How to Run a Start-Stop-Continue Retro After a Release

How to Run a Start-Stop-Continue Retro After a Release

The release is out, the dashboards are quiet and the team is already eyeing the next ticket. This is the exact moment when most lessons evaporate. A start-stop-continue retro is the lightest way to catch them before they do. What follows is a numbered routine for the end of any sprint or release, with concrete examples of what each list tends to contain on an engineering team. The whole thing takes roughly forty-five minutes and needs only a shared document.

Step 1: pick the slice of work

Decide what the retro covers. One sprint, one release or one migration works well. Avoid “the last six months”, because nobody can recall that much detail. State the scope in one opening line of your retro document, for instance “the checkout service release and the two weeks leading up to it”. Everyone then knows which events count.

Step 2: collect the raw material first

Before the meeting, ask each person to jot down three moments they remember: one that went well, one that was painful, and one that surprised them. Pull in the evidence you already have, such as the merged pull requests, the incident notes and the chat thread from release day. A short pre-read keeps the conversation anchored in events, not impressions.

Step 3: set up the three columns

Create a document with three headings: Start, Stop and Continue. Under each heading, leave room for a one-line entry and an owner. Some teams add a fourth empty slot labelled “needs a decision” for entries that require someone outside the room to act. That keeps the three main lists realistic.

Step 4: fill Start with concrete behaviours

Give people five quiet minutes to type entries. Good Start items for a technical team look like these.

  • Start running the database migration on a copy of production data before each release.
  • Start writing the rollback steps into the ticket before the work begins.
  • Start pairing on the first hour of any task that touches an unfamiliar service.

If you are wondering where repetitive manual steps hide, this explanation of how automating routine tasks trims wasted time and slips is a handy prompt. Scan your last release for any step a person performed by hand more than twice, and consider whether it belongs on the Start list as an automation task.

Step 5: fill Stop with practices, never people

Stop entries describe habits that cost the team more than they returned. Typical examples include the following.

  • Stop merging large changes on the afternoon before a freeze.
  • Stop discussing production issues in private messages where the rest of the team cannot see them.
  • Stop adding configuration flags without a note on when they will be removed.

Phrase each item about the practice. “Late merges made the release harder to review” is useful. A sentence naming a colleague is not. If the conversation turns tense, pause and ask the group to restate the entry in terms of the process.

Facilitation tips for step 4 and step 5

Timing matters more than most teams expect. Cap each column at ten minutes, and use a visible timer so nobody feels singled out when it ends. If a topic runs long, write it in the parking area and move on; you can return to it once every column has been heard. The facilitator should speak last in each round, so their view does not steer the others.

Sprint retros and release retros differ slightly in flavour. After a sprint, entries often concern planning and handoffs. After a release, they concern testing, deployment and what happened once real users arrived. Naming which kind you are running helps people calibrate the level of detail they bring.

Step 6: fill Continue with what quietly worked

This is the column teams forget. Ask what saved you during the release, and write it down so it survives the next reshuffle. Examples include a short daily sync during the final week, a habit of reviewing infrastructure changes with a second person, or a smoke test that caught a broken login before users did.

When traffic is a factor, it is worth recording what held up under load. Some background on how modern cloud setups cope with surges in web visitors can help you decide whether the scaling choices you made deserve a spot under Continue or a rethink under Start.

Step 7: narrow to three and assign owners

Read all entries aloud, group similar ones and then vote. Each person gets three dots to place on the entries they care about most. Keep the top three per column. For every surviving entry, attach three things: an owner, a first action, and a date. An entry with no owner is a wish. For example, “Start writing rollback steps into the ticket” becomes “the tech lead drafts a ticket template with a rollback section by Friday”.

Step 8: look sideways for ideas

Teams working in a single domain can grow blind to how others structure their work. If you like gathering ideas from unrelated fields, browse a site such as BestPilotsNow, an aviation discovery platform with pilots and flight schools among its sections, and see how a sprawling subject is carved into separate, easy-to-enter areas. It is only a prompt for fresh thinking, but it often sparks a Start entry about how your own documentation is organised.

Step 9: close the loop

Store the page where the team already looks, such as the project wiki or the pinned message in the team channel. Open the next retro by reading the previous page aloud and marking each entry done, dropped or in progress. This five-minute habit is what turns a one-off meeting into a real improvement cycle, and it is the first thing to go when schedules get tight, so protect it.

A quick example of the finished page

Here is what a completed page might look like for a small team after a release.

  • Start: rehearsing migrations on a production copy; adding rollback notes to tickets.
  • Stop: merging big changes right before a freeze; debating incidents in private chats.
  • Continue: the short daily sync in release week; the second-person review for infrastructure changes.

Six lines, six owners, six dates. That is all it takes. Each line is small enough to finish in a week, which is why the team will actually finish them.

Keeping the retro honest over time

Every few cycles, audit the audit. Count how many entries from the last three retros were completed. If fewer than half were, the team is writing wishes instead of commitments, so shrink the lists to two each for a while. If nearly all were finished, you are probably choosing safe items; invite a bolder entry or two and see whether the team can stretch. Either way, the retro should be judged by what changes afterward, not by how pleasant the meeting felt.

Troubleshooting a flat retro

If entries come out generic, ask for an example. “Communication was poor” becomes “we found out about the schema change from the build failing”. Specific stories produce specific changes. If the room feels cautious, try anonymous input through a shared form. And if you notice the same entry appearing in several retros, the item may be too big; break it into a first step that fits inside a single sprint.

Repeat the routine after every release for a few cycles. By the third round you should see the Start list shrinking and the Continue list growing, which is the clearest sign that your delivery habits are settling into something dependable.