1. LandOffer
  2. Blog
  3. All articles
  4. Run a Project Retrospective Without Turning It into a Complaint Session

Onboarding & Teamwork

Run a Project Retrospective Without Turning It into a Complaint Session

Turn retrospective concerns into inspectable work events, choose a capacity-backed experiment, and review mixed results while preserving disagreement and unresolved capacity questions.

9 min readLandOffer team

Run a project retrospective by turning broad reactions into specific work events, examining the process conditions around them, and choosing a small improvement the team can actually test. Give the experiment an owner, capacity, evidence to collect, and a review date. Let people disagree about the interpretation; a useful retrospective does not require a cheerful account of the project.

This LandOffer workplace guide follows a fictional team from complaints about review work to one bounded two-week experiment and a mixed follow-up decision. All people, work records, dialogue, commitments, and results are invented examples. No team meeting or workplace trial was conducted while preparing this article.

A facilitator and project team examining specific work samples and discussing a bounded improvement without forced agreement

Set a learning question with a clear boundary

A planned retrospective examines how the team worked over a defined period. Use the defined period to inspect work events, interactions, tools, and assumptions, including what helped and what caused difficulty. It is different from an incident postmortem investigating a particular service failure, and it should not become an individual performance-rating session.

In the fictional example, the team has completed an iteration revising in-app help panels. Facilitator Petra frames the question: “How did review requests reach the right person with enough context during this iteration, and what should we try next?” The team can inspect its own review process; it cannot use the session to promise staffing changes or set another group's priorities.

The 2020 Scrum Guide describes a Sprint Retrospective as inspecting interactions, processes, tools, assumptions, and how problems were handled, then identifying useful changes. That is a defined event within Scrum. A project team outside that framework can use a bounded learning question without claiming to follow Scrum's complete event structure.

Before the discussion, ask participants for concrete examples and provide a written contribution route. Include people who prepared, reviewed, and used the work. Tell them how notes will be shared and avoid promising confidentiality that the actual process cannot provide. Sensitive personnel concerns need their appropriate route, not a public exercise in naming who caused the most frustration.

Translate a complaint into inspectable evidence

Do not treat a complaint as useless because it is emotional or broad. It can identify a problem worth examining. Ask for the event and its consequence, then keep the speaker's interpretation open to checking.

Here is the completed fictional input grouping. The example references are invented review records rather than links to real employee work.

Initial contribution Specific event established in discussion Process question and retained limit
“Reviewers never answer.” Review request H-12 had no named reviewer; two colleagues each thought the other was handling it How should a request identify a reviewer who has accepted the work? This one record does not establish that reviewers never answer.
“Everything changed at the last minute.” H-15 received a new example after its original review; the request did not identify which version needed checking How should the author signal a new version and the specific question to revisit? The group has not established why the example changed.
“We keep explaining the same thing.” H-18 linked a discussion but did not state the review question; the reviewer asked for that question before proceeding What minimum context belongs in the request itself? Not every link requires copying an entire discussion.
“The examples session actually helped.” A short shared walkthrough established the expected help-panel behavior, and the note was attached to H-14 What should be retained from that preparation? One useful session does not imply every review needs a meeting.

Petra groups the first and third inputs under request readiness, and the second under version changes. She keeps the positive example as evidence about shared context. Grouping does not mean the events have the same cause, and the original comments remain available beside the summaries so participants can correct a misleading reduction.

A useful question is “What did the request contain when the reviewer first saw it?” A less useful response is “Who needs to communicate better?” The first can lead to an observable change. The second can invite general judgments without establishing what work behavior would improve.

Facilitate one disagreement all the way to a usable question

The following exchange is fictional. Petra does not require the team to soften its concern before it can be heard.

Hector: “We're being asked to add more fields, but the real problem is that nobody has time to review.”

Petra: “Which request shows the capacity issue most clearly?”

Hector: “H-12 sat until the end of the iteration. Even a better request might have waited.”

Sana: “I thought Tobin was reviewing it. I hadn't accepted it. Naming someone in a thread didn't establish that.”

Tobin: “I also thought it belonged to Sana. But I agree with Hector that a named reviewer could still be unavailable.”

Petra: “We have evidence of unclear ownership in H-12 and a separate capacity concern. Can we test whether a request with an accepted reviewer and a clear question makes the next step visible, without claiming it will shorten the queue?”

Hector: “Yes, if we record when someone cannot take the review, rather than count the filled field as success.”

Malik: “I can maintain a small request record for six reviews over two weeks, if Yvette confirms the time comes out of planned work.”

This exchange preserves the capacity objection. The three recorded items test question clarity, version visibility, and accepted ownership; they leave review capacity as a separate concern. A facilitator who writes “team agrees communication was the problem” would erase the useful distinction the discussion established.

Atlassian's retrospective play moves from patterns and insights to action items with owners and deadlines. Apply that step only after the group has identified an action within its influence. Limit the selected improvement until its owner, capacity, and review evidence are clear enough to act on.

