KPT retrospective
Separate practices to keep, observed problems, and changes to try, then agree how to check the next experiment.
Use Keep, Problem, Try after a shared activity when a discussion risks ending with vague advice. Record participants' observations, select practices worth continuing, describe the problems, and choose a manageable change for the next cycle. The useful output is a follow-up people can review together.
A defined activity or period, observations from participants, relevant results, and constraints on the next cycle.
An editable three-column retrospective with practices to keep, problems to investigate, and proposed experiments with review details.
On this page
How to use it
- Agree on the activity and period being reviewed. Invite the people involved to contribute observations, and distinguish recorded events from feelings or explanations.
- Under Keep, name specific practices worth continuing and the observations supporting them. Do not attribute every good outcome to those practices.
- Under Problem, describe what happened and its impact without turning a suspected cause into a fact or blaming a person.
- Under Try, propose a small change that responds to a Problem or extends a Keep. State the owner, timing, and check; mark any missing agreement as unconfirmed.
- Choose a manageable set of experiments. At the next review, inspect what happened and decide whether to continue, adjust, or stop each one.
Keep its limits in mind
KPT organizes a discussion; it does not establish root causes, assign blame, or prove that a proposed change will work. The worksheet does not assign tasks or send reminders. A Try remains a proposal until the responsible people agree to it, and missing details remain unconfirmed.
Test a replenishment routine at the next event
Fictional example. Four Qinghe staff members review a weekend member event. There were 60 reservations, 42 attendees, and two sample stockouts. Reminders ran smoothly, while on-site assignments changed at short notice.
On a small screen, scroll across the diagram to read it.
An observed operational gap: Two stockouts and an unassigned replenishment role give the team a concrete issue to address. The retrospective has not established every cause of the shortages.
One testable next step: The proposed stock checks respond to that gap. A review by the store manager does not mean a replenishment owner has already accepted the work.
Next step: Confirm the replenishment owner and how to record waiting time before the next event, then use the recorded results in the next retrospective.
Three reasons to slow down a retrospective
Use KPT when people need to move from a shared activity to a manageable next experiment. Choose a period that matches the evidence. A quarterly meeting can examine one event, but that event’s counts cannot stand for the whole quarter. Ask participants what they observed, which practices seem worth keeping and what needs investigation before choosing a change.
Pause if the activity or period is still disputed. A colleague describing setup work and another describing customer service may be reviewing different scopes. Pause if Problem becomes a list of personality judgments. Replace “unreliable staff” with the observed event, impact and missing agreement. Pause if Try contains more work than people can accept. Select a small set and retain unchosen proposals, rather than turning every suggestion into an assignment.
Keep a practice separate from the outcome it accompanied. Smooth reminders and good attendance can appear together without proving a causal link. Similarly, an unpleasant event does not by itself establish its cause. The purpose of the three columns is to support a better conversation and follow-up record, not to manufacture a success story or find someone to blame.
Choose discussion, diagnosis or follow-up
A retrospective can reveal that the next useful work is an investigation rather than an experiment. Choose a structure based on what is missing, and return with its results instead of trying to settle every question during one meeting.
| Method | Start with | Useful output |
|---|---|---|
| KPT | One activity, participant observations and constraints | Practices, problems and a proposed next experiment |
| 5 Whys | One bounded incident and inspectable records | Explanations tied to evidence, with unresolved links |
| PDCA | A chosen change and an agreed observation plan | A documented cycle of execution, checking and adjustment |
| RACI | A checkable deliverable crossing several roles | A responsibility draft requiring authority and acceptance checks |
Worked continuation: stockouts are known, waiting time is not
Use the fictional Qinghe event above as a bounded item in a quarterly review. Four staff members reviewed one weekend member event: 60 reservations, 42 attendees and two sample stockouts. Reminders ran smoothly; assignments changed at short notice. No replenishment owner was assigned, and staff noticed the shortage after a queue formed. Waiting-time measurements are not supplied. These are teaching inputs, not an audit of an actual store.
Keep the raw records separate from the proposals. Attendance is 42/60 = 70% of reservations in this event. It does not measure all visitors, explain absences or show that reminders caused attendance. Two stockouts are a count, not a stockout rate unless the relevant opportunities or exposure are defined. A queue being noticed does not supply its duration or the number of people affected.
Build the review around a specific gap: replenishment had no named owner and shortages were noticed after the queue formed. Keep the broader cause question open. The proposed response is an owner plus checks every 30 minutes at the next event. The manager’s proposed review role does not establish that someone has accepted the replenishment work.
| Column | Supporting material | What the review can say |
|---|---|---|
| Keep | Reminders ran smoothly; address instructions were clear | Continue the practice provisionally, without attributing attendance to it |
| Problem | Two stockouts; ownership absent; queue noticed | A concrete operational gap, with waiting time still unknown |
| Try | Proposed owner and 30-minute stock checks | A next-event test, pending owner, capacity and recording agreement |
| Later review | No next-event observations supplied | Implementation and outcome remain unreported |
Check the action and its effect on separate lines
Before the next event, ask who can accept replenishment responsibility, what a stock check includes and how an exception will be handled. Agree how to record scheduled and actual checks, stock availability, stockout start and end, and customer waiting where it can be observed appropriately. Do not invent waiting times to complete the baseline. Record the extra staff effort of checking stock as well as the intended benefit.
At the next retrospective, first ask whether the owner and checks were actually in place. Then inspect outcomes using a comparable scope. A hypothetical fall from two stockouts to one would not by itself prove the routine worked: attendance, sample quantities and event length could differ. That hypothetical comparison is a review question, not a reported follow-up result.
Decide whether to continue, adjust or stop using the observations and constraints. If the action was not implemented, report that and investigate the obstacle before evaluating its effect. If waiting time is still missing, retain the gap. In a quarterly summary, identify this as one event-level experiment and state which other activities were not assessed. Preserve the original Try and append the later decision, actual agreement and reasons.
Check your work
- All three columns refer to the same activity or review period.
- Keep names a repeatable practice; Problem describes an observed event and separates causal assumptions.
- Each Try connects to a stated Problem or Keep and describes an observable change.
- Ownership, timing, and review measures are agreed or explicitly unresolved.
- The team records experiment results later rather than treating a proposed Try as completed work.
Framework references
References explain the framework and its scope; they do not validate the fictional example's results.
Put the framework to work
Copy the outline and add your question, evidence, and open questions.
Read a practical guide →Practice with an example
Choose a preparation route for progress, a resource or decision request, or a period review. Use one service case to inspect openings, review checks and focused guides, and keep a preparation note.
Prepare a work update: start with the purpose →