How to run a retrospective
A retrospective is an hour the team spends deciding what to change. Most of them fail in the same few ways, and all of those are avoidable. Here is the shape that works, what each part is for, and the mistakes worth knowing about before you make them.
The five phases
- 01
Set the stage
5 minSay what the retro covers and how long it runs. Read the prime directive out loud if the team is new to it: everyone did the best they could with what they knew at the time. This is not ceremony: it is what makes it safe to say the difficult thing twenty minutes later.
- 02
Gather data
10–15 minEveryone writes, silently and at the same time. Silent writing is the single highest-leverage habit in a retro: it stops the loudest person setting the agenda, and it gets the quiet people's observations onto the board at all. Keep cards hidden until everyone has finished so nobody anchors on the first opinion they read.
- 03
Generate insight
10 minReveal, read out, and group related cards into themes. You are looking for the pattern under the individual complaints: four cards about waiting is one card about handoffs. Then vote, with a fixed budget per person, so the discussion goes where the team's attention actually is rather than where the argument is loudest.
- 04
Decide what to do
15 minTake the top two or three themes and turn them into actions. An action needs an owner and needs to be small enough to finish before the next retro. "Improve communication" is not an action. "Ana posts the deploy checklist in the team channel by Friday" is.
- 05
Close
5 minRead the actions back, confirm the owners, and say where the notes will live. Then actually put them there. A retro whose output never leaves the tool teaches the team that retros do not matter.
Which format should you use?
The format decides what the team notices. Rotate it when the same cards keep appearing.
Start / Stop / Continue
The default. Fast to run, hard to get wrong.
Start · Stop · Continue
How to run it →
Mad / Sad / Glad
Surfaces how the sprint actually felt.
Mad · Sad · Glad
How to run it →
Liked / Learned / Lacked / Longed for
Deeper reflection for the end of a milestone.
Liked · Learned · Lacked · Longed for
How to run it →
Good / Bad / Start / Stop
Separates what happened from what to do about it.
Good · Bad · Start · Stop
How to run it →
Sailboat
Wind, anchors, rocks, island. Good for strategy retros.
Wind · Anchors · Rocks · Island
How to run it →
Side by side, with what each one surfaces: all retrospective templates. None of them fit? Build a custom board with your own columns.
Five ways retros quietly fail
Skipping the actions
The most common failure by far. A retro that ends in discussion and no owned action is a support group. If you only have time for one thing, cut the discussion short and spend the last ten minutes on actions.
Never revisiting last time
Open the next retro by reading out the previous actions and their status. Two minutes. It is the only thing that makes the team believe actions are real.
The manager writes first
Whoever speaks or writes first sets the frame. Silent writing with cards hidden until reveal removes the problem structurally, rather than relying on everyone being brave.
Same format every time
Teams stop seeing what the columns stop asking about. Rotating the format, say a sailboat retro after a quarter of Start/Stop/Continue, surfaces different material from the same people.
Too many people
Above about ten, most people stop talking. Split into two retros and have the facilitators compare notes afterwards.
Common questions
›How long should a retrospective be?
About an hour for a two-week sprint. Forty-five minutes works if the team is practised and the format is familiar. Under thirty minutes you will get through gathering data and not reach actions, which defeats the point.
›Who should facilitate?
Ideally not the person the team reports to, and ideally rotating. A facilitator who is also the loudest stakeholder cannot run a neutral room.
›What if nobody says anything?
Silence usually means it is not safe, or the question is too broad. Switch to silent writing before discussion, and ask a narrower question: not "how did the sprint go" but "what took longer than you expected".
›Should retrospectives be anonymous?
Anonymity helps a team that is not yet safe, and hurts one that is, because you cannot follow up on a card nobody will claim. Hiding cards until everyone has written gets most of the benefit without the cost.
›How do we run a remote retrospective?
Use a shared board everyone can write on at once, keep cameras on for the discussion, and be stricter about timeboxing than you would in a room. The failure mode of a remote retro is one person narrating while everyone else waits.