Managing
A retrospective is a short, structured meeting where a team examines how it worked, not what it produced, and commits to two or three changes for next time. Run it in five moves: frame it, gather what happened, find the pattern, decide the changes, close. The format matters far less than the first sentence you say, because that sentence decides whether anyone tells you the truth.
Most retrospectives fail politely. Everyone turns up, everyone agrees the last few weeks went fine apart from some pressure at the end, somebody writes “improve communication” on a board, and nobody looks at it again. That is rarely a format problem. It is a script problem: the person running the meeting never says anything specific enough for anyone else to risk being specific back. So this guide is mostly sentences. Steal them, change the nouns, say them out loud.
For five to nine people looking back over two to four weeks, book sixty minutes. Ninety if the period was rough or the team is new. Under thirty minutes you are running a status update in a costume.
People arrive guessing at what the meeting is for and how safe it is. Answer both, explicitly, before you ask a single question. Something close to this:
That last one carries most of the weight, and only if it is real. “I could have communicated better” is not a confession, it is a shield. “I sat on the client's scope change for four days hoping it would go away, and that's why the last week was compressed” is a confession, and it changes what the room believes is permitted. That belief has a name: psychological safety.
A retrospective that produces no change is worse than no retrospective. It teaches the team that saying the true thing costs something and buys nothing.
The difference is almost always specificity. Vague questions get vague answers, and vague answers cannot be acted on.
| Instead of | Say | Why it works |
|---|---|---|
| “So, how did everyone think that went?” | “Let's start with the Tuesday of week two, when the build broke. Walk me through what you saw from where you were sitting.” | A concrete moment gives people something to describe instead of something to judge |
| “Any blockers?” | “What did you wait on last week, and how long did you wait?” | Turns a yes/no question into a measurable one |
| “We should improve communication.” | “Priya posts a two-line status in the channel every Thursday by 4pm, from this week.” | A person, a verb and a date is the only kind of action that survives the meeting |
| “That's not really what happened.” | “That's not how I saw it, which is interesting. Say more about what you were seeing.” | Keeps two accounts on the table instead of ending the discussion |
| “Who missed the handover?” | “At what point could we have caught this, and what would have had to be different?” | Moves from person to system without pretending the failure didn't happen |
| “Great, thanks everyone.” | “Three actions: Sam, Priya, me. Dates on all of them. First item next retro is whether they happened.” | Closing the loop is what makes the next retrospective honest |
Formats are scaffolding for the “gather what happened” stage. Rotate them; the same three columns every fortnight stops producing new information after about the fourth run.
Silence is almost never “nothing to say”. It is “I'm not going first”. These prompts move it:
Then the discipline everything else rests on: after you ask, stop talking. Count to ten. Managers fill silences at about the four-second mark and conclude nobody had anything to say. See active listening for what to do with the answer when it arrives.
Aim for the smallest change that would have altered the outcome. Ambitious process redesigns die within a week; “we move the standup to 9:30 so the Manila team isn't dialling in at 7pm” survives, because it is one decision, already made.
Four failure modes cover most of it. Each has a recovery move you can make in the room.
One person becomes the subject and the tone shifts from examining to prosecuting. Interrupt early: “We're three comments into one person's work now. I want to move to the process that let this reach a customer. What would have had to be in place to catch it?” Then take the individual conversation offline the same day, using SBI feedback. Do not leave it hanging.
You get shrugs and “yeah, fine”. Switch mediums: five minutes of anonymous silent writing, then you read the cards out. If it stays flat across two retrospectives, the problem is upstream of the meeting. Ask one person privately: “In the retro, what didn't get said?” The answer usually arrives in about thirty seconds.
Usually something outside the team's control: another department, a tool, a client. Name it and split it: “Fourth retro running, the data handover has come up. We can't fix their side from this room. So, two things: what makes their delay hurt us less, and what do you need me to take to their head of department, in writing, this week?” A recurring complaint you never escalate becomes proof the meeting is theatre.
You will not spot this from inside; it looks like consensus. Prevent it structurally: written positions before discussion, most junior speaks first, or hand facilitation to someone else while you take notes. In a new team, expect it. Unearned agreement is a signature of the early stages in Tuckman's model, and it burns off only once disagreement is visibly survivable.
Everything above is one skill wearing several hats: asking a question that makes honesty cheap, then not flinching at the answer. You can read that in four minutes and still not do it, because under mild social pressure people fall back on what they have practised, not on what they have read.
So this changes through rehearsal and feedback, not through articles. Run one badly. Have someone watch and tell you where you rescued the silence, where you defended your own decision, and where you took a vague action because the hour was nearly up. Then run it again on Thursday. That is what experiential learning means: behaviour shifts in the doing and the debrief, not in the reading.
Tour De Force runs this as live, experiential training for teams worldwide — online and in person. Talk to us, or play Gamified learning appsThe Weekly Challenge to see the method in ten minutes.
Questions
A short, structured meeting where a team examines how it worked over a recent period and agrees a small number of changes. It looks at process, collaboration and decisions, not at output or individual performance. It comes from software development but works for any team that repeats a cycle of work: sales, operations, marketing, events.
Sixty minutes for a two to three week period with a team of five to nine. Ninety if the period was difficult or the team is new to the practice. Under thirty minutes you cannot get past the first round of polite answers, and beyond ninety attention collapses and people start agreeing to things to end the meeting.
Specific ones anchored to real moments. "Walk me through the Tuesday the build broke." "What did you wait on, and for how long?" "What did you spend time on that didn't matter?" "What nearly went wrong and we got lucky?" "What did you disagree with and not say?" Avoid "how did everyone think that went", which reliably produces nothing usable.
Went well / didn't go well / try next is the fast default. Timeline works best after a messy period because it preserves the order of events. Start / stop / continue converts an agreed diagnosis into behaviour. Mad / sad / glad suits emotionally loaded periods. Sailboat looks forward as well as back. Rotate them: the same format repeatedly stops producing new information.
You can, but you have to work harder at it, because your reaction to the first piece of criticism sets the ceiling for the rest. Going first with a real admission of your own helps. If the team has gone quiet across two or three sessions, hand facilitation to someone else, or rotate it, and take notes instead of steering.
A retrospective is regular and rhythmic: every sprint, month or project phase, whether things went well or badly. A post-mortem is triggered by a specific event, usually a failure or incident, and digs into one causal chain. Teams that only run post-mortems learn exclusively from disasters, which is an expensive way to learn.
Keep reading
The individual conversation you take offline when a retrospective starts to circle one person.
GlossaryThe condition that decides whether anyone tells you what actually happened.
Free · 10 minutesPlay one crisis week and find out what your instincts do under time pressure.
Get in touch
Book a 30-minute discovery call and we'll shape an experiential programme around your goals.
Book a discovery call →