 90 to 120 minute sessions tend to rubber-stamp estimates near the end. Split long backlogs into two shorter sessions.
How do I make planning poker faster? Share stories ahead, keep clarifying questions to "what" rather than "how," take the higher card when votes split across two adjacent numbers, batch similar stories, and park anything that needs outside information. Most slowness comes from unready stories, not from voting.
How do you keep the team engaged during estimation? Make every vote matter by always hearing outliers, cut waiting time with pre-reads and short voting windows, stop two-person debates with a questions-only rule, and keep sessions under an hour. Avoid leaderboards and penalties, which reward speed over thinking.