In order to generate the best results from a retrospective, the following elements must be present:
A safe environment for feedback and discussion
A plan for advancing discussion from facts to conclusions
Not every retrospective will require both of these elements - for example, an iteration where "nothing went wrong" may see retrospectives that don't benefit from a safe environment. Similarly, an iteration that has nothing but "obvious" problems will probably have a productive retrospective even without a strong plan for the discussion. However, even iterations that seem fine on the surface can harbor deep problems, so it is always best to be prepared.
Retrospectives are inherently conversations about what went well and what went wrong in a project or iteration. While it's easy to talk about what went well, what went wrong can be a touchy subject. Without a safe environment, issues may go unmentioned and the team won't improve as rapidly as they otherwise could. In order to collect as much feedback as possible, we recommend that you observe the following:
All retrospectives should have an impartial moderator. This moderator actively attempts to remain objective and is focused on guiding the conversation forward throughout the meeting. If participants feel that the moderator has a stake in the conversation, they may feel as though it is not safe to voice dissent or share concerns. Normally the manager should act as this moderator - however, in some circumstances the manager may feel a strong need or desire to participate in the conversation. In that case they should find a peer, Director, or other member of the team willing to moderate.
Emotions are not only allowed in retrospectives, they should be encouraged. Emotional language ("I was angry when…," "It made me sad that…") not only helps convey intensity, it also helps expose issues that may have been difficult to sort out otherwise. Make sure all participants know they are free to express their feelings, although we do expect them to stay consistent with our values.
When possible, all parties should be present. No one should have to worry about sharing concerns or experiences because another party isn't there to represent their side of the story. This includes stakeholders outside of the team, where necessary - if someone's role or contributions are going to be discussed, they should have the opportunity to contribute to the retrospective as well.
When necessary, get people face to face. After a particularly difficult iteration, or when there's a strong risk that emotions will be running high, it's almost always worth the cost to have everyone in a video meeting to talk through the retrospective in real-time.
Having a plan
It's easy for retrospectives to go off the rails if there isn't a good plan for collecting actionable insights. If the moderator doesn't guide the conversation successfully, the retrospective could be dominated by a few "loud voices" or could stay focused on the facts and feelings about the past iteration without drawing any conclusions. To make sure the conversation is productive, we recommend that you have a clear agenda for the conversation. The moderator for the retrospective is then responsible for ensuring the agenda is followed and completed. While the specific nature of this agenda will vary from team to team, we recommend something that follows this general pattern:
Introduction - remind people of the purpose of the retrospective, and that the conversation should be scoped to the project/iteration under review. Ensure everyone understands the rest of the agenda.
Gather data - don't try to draw any conclusions up front, simply collect facts. This can be done by constructing a timeline, soliciting impressions ("what made you mad, sad, or glad in this iteration?," etc.), or any other number of fact collecting exercises.
Generate insights - now that you have all of the facts, try to work together to identify patterns or causal relationships (because we did x, y happened).
Decide what to do - with the insights firmly in hand, it should be fairly easy to identify action items for the team. This could be things to change with our process, it could be things to experiment with in the next iteration, or any other number of things.
Close the retrospective - make sure everyone who participated receives positive feedback about how they helped create these results. Observe any surprising or unexpected results from the retrospective, which confirms that the retrospective was valuable. Make sure all of the action items are appropriately assigned to one or more team members with clear expectations for when they should be completed.
Right now we are leaving it to individual Engineering Managers how they would like to collect retrospective notes - GitLab issues, Google docs, etc. After the retrospective is complete, the Manager has until the public retrospective call (shortly after the 22nd every month) to make relevant notes from the retrospective public. Teams are encouraged to link to these notes from their team page in the handbook.
We recommend the following resources if you'd like to learn more about running effective retrospectives: