Async vs. live: which one fits
Live planning poker is still the gold standard when a team shares enough working hours to meet. The simultaneous reveal is instant, discussion is immediate, and consensus lands in one sitting. Async trades that immediacy for reach: nobody wakes at 3am for a call, and quiet voters get thinking time instead of being anchored by whoever speaks first. Reach for async when your team's working-hour overlap is under an hour a day, or when most of your backlog is routine and only a few stories need real debate. If you can meet live, read planning poker for remote teams first — async is the fallback when live genuinely doesn't fit.
Prepare the stories first
Async estimation is unforgiving of vague stories. In a live session someone asks a clarifying question and the room adjusts; async, a confused voter either guesses or stalls the whole window. Before you open voting:
- Each story has a title, a short description, and at least one acceptance criterion.
- Anything genuinely unknowable is flagged for a spike, not dropped into the window.
- Oversized stories are split first — see how to split user stories.
Set a voting window
A window is a deadline, not a meeting. Twenty-four hours works for most distributed teams: open the session Monday morning in the earliest time zone, close it Tuesday morning. Everyone votes once within that span, on their own time, from a focused single-story view. Longer than 48 hours and the backlog goes stale; shorter than a full working day and you exclude someone's schedule.
Collect rationale and confidence
The single biggest thing async gets wrong is losing the "why". In a live session the discussion carries it; async, the number arrives with no context. Ask every voter for a one-line rationale and a confidence signal alongside their card. A 5 with "done this twice, low risk" is a very different estimate from a 5 with "no idea, guessing" — and only the async format makes that difference visible before the discussion even starts.
Set a per-story voting window when you create the session. Teammates vote on their own time from a single-story focused page — rationale and confidence included — and cards stay hidden until the window closes.
Nudge, don't nag
Someone always forgets. One reminder at the halfway mark and one an hour before close is plenty — more than that and people tune it out. Keep the reminder a single link straight to the unvoted stories, never a "please vote" message with no context. If a voter misses the window entirely, proceed without them rather than reopening; a reopened window re-anchors everyone who already voted.
Reveal and read the spread
When the window closes, reveal all at once. The rationale and confidence you collected turn the reveal into a triage list: stories where everyone converged with high confidence are done — record the estimate and move on. Stories with a wide spread, or a tight spread but low confidence, go to the discussion list. Most backlogs converge on 70–80% of stories without any conversation at all.
Converge in a short live call
Reserve synchronous time only for the contested stories. A 20-minute call covering the five widest-spread items beats a two-hour call covering everything. The outliers explain their reasoning first — that's usually where a hidden assumption or a missed edge case surfaces — then re-vote once. If a story still won't converge after one re-vote, it's a signal to handle the disagreement deliberately: split it, spike it, or park it.
Failure modes to watch
- Votes visible before the window closes. If people can see each other's cards mid-window, it's just slow open voting with anchoring. Cards must stay hidden until reveal.
- No rationale. Numbers without context can't be triaged; you end up discussing everything, which defeats the point of async.
- Windows that never close. "We'll leave it open until everyone votes" turns a 24-hour window into a two-week limbo. Set the deadline and honor it.
- Estimating a backlog nobody refined. Async amplifies every gap in story prep. Refinement is not optional here.
Done well, async planning poker gives distributed teams most of the consensus quality of a live session for a fraction of the synchronous cost — the discussion time goes only where the disagreement actually is. The Agile Alliance's planning poker overview is a good primer on the underlying technique if your team is new to it.