All posts

Survey Whether Last Retro's Actions Actually Happened

Why a short anonymous survey on last retro's actions tells you more than asking the room, how to read the spread of answers, and what makes an action that actually gets done.

Here is the version of this that most teams run. The facilitator opens the retro, pulls up last time's board, reads out three actions and asks "did we do these?" Someone says "mostly," someone else says "I think that one's blocked," and the group moves on in about ninety seconds. Everyone in the room now knows the actions did not happen, and nobody has said so.

The problem with asking out loud is that the owner of each action is sitting right there. Saying "no, that didn't happen" costs a colleague something, and most people on a decent team will pay a small lie to avoid it. The other half of the problem is that people genuinely do not know. If the action was "Priya quarantines the flaky integration test," four people have no idea whether that landed. In a verbal round they will stay quiet rather than admit they weren't tracking it.

Send it as a survey an hour before

I send one short form before the meeting, one question per action from the previous retro, four options each: done, partly done, not started, I don't know. Anonymous. One free text field at the end: what got in the way?

A vote would be wrong here. Voting asks the group to form a judgment, and I don't want a judgment, I want each person's independent observation. A topic column would be wrong too, because it invites the two people with the most to say to say it, and the rest of the room nods. The survey gets me seven separate readings of the same fact, and the disagreement between them is the useful part.

The spread tells you which problem you have. If five say done and three say I don't know, the work happened and nobody announced it. That is a visibility problem and the fix is thirty seconds in standup, not a new process. If everyone says not started, the action was bad. If two say done and two say not started, you have two different understandings of what was agreed, which means the wording was vague and you wrote it down anyway.

I put my own actions in the survey. When one comes back as not started, I say so first, before anyone else's item comes up. It costs me about four seconds of discomfort and it changes what the next ten minutes sound like.

When actions keep not happening

The instinct is to add accountability: a tracker, a nag, a standing agenda item. That usually makes the next retro quieter, because people learn that proposing an action means being chased.

The causes I actually find, in rough order of frequency: there were six actions and a team can carry one or two. The action had no named person, or worse, the team was the owner. The action required someone who was not in the room and never agreed to it. The action was a wish, phrased as a behavior change, like "we'll communicate earlier about schema changes." And the honest one, the action was never agreed at all, it was something one person said aloud while the facilitator was typing and nobody objected.

That last one is why actions have to trace back to the discussion. If I can't point to the moment in the conversation that produced an action, I cut it before the meeting ends. An action with no story behind it has no owner who cares about it.

The actions that survive to the next retro share a shape. They are finishable in less than two weeks. They have one name on them. They change an artefact or a default rather than asking people to be better: a checkbox on the PR template, a renamed alert, a deleted dashboard, a time-boxed experiment with a review date. "Ben adds a needs-schema-review checkbox to the PR template by Friday" gets done. "Be careful with migrations" gets surveyed as not started for three retros, and then the team stops writing actions, which is the expensive failure.

Cap it at two. Survey both. Read the spread.