The four formats, and when to pick each
Start / Stop / Continue is the fastest format to run and the easiest to close out, because every card is already phrased as a behaviour change rather than an observation. Its weakness is the same thing: it jumps straight to solutions, so a team that hasn't yet agreed on what happened will argue about the fix instead of the facts. Use it for a routine sprint where nothing dramatic went wrong.
What went well / What didn't go well / Action items is the neutral default, and the one to reach for if you're not sure. It deliberately separates observation from decision: the first two columns describe the sprint, and nothing moves into the third until the group has agreed it's worth changing. That separation is what stops a retro turning into a list of grievances with no owner.
Mad / Sad / Glad asks for feelings rather than facts, which sounds soft and is in fact the most useful thing you can do when something went badly and nobody is saying so. "Mad" gives people a sanctioned place to put frustration, and frustration is usually pointing at a real process problem. Use it after a painful release, a re-org, or a stretch of quiet tension.
4Ls — Liked, Learned, Lacked, Longed for — is the format for the end of a long project rather than a two-week sprint. "Learned" is the column that earns its keep: it captures knowledge the team gained that would otherwise evaporate, and it's the only one of these formats that treats learning as an outcome rather than a side effect. "Lacked" and "Longed for" separate what was missing from what you wish existed, which are different asks with different owners.
How to run one that isn't a blame session
The single most effective move is to set the frame before anyone writes a card: the retro examines the system, not the people in it. Norm Kerth's Prime Directive is the usual wording — everyone did the best job they could with what they knew and the resources they had. Reading it out loud feels slightly awkward the first time and noticeably changes what gets written.
Then have people write in silence for five minutes before anyone speaks. Silent writing is doing real work: it stops the loudest person setting the agenda, it gives quieter and more junior people a way in, and it surfaces duplicate cards, which are the strongest signal you'll get all session about what actually mattered.
Keep cards about events and effects rather than individuals. "The deploy broke and we found out from a customer" is workable; "Sam broke the deploy" is not, and it ends the useful part of the conversation. If a card names a person, ask what the system did that let it happen — usually there's a missing check, a missing handover, or a missing pair of eyes underneath it.
If you're a manager, be careful about going first. Whatever you say becomes the frame, and people will calibrate their cards to it. Write silently like everyone else, and let someone else facilitate if you can.
How long a retro should take
For a two-week sprint, an hour is plenty and 45 minutes is often better. A rough shape that works: five minutes to set the frame and recap what happened, ten minutes of silent writing, fifteen to twenty minutes grouping and discussing, five minutes of voting, and the last ten minutes turning the top votes into concrete action items with names attached.
The timebox matters more than the total. Discussion expands to fill whatever space it's given, and a retro that runs long tends to spend its extra time on the item that generated the most heat rather than the most value. Set a timer per phase, and when discussion overruns on one card, park it and move on — parking is not the same as ignoring, as long as the parked card goes somewhere real.
For a project retro or a quarter's worth of work, budget 90 minutes and expect to use the 4Ls. Longer retrospectives need a break in the middle and, ideally, a facilitator who isn't also a participant.
Dot voting: why prioritizing is half the job
A well-run retro will produce twenty or thirty cards. A team can meaningfully change two or three things before the next retro. Voting is how you get from one number to the other without the decision being made by whoever argued hardest.
The usual mechanic is three dots per person, spendable however they like — all three on one card is allowed and is a strong signal. Group near-duplicate cards before voting, or you'll split the vote across three phrasings of the same problem and it will lose to something less important. Turn on dot voting in this tool and each card gets a counter you can bump; columns then sort by vote count so the top of each column is the top of the discussion.
Then stop at the top two or three. The temptation to action-item everything is the most reliable way to end up doing none of it, and a team that watches its retro output get ignored will stop bringing anything real to the next one.
Turning the board into action items people actually do
An action item that survives contact with the next sprint has three properties: a single named owner, a change small enough to finish inside one sprint, and a way to tell whether it happened. "Improve our testing" has none of these. "Ali adds a smoke test to the deploy pipeline by Thursday" has all three.
Prefer changes to the system over changes to intention. "Remember to update the changelog" depends on everyone's memory forever; "the release script refuses to run without a changelog entry" is done once. Anything you can encode in a checklist, a template, a default, or a script will outlive the enthusiasm of the meeting that produced it.
Carry the previous retro's actions into the start of the next one and read them out. This is the step teams skip, and skipping it is why retros lose credibility — nothing teaches a team that the retro is theatre faster than watching last month's items quietly disappear. If an item didn't get done twice, that's information: either it wasn't important, or something is blocking it, and both are worth two minutes.
Saving the board to an OmniCanvas canvas keeps the columns and cards as they were, so the next retro can sit beside the last one and the follow-up is a glance rather than an archaeology exercise.
Remote and hybrid retros
Distributed teams lose the two things a physical wall of stickies provides: everyone writing at once without being influenced, and everyone seeing the whole board at a glance. Silent writing handles the first — ask people to type their cards before the discussion opens rather than as it goes. Sharing one screen with the board on it handles the second.
In a hybrid room, put everyone on the same board even if half the team is in the office, and don't run a physical wall in parallel. The moment there are two boards, the remote half of the team is watching a meeting rather than in one.
This board is a single-browser tool by design — nothing is uploaded, so there's no shared live cursor. For a remote retro, one person drives while the group calls out cards, or each person drafts their own and reads them in. If you want a persistent shared board across devices and sessions, that's what saving to a canvas in the app is for.