Choose one experiment and account for its cost

The team considers three options in the fictional case. Requiring a full history on every request is rejected as too much administration for the question being tested. Adding a daily review meeting is deferred because the group has not established that another standing session would resolve ownership or capacity. It selects a small request-context experiment instead.

The hypothesis is precise: making the review question, current version, and accepted reviewer visible will help the team identify what should happen next. It is not a claim that every review will be faster or that an unavailable reviewer will acquire more time.

Filled experiment agreement: For the next two weeks, Malik will track six ordinary help-panel review requests. Each request must state the question to answer, identify the current version, and record a named reviewer's acceptance or an explicit unassigned state. Changes after review must identify what changed and which question needs another look. No sensitive personnel commentary goes into the shared record.

Ownership and capacity: Malik owns setup, the six-entry record, and the follow-up summary. Team lead Yvette confirms a two-hour allocation by deferring Malik's optional screenshot refresh for this cycle. Authors and reviewers retain their existing work responsibilities; the experiment does not silently assign all requests to Malik or increase another person's review capacity.

Evidence and boundaries: Record whether the three items were present before review, whether the reviewer needed a clarification, and whether a capacity or reassignment problem appeared. Keep the six request identifiers and short observations so counts can be checked. Do not treat a completed field, an emoji, or a tagged name as accepted ownership unless the reviewer actually confirms it.

Review and stopping rule: Petra will facilitate the follow-up on Friday at the end of the second week. Continue only the parts that help identify a next step without creating disproportionate administration. Revise or stop the experiment if people still cannot identify the current version, if ownership is merely nominal, or if maintaining the record exceeds Malik's agreed allocation. The team has not approved a permanent process change.

GitLab's group retrospective guidance describes gathering facts, generating insights, selecting actions or experiments, and assigning completion expectations. Those are that employer's practices. They do not authorize the fictional team to change other groups' review obligations or establish causal conclusions from a small trial.

Colleagues reviewing a small experiment's work samples and keeping an unresolved issue visible beside completed observations

Review the mixed result after two weeks

The following completed follow-up is fictional. It uses the six agreed review requests rather than replacing the sample with only the successful cases.

Request What the record showed Follow-up interpretation
R-1 Question, version, and reviewer acceptance present; reviewer answered the stated question The next action was identifiable for this request
R-2 All three items present; author marked a later version and its changed question The version change remained visible rather than silently replacing the reviewed material
R-3 All three items present; reviewer asked one substantive design question A question during review is not automatically evidence that the request lacked context
R-4 All three items present; review proceeded on the recorded version Another usable request record, without a measured queue-time conclusion
R-5 All three items present; reviewer was reassigned after accepting, and the response arrived later than expected Clear ownership did not prevent a capacity change; reassignment handling needs work
R-6 Review question and acceptance present, but the current version was missing The reviewer needed a version clarification before continuing; the experiment was not followed consistently

Five of the six requests had all three readiness items before review. That count describes this sample; it does not establish a success rate for future work. R-5 shows why Hector's capacity concern still matters. R-6 shows that adding a field does not ensure people use it.

Malik's fictional effort record totals seventy minutes: twenty for setup, thirty across the six entries, and twenty for the summary. This is within the agreed two-hour allocation. The record does not measure the authors' or reviewers' additional effort, so the team cannot claim net time savings or improved productivity from those numbers.

At follow-up, Sana supports keeping the concise review question and version pointer. Hector still objects to implying that request readiness fixes the queue. Tobin asks for a visible reassignment state rather than leaving an old acceptance in place. Petra records all three views instead of describing the discussion as unanimous approval.

Make a bounded continue, change, or stop decision

The fictional team continues the three-item request format for one further two-week cycle, with one revision: if an accepted reviewer becomes unavailable, the author marks the request unassigned and asks Yvette to route the ownership question through the team's normal planning process. Nobody is automatically reassigned merely by being named.

Yvette confirms one hour of Malik's time for maintaining the next six-entry record and summary, again by leaving the optional screenshot refresh deferred. Malik accepts that narrower continuation; no new setup is planned. Petra owns the Friday review at the end of the next two-week cycle, using the six request records and the maintenance-effort note. If the record cannot fit the hour or adds no useful visibility, the team will simplify or stop it rather than extend it by habit.

The separate capacity question remains open. Yvette will bring R-5 and the existing review-demand record to the next planning discussion. Staffing and queue expectations remain subject to the separate planning decision. Its result is a small revised practice, an owned open question, and evidence to inspect next time.

Before closing your own retrospective, read back the chosen experiment, its owner, the work displaced to fund it, the observations to collect, and the review date. Ask whose concern the summary still fails to represent. Save the experiment agreement, dissent, displaced work, and follow-up observations together, including the decision to continue, revise, or stop. More workplace guides are available through LandOffer.

Sources and Further Reading