Votenir
You reveal the cards and the table looks like this: two 3s, a 5, an 8 and a 13. Someone sighs. Someone else says "let's just call it a 5". Five minutes later nobody remembers why the 13 was played, and three sprints later that story is still in progress.
When planning poker estimates disagree, the meeting is working. The spread is the point. It tells you the team doesn't share the same picture of the work. This guide shows how to turn that into a better estimate in two rounds or less.
Why disagreement is good news
Planning poker exists to surface hidden assumptions. A wide spread usually means one of four things:
- Someone knows about a risk the others don't (a fragile legacy module, a missing API, a compliance step).
- Someone knows a shortcut the others don't (an existing component, a library, a previous similar change).
- People are estimating different scopes, for example with or without tests, migration or documentation.
- The story is unclear, and everyone has filled in the gaps differently.
Averaging the numbers hides all four. Talking about them fixes all four.
The playbook: from spread to agreement
Step 1: Look at the spread, not the average. A 5 average from five 5s is very different from a 5 average made of a 2 and a 13. If the highest and lowest cards are next to each other on the deck (say 3 and 5), you're close enough: pick one and move on.
Step 2: Ask the outliers to explain, high first. Ask the highest voter "what are you seeing?", then the lowest. Keep it to one or two sentences each. Starting with the high estimate tends to bring out risks before optimism takes over.
Step 3: Check scope. Ask, "Are we all including testing, review and deployment?" Many disagreements disappear the moment the team agrees on what "done" covers.
Step 4: Re-vote once. Hide the cards again and let everyone choose with the new information. In Votenir, the owner can un-reveal the round so people can change their card, then reveal again.
Step 5: Decide, don't loop. If the second round converges, take it. If it doesn't, use one of the tie-breakers below. Never go to a third round hoping the numbers will sort themselves out.
Tie-breakers when the second round still disagrees
| Situation | What to do |
|---|---|
| Spread of one step (e.g. 5 and 8) | Take the higher number; it's usually closer to reality |
| Spread of several steps, clear reason | Split the story along the line of disagreement |
| Several ? cards | Stop; the story needs refinement before estimating |
| Nobody knows how hard it is | Add a timeboxed spike and estimate after |
| One expert vs the rest | Ask the expert to pair with someone; estimate with that in mind |
Common mistakes
- Averaging the result. "8 and 3, so let's say 5" produces a number nobody actually believes.
- Letting the most senior person decide. That's just anchoring with extra steps.
- Asking the outlier to justify themselves. Ask what they're seeing, not why they're wrong. The person with the odd card is often the only one who's right.
- Discussing before the reveal. Talking about size before voting removes the independent estimates planning poker relies on.
Use your history to spot patterns
One disagreement is a conversation. The same disagreement every sprint is a pattern. Look back at past rounds and ask:
- Do we always split on a particular kind of work (database changes, UI polish, integrations)?
- Was the high voter right more often than not?
- Which stories that we "settled" on a middle number ran long?
Votenir saves every revealed round with each person's vote, so you can bring real examples to your next retrospective rather than impressions. For more habits that keep sessions short, see 10 planning poker best practices.
Frequently asked questions
Should you take the average in planning poker?
Not as the final estimate. The average hides the information you need. Use the spread to start a discussion, re-vote, then pick a value the team agrees with.
How many times should you re-vote?
Once. If the team still disagrees after one re-vote, split the story, take the higher value, or add a spike.
What if one person always votes very differently?
Ask them what they're seeing. Their view may come from experience the team lacks. If it's a calibration problem, a shared reference story usually fixes it.
Is it okay to just pick the higher estimate?
For a one-step gap, yes. Most work takes longer than expected, so the higher estimate is usually the safer choice.