Library.Reference:6.5.3.1 - Process performance, an improvement test and sustained use
In the Model for Improvement, an improvement aim, useful measures and an idea for change guide PDSA. Performers predict the result of a bounded test, conduct it, compare observations with the prediction and select the next move. The test supplies grounds for a decision. Sustained use of the change requires everyday work to be organized. IHI distinguishes testing from implementation, including training, documentation and supporting conditions.
Choose what the next test must establish. An early PDSA can answer a small question about feasibility or the participants’ response. A claim that the change caused an improvement needs observations and a comparison that can distinguish the relevant alternatives; the number of completed cycles does not supply that basis. Implementation establishes continuing use through the actual workflow, responsibilities and support. It can itself require tests. When spreading a change, the receiving participants must establish how it fits and works in their conditions. Knowledge from a trial, sustained practice and the organization’s ability to conduct further improvement are different results.
Connect the change with its implementation. IHI’s Core Components Guide distinguishes a content theory, explaining why selected changes should improve the system, from an execution theory, explaining how the initiative will make their use obtainable. Develop these together with people who perform and receive the work: connect the problem, aim and proposed changes with time, data, learning, cooperation and responsibility. Observe both the system’s results and the conduct of the initiative. Revisit their connections as understanding changes. The components support an iterative design; their numbered presentation does not prescribe a linear sequence.
Return evidence to the contribution it can change. A.3.1.MI:4.5 supplies the fuller comparison when the next action depends on distinguishing delivery, mechanism and context. If the intended action did not occur, inspect access, allocation, capability and support before rejecting its proposed mechanism. If it occurred but the expected consequence did not, examine that mechanism, the relevant conditions and competing explanations. A useful result obtained through additional help can justify retaining that help while leaving the original explanation unresolved. Distinguish improvement in the service, the intended performer’s acquired ability and the organization’s ability to obtain the work. Meetings, coaching hours and completed tests can describe activity; later performance under the required conditions is needed for the corresponding ability claim.
Use IS.3 when the missing contribution is how the practice becomes usable. Use a result already in hand when it answers the present need under current conditions and the recipient can act on it. If a fresh result is needed, check whether an existing service can supply it in time. If access or another prerequisite for use is missing, check whether restoring it would suffice. An instruction, record or observation may already show how the required contribution reaches its user. If a consequential gap remains, obtain the missing evidence. Work through a representative case together when the gap calls for joint interpretation, design or agreement. When the aim is for the recipient to learn the operation, arrange an attempt and feedback on the recipient’s work. A different introduction can preserve the target way. A change to its decisive operation or criterion returns to that way’s owner. OPS.22 supplies interpretation of actual operating results; OCE.13 compares consequences when the organization arrangement changes, including work transferred to other participants. Use their sufficient existing results directly.
Connect continuing improvement to organizational decisions. Quality as an Organizational Strategy connects the need served and the organization’s purpose with its interdependent processes, measures, information and improvement work. Information reaches distinct uses: an immediate response to a problem, feedback to an existing improvement effort, or analysis for future planning. Planning connects strategic objectives to the processes that need change and produces both operating plans and a resourced improvement portfolio. Leaders sponsor the work, help remove obstacles and use its results to learn about the organization. Existing improvement projects can provide the starting point while the other connections develop. A process map makes dependencies discussable; its arrangement on the page does not establish decision rights.
STR.12 connects a qualified signal to the premise and authorized response it can change. Use STR.16 when several worthwhile undertakings and their support still need a feasible combination. Separate one undertaking’s need for another’s result from competition for the same person or resource. Obtain the required contributions and commitments, then observe whether they let the receiving work proceed. Experience can warrant revising an initiative, its support, the portfolio or the strategic direction. Each revision needs the people authorized to decide it and those who can perform the change. These relationships can be distributed among participants or supplied by the same people.
A local test with adequate provision can finish its bounded question without constructing a new organization-wide programme. Retain a sufficient current method and arrangement. Expand the inquiry or arrangement when the needed evidence, continuing work, shared capacity or decision right is missing or changes. Close the present question at its supported scope while preserving the later observation or dependency that can reopen it.
