How to do a gemba walk: from observation to verified action

Why it matters

A report saying material delay does not explain where the delay occurs. Material may be ready at the entrance but not delivered to the station. A gemba walk involves going to where work happens, observing actual flow and understanding it together with the people doing the work. This is not a visit to confirm a decision already made. The point where a problem becomes visible may differ from where it originates. Therefore, select the observation question and route together. If investigating why the first product leaves late, examine startup preparation and material arrivals. Walking only through the cleanest or fastest section may not answer the question being investigated.

Drawers being opened until a wrench is found is an observation. Saying the worker is disorganized is an interpretation. The worker may explain that the tool is shared across shifts; this explanation is a valuable investigative lead, not yet a verified cause. A good shop-floor record keeps what was seen, what was said and what was inferred separate. Recording time, location, product and event boundaries allows someone else to check the same finding. Unmeasured time is marked as an estimate; unknown sections are not filled with zero. The certainty of an explanation must not exceed the strength of the evidence. A concrete event record, rather than an opinion about a person, is a more useful starting point for reliable learning and a shared solution.

Taking many notes during a walk does not mean collecting much evidence. The same event may have been written down several times; some notes are interpretations and others incomplete records. Before calculating the rate of events supported by evidence, check the uniqueness of source records and the classification criterion. Having a photograph alone does not prove duration, recurrence or cause either. State which evidence supports which claim. In this lesson, the rate shows how many of the reviewed records meet the defined evidence requirement. It is not a defect rate or a reliability score for all production. A low rate may indicate a need for more observation; it does not show that workers do their jobs poorly.

The impact of the walk is understood through implementing a proposal and following up the result. Work described as done may not have produced the expected result. Define closure conditions in advance: implementation must be seen, the result measure checked and safe working preserved. Follow-up easily becomes unclear if the owner and date are not explicit. Selecting a small number of feasible trials is more instructive than opening many issues and verifying none. If a positive result is not visible, reexamine the assumption rather than forcing the record closed. A gemba walk then becomes a way of working that links observation, trial and verification, instead of a visit measured by the instructions given.

Formula

How to do it

1. 1

   Choose an investigation question and a safe route

   Write an observable question such as why the first conforming product is delayed. Define the relevant product, shift and process boundary. Choose where to start according to the question. The figure shows a route from the customer end upstream; this route is not required for every question. Learn the entry rules and safe observation position. Do not block material routes, the worker's view or emergency access. Do not allow a hazardous situation to continue for better measurement; use the appropriate safe intervention procedure. Clarify the permissions and notification requirements for observation records in advance. A narrow investigation question does not mean ignoring safety or other important findings.

2. 2

   Explain the purpose and observe normal work

   Explain to the people doing the work why you are there and what you are trying to understand. Observe normal conditions rather than testing a worker or watching work arranged only for demonstration. Wait for a safe and suitable time before interrupting an activity. Do not assume one repetition represents the entire situation. Record changes in product type, material supply and starting conditions. If you notice work being performed differently because an observer is present, explain this limitation. Preserve the worker's ability to stop, ask questions and report difficulties. Use codes such as K01 and K02 for people; focus on working conditions without collecting unnecessary personal information. Apply the same principle during follow-up visits.

3. 3

   Record observation, explanation and interpretation separately

   Describe what was seen in the note using a concrete verb. Add the location, time, product being followed and event start and end. Label what the worker said as an explanation and your possible cause as a hypothesis. Before calling a record supported by evidence, show which evidence requirement it meets. Seeing the same event again in another note must not count as a new event. Do not create agreement by deleting contradictory records; investigate why they conflict. Unmeasured time remains unknown. Even if an event's existence is supported by evidence, its duration may be unknown; do not treat these as the same deficiency. The scope of the claim must match the scope supported by the evidence.

4. 4

   Verify the explanation with the worker and source records

   Ask open questions such as when material arrives and in response to which request. Do not declare the first answer the root cause. Read your note to the worker and check that the event has been understood correctly. Test the explanation against the previous process's record, another direct observation or relevant timing information. If sources say different things, keep the difference visible. A photograph may show the layout; it does not prove the repetition count across a whole shift. Explain what the evidence demonstrates. Workers' knowledge helps develop the change to be tested; treating a person's explanation as a hypothesis to test does not mean distrusting them. The aim is to strengthen shared understanding. Record the need for further observation when a matter cannot be verified.

5. 5

   Summarize record quality with the correct denominator

   Count all reviewed records, then separate those meeting the defined evidence requirement. Show interpretations and incomplete records separately. Do not silently remove records from the denominator. If the review boundary changes, the new rate must have an explicitly different name and scope. If record counts differ by shift, the simple average of shift percentages may not give the overall rate. Calculate from the total numerator and total denominator. This rate describes the evidence status of observation records, rather than the size of process losses. Selecting a priority also requires other data such as time, frequency, conformity and safety. Counting records does not replace measuring event impact.

6. 6

   Define an owner, date and closure condition

   Record which condition a trial will change, who will implement it and when it will be checked. Identify the owner as a function or a code such as K02; do not use an unclear term like everyone. Check implementation separately from achievement of the expected result. If material supply changes, monitor wrong parts and safe access as well as waiting time. Setting closure criteria in advance prevents every apparently positive situation at the end of a visit from being counted as a success. Work reported as done but whose result has not yet been measured stays open. If delayed, record the reason and new follow-up date. Changing the date does not erase the previous delay from the record.

7. 7

   Measure the result on site and share the decision

   Take follow-up measurements with similar products and starting conditions. Compare before and after records in the same unit. Shorter waiting alone does not prove the change is lasting under all conditions. Investigate the effect of different products or shifts. Share the result with workers and listen to daily difficulties in using the change. If the expected result has not occurred, reconsider the hypothesis; do not change the criterion afterward to make completed work look successful. Keep closed, implemented but unverified, and not yet implemented actions separate. This ensures the follow-up rate represents actual verification. The next visit then aims to resolve remaining uncertainty rather than repeat the same instruction.

Worked example

When the method misleads

Common mistakes

• Counting judgment as observation: Record the activity seen, conditions and time instead of applying adjectives to a worker or department; separate a possible cause as a hypothesis.
• Zeroing an incomplete record: Do not enter zero for unknown time; assess an event's existence separately from knowing its duration, and state the additional evidence needed.
• Stopping at the first answer: Treat a worker's explanation as valuable investigation input; do not declare it a definite cause without verification through another source or repeated observation.
• Confusing a completion report with closure: Done reports implementation; do not close an action or reduce the open follow-up count before verifying the result criterion on site.
• Making the route the purpose: Do not treat walking the same route on every visit as success; select points suited to the investigation question together with safe access conditions.