01 · Connect stability with learning
Connect stability with learning
A continuous improvement process is the recurring way an organisation notices gaps, tests better methods, learns from results and updates its operating standard. It continues after an individual DMAIC project closes. Control does not mean freezing the process forever. It means preserving an understood method while changing it deliberately, so the team can distinguish improvement from uncontrolled variation.
Use a cycle such as Plan–Do–Check–Act to organise learning. Plan defines the problem, prediction and test; Do carries out a bounded change; Check compares the observed result with the prediction; Act decides whether to adopt, adapt or abandon the change. The cycle should produce an explicit learning decision. Repeating activities without examining evidence is not meaningful improvement.
Separate immediate recovery from experimentation. A parcel at risk of missing collection may require authorised containment now. A redesigned release sequence needs a planned test. Both matter, but emergency improvisation should not become the default improvement method. Reliable standards and a safe route for suggestions give people a practical basis for learning.
02 · Choose a worthwhile, testable question
Choose a worthwhile, testable question
Use customer gaps, recurring abnormalities, operator observations and strategic priorities to choose the next problem. Describe the condition and its consequence before naming a solution. “Buy another scanner” is a proposed action; “verification queues exceed safe lane capacity during shift overlap” is a process condition that can be investigated. Check whether the apparent problem is supported by consistent definitions and representative evidence.
Develop a prediction linking the change to the outcome. If two shifts wait for one supervisor to approve reprints, a clearer authorised decision route may reduce waiting. The prediction should include an expected result and the conditions under which it should occur. Identify balancing measures such as release accuracy, safety and workload. A faster process that transfers risk or burden elsewhere is not an adequate improvement.
Choose the smallest test that can answer the question without unacceptable exposure. Define duration, population, owner, rollback criteria and required approvals for the actual risk. Small does not mean careless: a test affecting customer identity or safety needs appropriate protection even if only a few orders are involved.
03 · Worked example: testing shift-overlap support
Worked example: testing shift-overlap support
Fictional Northbank’s release process is stable after its improvement project, but verification waits rise during shift overlap. The team observes that both shifts rely on one supervisor for reprint approval. It predicts that training a second authorised approver for a defined two-hour overlap will reduce waiting without changing the required checks. The test covers three comparable overlap periods on one carrier route.
The team records the time each reprint enters hold and the time it becomes eligible for release. It also tracks incorrect releases, the number of reprints, approver workload and interruptions to other duties. The second approver receives observed competency signoff before the test. Parcels with uncertain identity remain held; the test does not relax the identity requirement.
Waiting falls in the test periods, but the approver’s ordinary work accumulates. The team therefore does not adopt the arrangement immediately. It learns that the original prediction about waiting was plausible, while the staffing assumption was incomplete. A second cycle tests a clearer reprint decision aid and scheduled support from a role with available capacity. The first cycle has produced useful evidence even without permanent adoption.
04 · Interpret results without forcing success
Interpret results without forcing success
Compare results with the prediction and baseline conditions. Look at counts, variation, case mix and the timing of events. A short test with few cases may reveal feasibility or failure modes without establishing a durable performance gain. Label that level of evidence. If demand was unusually low, do not attribute every improvement to the change.
Northbank’s second cycle reduces waiting without increasing outstanding work or errors. The team examines whether the result persists across both shifts and whether the decision aid is understood without coaching. It retains the change provisionally and expands observation to a busier period. The decision is based on converging evidence rather than a single favourable average.
Ask what was learned when the result differs from the prediction. The change may have been poorly implemented, the causal assumption may be wrong or a balancing effect may dominate. Adapt the test based on that explanation. Avoid changing several elements at once and then claiming to know which caused the result. A failed prediction is valuable when it sharpens the next question.
05 · Adopt through the control system
Adopt through the control system
Once evidence supports adoption, update the standard, training, configuration and control plan together. Specify the implementation boundary and effective date. Remove obsolete instructions and temporary trial arrangements. The process owner accepts the change and assigns monitoring. Improvement is not complete when the trial ends; it is complete when the better method becomes a reliable operating method.
Northbank incorporates the reprint decision aid into its release instruction and trains all relevant approvers. The dashboard retains waiting time and accuracy as balancing measures. The shared board explains the changed route and owner. After sufficient representative operation, the team assesses whether a new reference is justified for its monitoring chart rather than automatically recalculating limits because the intervention occurred.
Store the test question, result and learning in a concise record. Include negative findings and boundaries. Another warehouse may have different staffing and reprint frequency, so it should test the idea in its own conditions. Sharing a successful mechanism is more useful than announcing that every site must copy an identical layout or staffing pattern.
06 · Build a sustainable improvement routine
Build a sustainable improvement routine
Provide time and capability for learning. If suggestions are welcomed but nobody can test them, the process becomes a collection box. Review ideas regularly, prioritise them transparently and respond to contributors. Some ideas can be tested locally; others need cross-functional work or a larger DMAIC project. Explain the route and the reason for prioritisation.
Avoid measuring success only by the number of ideas or events. Those activity counts can encourage trivial changes. Review customer outcomes, reduced burden, reliability and the quality of learning. Recognise evidence-based abandonment of a weak idea as responsible improvement. Respectful experimentation depends on people being able to reveal problems and uncertain results.
Advanced improvement systems connect local cycles with wider goals and recurring risk themes. Maintain enough common definitions to compare learning, while allowing local tests to reflect actual conditions. Common errors include uncontrolled trials, no prediction, no balancing measure, cherry-picked results and adoption without standardisation. A disciplined continuous improvement process makes change safer, more understandable and more likely to persist.
07 · Use stop criteria as learning boundaries
Use stop criteria as learning boundaries
A planned test should specify conditions that require stopping or restoring the previous method. Northbank stops the overlap experiment if identity checks become incomplete, the hold lane loses safe access or other essential work cannot be maintained. These criteria protect the operation and prevent the team from continuing simply because the test calendar has not ended. Record why a test stopped and what was observed before the stop. Do not treat early termination as a reason to discard uncomfortable evidence. The next cycle can then address the demonstrated constraint with a clearer hypothesis and more suitable protection.
Connections
Related methods and phase use
Placed under Control because of the outcome it supports. Use the method in other phases when its purpose and assumptions fit the work; the phase list is a guide to selection, not a compulsory sequence.
Sources
Professional foundations and further reading
The explanation and fictional teaching examples are written for LeanTrainingLab. Linked primary sources provide professional foundations; example values and local conventions are stated in the guide. The accepted toolset draws on Six Sigma+Lean Toolset, second edition (Lunau, editor; Springer, 2013). No book text or examples are reproduced here.