Recover the selected formulation of PDCA or PDSA. The Deming Institute’s PDSA account emphasizes study of a prediction and permits reconsidering theory and goals. ASQ’s PDCA also includes analysis and learning in Check. Treating every PDCA as compliance checking would lose that variant. An experiment can inform a goal change; the actual organization must still provide the decision right.
Imai’s SDCA/PDCA distinction makes the relationship between maintaining and changing a standard explicit. Performers sustain work under the current standard and discover departures; improvement supplies grounds to revise that standard. Further work uses the revision, and its results inform the next reconsideration. One process owner can participate in both relationships. The basis is Gemba Kaizen, second edition, chapter 1, and the Kaizen Institute’s account of standardization and improvement. Maintaining an adequate way need not carry the burden of another improvement investigation.
DMAIC adds useful measurement and investigation of causes before changing the process. Its Control phase organizes sustained results and response to later departures. ASQ distinguishes DMADV where designing a new process is required. A3, particularly pages 3–9 of the LEI account, connects problem investigation with participants’ examination of the reasoning, agreement and responsibility for action. The sheet supports those interactions. These variants explain how a PDSA test can contribute inside a more developed undertaking.
Distinguish the difficulty before choosing the response. Begin with the result, population, period and requirement that matter to the recipient. For a defective item, apply the relevant acceptance or protective rule. A statistical signal concerns departure from an established process behavior. An unreliable measurement leaves the attributed property uncertain. A missed operation raises a question about actual execution and its conditions. These difficulties can coexist. OPS.18 connects each evidence question to its available action and authority. Act on a known requirement violation when its response is already warranted, while obtaining any diagnosis still needed for prevention. A stable process can repeatedly produce unacceptable output; statistical stability alone does not settle acceptance or process capability.
Establish what the observations represent. Recover the unit or event, inclusion rule, time order, measuring procedure and relevant operating conditions. OPS.15 supplies the population and event account; C.16:5.1–5.4 connects an indication to the intended property, calibration and uncertainty. Repeated readings of one unchanged item can expose measurement variation; they do not sample the variation among produced items. Before adjusting the process, test whether a change of gauge, operator or recording rule contributed to the apparent change. If the rule changed, apply both rules to the same retained observations where possible. If a new gauge gives a different result on a later batch, use an established comparison of the gauges under the relevant conditions, or compare them on the same stable specimen when needed. Both the measurement and the process may have changed. NIST’s Gauge R&R guidance identifies repeatability, reproducibility, stability and bias as distinct measurement questions. More readings under the same shared bias leave that bias unresolved.
For monitoring, establish the baseline and a suitable observation and response rule, then compare later behavior with that basis. NIST distinguishes baseline establishment from subsequent monitoring. Dependence, changing case mix or grouping can change the interpretation and the monitor needed. A new baseline requires grounds for the changed reference; an unfavorable observation alone is no reason to erase the departure. Use the existing qualified monitor or acceptance plan when it already answers the question.
Connect the investigation to the explanation or change required. DMAIC connects a defined problem to usable measurements, an account of causes, a proposed remedy and continuing control. Return between these contributions when a finding changes the question: a recording error can invalidate the apparent defect pattern; a failed remedy can challenge the proposed mechanism. C.28.CM supplies the comparison of material causal explanations, including measurement and selection. Choose another observation for what it could reveal about the mechanism or the consequences of a remedy. If the retained explanations already support the same sufficient protection, use it with its stated limits. A commission to establish the cause of the defect still calls for an account of how it arose. A commission to prevent recurrence calls for a way to keep the process effective under the relevant conditions. Either may require distinguishing mechanisms that support the same immediate protection.
Compare interventions through their consequences. Finding that an input affects an output does not yet establish how much changing that input would improve ordinary production. For an output Y that depends linearly on input X, write Y = a + bX + e: a is an offset, b the sensitivity and e the remaining influence. With independent X and e and finite variances, Var(Y) = b² Var(X) + Var(e). A suitable experiment can estimate the effect b, while the ordinary distribution of X needs its own basis; experimenter-selected levels need not reproduce it. This is a useful distinction in Panahi, Steiner and de Mast’s verification study, §§2–3.1. Compare feasible changes to the input distribution, sensitivity, centering or process design against the required output and their full consequences. A statistically significant effect or a large explained share alone does not select the remedy.
