Why Most Retrospectives Produce Documentation Instead of Learning

Most project teams hold retrospectives. Far fewer can name a process change that originated in one and was still in place three months later.

The gap is rarely a matter of attendance or goodwill. Teams show up, answer the three standard questions (what went well, what did not, what could improve), record the answers, and return to work.

When the next cycle surfaces the same issues, the session has produced documentation without producing learning.

The Agile Manifesto frames the purpose of the practice in its twelfth principle:

"At regular intervals, the team reflects on how to become more effective, then tunes and adjusts its behavior accordingly."

The principle carries two actions of equal weight, reflecting and adjusting. Most retrospectives deliver the first and skip the second.

Why retrospectives stay shallow

Donald Schön, in The Reflective Practitioner (1983), distinguished reflection-in-action from reflection-on-action. The second, the deliberate interpretation of experience after the fact, is where professional learning takes place.

Without it, teams accumulate events without drawing conclusions from them.

Deliberate interpretation carries a psychological cost. Findings that threaten self-image, such as a flawed plan or a failed communication, trigger cognitive dissonance, the tendency described by Leon Festinger to resolve discomfort by rationalizing or attributing causes externally.

In a group setting the result is predictable: discussion stays at the level of symptoms, and vendors, tools, or shifting priorities absorb the blame.

The corrective is structure. A predictable, transparent process signals that the session is fair and that its target is the system. Psychological safety, in this sense, is partly a product of design.

A structure that produces change

Esther Derby and Diana Larsen, in Agile Retrospectives: Making Good Teams Great (2006), propose a five-stage model that remains the most widely used reference for facilitation.

• Setting the stage uses a brief check-in and a restatement of purpose to move participants from task mode into reflection and to establish that the process is under review.
• Gathering data builds a shared timeline of releases, blockers, decisions, and incidents, supported by metrics, so the discussion is grounded in events before opinions enter it.
• Generating insights applies techniques such as the Five Whys, drawn from the Toyota Production System, to move the group from what happened to why it happened, past the first plausible explanation.
• Deciding what to do limits the team to two or three priority issues, each with one specific, time-bound action and a named owner.
• Closing the retrospective acknowledges contributions, captures individual takeaways, and confirms when the agreed actions will be reviewed.

The sequence matters. Skipping data collection turns insight generation into an exchange of impressions, and compressing the decision stage tells participants that the discussion carried no consequence.

Two conditions determine whether this structure works. The first is safety. Amy Edmondson's research on psychological safety found that higher-performing teams reported more errors, because members felt able to discuss them openly.

Facilitators reinforce this condition by stating Norm Kerth's Prime Directive at the start of each session:

"Regardless of what we discover, we understand and truly believe that everyone did the best job they could, given what they knew at the time, their skills and abilities, the resources available, and the situation at hand."

When disagreement arises, the principle from Roger Fisher and William Ury's Getting to Yes applies directly: separate the people from the problem. A facilitator who names the dynamic ("there is a lot of frustration around this topic") without attributing it to an individual keeps the conversation on the system.

The second condition is follow-through. Agreed actions belong in a visible improvement tracker, each with a single owner, and the next retrospective opens by reviewing their status before any new topic is raised.

This mirrors the Plan-Do-Check-Act cycle from quality management, and a retrospective without that review step has no Check and no Act.

Rotating formats such as Start-Stop-Continue, the 4Ls, or Mad-Sad-Glad helps sustain engagement over many cycles. These formats are secondary, and their value depends entirely on whether the underlying structure is in place.

For early-career project managers, the practical implication is direct. The quality of a retrospective is measured by the actions still active a month later, and the facilitator's primary responsibility is protecting the two stages most often compressed: data gathering at the start and decision-making at the end.