Close the loop between a finding and a result
A chart doesn't improve delivery by itself. A slow stage, a data gap, or low use can remain on a dashboard because nobody owns the next step. Use records and team feedback to choose a change, assign the work, and return to decide whether it helped.
Large transformation backlogs can hide the same problem. “Improve developer productivity” has no finish line, while a list of requested tools can lose the delivery problem that led to it. Each improvement needs a clear condition, an owner, a result, and a review point.
Not every finding is ready for action. A weak signal may need more investigation, and a valid problem may not be important enough to address now. Record that decision instead of creating work merely to clear a dashboard.
Make the next step clear and achievable
Start with what you can inspect. Move from a trend to the issues, merge requests, pipelines, deployments, incidents, settings, or feedback behind it. Ask the people who do the work whether the sample is normal and what changed. Write the finding as a condition, not a preselected fix.
“Production deployments from the external release tool are missing from GitLab” can be checked. “Build a dashboard” assumes a solution before the data path is understood.
Give every finding the team agrees to act on a place in its existing backlog. A GitLab issue is often enough. An epic can group several independent changes that share an outcome. Another planning record can work when the team already uses it, as long as the supporting records and decision remain easy to find. Keep enough detail for the owner and contributors to understand the next step:
| Field | What to record |
|---|---|
| Problem and support | Current condition, affected work, and links or summarized feedback that support it. |
| Starting point | Scope, period, current result, and known data gaps. |
| Expected result | The user, delivery, quality, security, or risk change the team wants. |
| Trial | A small change that addresses the problem, tried with a project or team that follows the normal process. |
| Owner and dependencies | One person responsible for the next step, contributors, and work that must happen first. |
| Checks and reversal | Measures reviewed beside the result and how to undo or contain the change. |
| Review point | Date or event when the owner returns with results and a keep, change, stop, or expand decision. |
Choose work that has value on its own and can be completed with the people and time available. A representative trial gives the team a way to challenge its explanation of the problem before committing to a wider change. Decide what you'll inspect at the review and how you'll contain an unsuccessful trial. If the trial improves the intended result, expand to similar projects, checking dependencies and results again in each wave.
Consider missing deployment records. GitLab shows fewer production deployments than the release system, and a sample confirms that an external tool doesn't create GitLab deployment records. The backlog record links both samples, names the affected project and environment, and states that the current DORA result is incomplete. The trial is to follow GitLab's external deployment tracking for one normal project. The expected result is that successful production releases have matching GitLab deployment records and any difference can be explained. The checks confirm that failed deployments are not counted as successful and the environment uses the intended deployment tier.
At the review point, the owner compares the two systems again. If the records match, the team can plan a wave of similar projects. If they don't, the record keeps the remaining gap and the team changes or stops the approach. Installing the integration is an action; agreement between the records is the result.
When several improvements compete, discuss business or delivery value, risk, strength of support, effort, dependencies, and available owners. Keep those factors visible so the ordering can be understood and reconsidered. If no one has time to own the next step, describe the work as deferred or still being investigated; an assignment without capacity won't make it happen.
Treat follow-up as part of the improvement
Investigation and follow-up use time that could go to feature work. Choose improvements by likely value and risk, and make sure the owner has time to complete the trial and review. Keep the backlog small enough that its owners can return to the work and review the results.
Some findings are valid but can't be addressed now. Record the reason, owner, and next review point. For an unsuccessful trial, record what was tried, what happened, and what the next team should know.
Keep sensitive operational records and interview notes in a project with suitable access. Put only the records and context needed to understand the finding in the backlog record, and summarize personal feedback so it doesn't identify a participant without their agreement.
The owner is responsible for returning with results, not for doing all the work alone. Name contributors and dependencies, especially when a platform integration, security policy, product team, or service owner must take part.
Expansion is a new decision. A trial can work for one branch model, runner setup, or release path and fail elsewhere. Group similar projects into waves, define checks and a reversal action for each wave, and carry lessons forward before increasing variation.
Questions to ask
- What examples and source records support this finding?
- Is the finding ready for action, in need of more investigation, or accepted for now?
- What result should change, and what must not get worse?
- Who owns the next step, and which people or dependencies are needed?
- What is the smallest normal scope that can test the change?
- How can the team undo or contain the change if the trial fails?
- When will the owner return, and what decisions can be made then?
- What did the follow-up cause the team to keep, change, stop, or expand?
- What did an unsuccessful trial teach, and where is that learning kept?
Further Reading
A value-stream improvement cycle shows how to give an improvement an owner, try it in normal work, and decide whether to expand it.
Choosing measures that guide improvement helps you establish the starting point and review the result alongside unwanted effects.
Issues and epics provide GitLab records for focused work and related groups of work.
Track deployments of an external deployment tool supports the worked data-path example.
Description templates explains how to prompt people for the information needed to investigate recurring improvement issues.