Use C.28.MR when the consequence still needs to be derived from a model. An applicable estimate or earlier derivation can already answer a request to compare the interventions. For a new derivation, identify the mechanism replaced and the other mechanisms and input conditions that still apply. Follow those relations to the result the question needs. They may establish a direction of change, a bound or a quantitative effect. A common cause of X and Y can leave output variation after X is fixed; their squared correlation alone does not quantify the reduction from that intervention. Carry model uncertainty and uncertainty in estimated quantities into the comparison when they can change the decision. MMP.13 supplies inference under the actual observation law. Process requirements can concern tails, centering or asymmetric consequences; a smaller variance is useful only through the result it improves.
Select observations for the distinction needed. A study aimed at separating causes can deliberately choose contrasting cases. The 2025 review of leveraged sample selection explains how parts near opposite ends of a known baseline can help investigate variation. That choice needs a justified baseline, an appropriate study and analysis that account for selection. Very extreme cases may belong to another failure mechanism. Such an investigation answers a different question from estimating the ordinary frequency of defects. MMP.16 connects the alternatives, obtainable records and receiving decision. Use existing information when it is sufficient; obtain a new study when its attainable contribution warrants its cost and delay.
A conditional calculation can reject an insufficient remedy. Suppose a centered, stable process is modeled by Y = 10 + 0.5X + e, with independent normal X and e, means zero, Var(X) = 4 and Var(e) = 1. The measurement basis and execution are assumed adequate. Individual output must lie between 8 and 12, and the process objective is at most 2% outside that interval. Under these assumptions:
| Compared condition | Output variance | Modeled fraction outside 8–12 |
|---|---|---|
| Current process | 2 | 15.73% |
| Reduce Var(X) from 4 to 1; retain sensitivity 0.5 | 1.25 | 7.36% |
| Reduce sensitivity from 0.5 to 0.25; retain Var(X) = 4 | 1.25 | 7.36% |
| Hold X at zero; retain the residual mechanism | 1 | 4.55% |
The alternatives leave the stated centering and residual distribution unchanged. Even ideal control of X misses the 2% objective. To meet the 2% objective, obtain another effective change, such as reducing residual variation. If the requirement itself is reconsidered, the authorized decision changes the target of the comparison. The calculation can end the search for a sufficient X-only remedy while containment continues. It does not choose among the remaining changes or establish that the assumed model describes an actual process. A measured improvement would still need its own observations and comparison.
Return the chosen change to sustainable work. DMAIC Control and SDCA connect the selected way to actual performance, observation and a response to departure. Make the operating method and required inputs obtainable, assign capable performers, and provide the time, means and rights needed to act. State who can contain output, restore operation, authorize restart or revise the standard. Observe whether the changed practice continues and whether the required result follows. OPS.16 compares a reusable method change with alternatives; OPS.18 supplies containment and restart. Use the implementation connection above when the missing contribution is how the practice becomes usable.
When a departure from a standard leaves the next response unresolved, distinguish the missing contribution. Restore missing tools, access or time when these prevented performance. If the difficulty concerns an operation the person has not yet mastered, arrange an attempt, usable instruction, correction and follow-up. The TWI Institute’s account of Job Instruction connects preparation and demonstration with trying the job and subsequent help; HCD.9 supplies the feedback and retry under stated support conditions. When adequate performance under the intended conditions still fails the requirement, reconsider the way and its causal grounds. Statistical process capability and a person’s ability to perform the work are different claims. Retain a useful reference, colleague or other support when it belongs to the intended work; successful assisted performance need not imply independent ability.
Maintaining a sufficient standard and its provision is an ordinary completion of this control question. A local correction can finish without launching a new DMAIC project. Reopen the affected relation when measurement, process behavior, requirements, execution or provision changes enough to alter the warranted action. A missing decision right or shared resource returns to the organizational connections above; another repetition of the local cycle cannot supply it.