Lean & Six Sigma Systems

Chapter 11
DMAIC: Control and Benefit Sustainment

Post Graduate Program in Total Quality Management
Lean & Six Sigma Systems • Post Graduate

11. DMAIC: Control and Benefit Sustainment

Introduction and Context

The Control phase is the point at which a Six Sigma project becomes an operating discipline rather than a temporary improvement exercise. Earlier DMAIC phases establish the problem, identify causes, test solutions and demonstrate improvement. Control ensures that the improved process continues to perform when the project team withdraws, operating conditions change, personnel move, demand fluctuates or management attention shifts elsewhere.

Control is therefore more than the final step of DMAIC. It is a system of ownership, measurement, response, standardization, governance and learning. A process may achieve an impressive short-term reduction in defects and still fail to deliver lasting value if its new method is not embedded in daily work. Conversely, a well-designed control system makes performance visible, signals deterioration early, guides corrective action and verifies that expected operational and financial benefits have actually been realized.

The phase connects statistical thinking with management practice. Statistical process control (SPC) helps distinguish common-cause variation from special-cause variation. Control plans translate improvement knowledge into routine operating requirements. Standard work makes the improved method repeatable. Audits test whether the system is being followed. Dashboards connect process measures with customer, financial and strategic outcomes. Process ownership ensures that accountability remains after formal project closure.

Learning Objectives

After studying this chapter, the learner should be able to:

1. Explain the purpose and logic of the Control phase in DMAIC.

2. Design a control plan linking critical process inputs, process measures and outputs.

3. Select and interpret suitable control charts for continuous and attribute data.

4. Distinguish common-cause from special-cause variation and develop response plans.

5. Construct useful operational dashboards and audit systems.

6. Assess post-improvement process capability and interpret its limitations.

7. Standardize work, transfer ownership and manage the transition from project to process.

8. Verify financial, operational and customer benefits without overstating project results.

9. Evaluate risks, governance requirements and digital opportunities in sustainment.

The Logic of Sustaining Improvement

An improvement is sustained when the process continues to produce the intended level of performance under ordinary operating conditions. Sustainability has three dimensions. First, the process must remain statistically stable. Second, it must remain capable of meeting customer and business requirements. Third, the organization must preserve the behaviors, controls and ownership that produced the result.

These dimensions should not be confused. A stable process is not necessarily capable: it may consistently produce output outside specification. A capable process may not be stable: its average and variation may change unpredictably. Both statistical performance and organizational discipline are required.

The central question in Control is not simply, “Did the process improve?” It is, “What must be monitored, by whom, how often, using which decision rules, and with what response when performance changes?” A control system that measures everything may create administrative burden without improving control. A system that measures too little may discover failure only after customers or financial results have been affected.

A useful sustainment logic is:

Critical requirement → process risk → control method → responsible owner → response → verification.

For example, if a customer requires delivery within 48 hours, the organization may identify order-release delays as a major risk. The control method may include a daily cycle-time chart and an escalation rule for two consecutive breaches. The process owner investigates the cause, records action and verifies that the process returns to its expected state.

Designing the Control Plan

A control plan is a structured description of how an improved process will be maintained. It should be derived from the findings of the Measure, Analyze and Improve phases rather than created as a generic checklist. The plan normally includes the process step, critical input or output, specification or target, measurement method, sampling frequency, control chart or other monitoring method, responsible person, reaction plan and records required.

The plan should distinguish between three kinds of measures:

Input measures: conditions entering the process, such as material quality, staffing level or data completeness.
Process measures: variables describing how the work is performed, such as queue time, temperature, setup duration or first-pass yield.
Output measures: results experienced by the customer or business, such as defects, lead time, complaints, cost or on-time delivery.

Output measures are essential, but they may be too late for effective intervention. A mature control plan therefore combines lagging measures, which confirm results, with leading measures, which indicate whether the process is being operated correctly.

A control plan should be risk-based. High-risk characteristics deserve more frequent measurement, stronger error-proofing, clearer escalation and greater management visibility. Low-risk characteristics may be controlled through periodic verification. Excessive controls can cause measurement fatigue and encourage employees to treat the system as paperwork rather than as operational support.

The measurement system must also be adequate. If a gauge, inspection method or information system is unreliable, the control chart may display measurement noise rather than process behavior. Where relevant, measurement-system analysis should be repeated after process changes, especially when new devices, software or operators are introduced.

Control-plan example

Consider a service process for approving commercial loans. The improvement project reduces approval lead time by removing duplicate reviews. The control plan might include:

|---|---|---|---|---|

The plan integrates speed, quality and compliance. Focusing only on approval time could encourage inappropriate approvals or incomplete documentation.

Statistical Process Control in the Control Phase

SPC is a method for understanding process variation over time and using that understanding to guide action. A control chart contains a central line representing the process average or expected level and upper and lower control limits representing the range of variation expected when only common causes are present. Control limits are calculated from process data; they are not the same as customer or engineering specification limits.

A process is generally interpreted in two stages. In the first stage, historical or baseline data are examined to estimate the process’s natural behavior and identify special causes. In the second stage, future observations are compared with the established limits to determine whether the process remains stable. Limits should not be recalculated merely to hide deterioration. They may be revised when the process has undergone a verified, fundamental change and sufficient new data have been collected.

Choosing the chart

The correct chart depends on the type of data and the sampling structure.

Individuals and moving range (I-MR) charts: one observation at a time, often used for cycle time, response time or chemical concentration when subgrouping is impractical.
X-bar and R charts: continuous measurements collected in rational subgroups, with the average and range monitored separately.
X-bar and S charts: continuous subgroup data, particularly when subgroup sizes are larger.
p charts: fraction or proportion defective when sample size can vary.
np charts: number defective when sample size is constant.
c charts: number of defects per constant inspection unit.
u charts: defects per unit when the opportunity or sample size varies.
Rare-event or time-between-event charts: useful where defects are infrequent and ordinary attribute charts may be insensitive.

A rational subgroup should contain observations produced under sufficiently similar conditions so that within-subgroup variation represents short-term noise. Poor subgrouping can conceal meaningful changes or create false signals.

Interpreting signals

A point outside a control limit is a strong signal of a special cause, but it is not the only signal. Long runs on one side of the center line, sustained trends, cycles or unusual clustering may also indicate a change. Rules should be selected carefully and documented before use. Applying too many sensitive rules increases false alarms; applying too few delays intervention.

When a signal occurs, the first response is not automatically to adjust the process. The organization should confirm the data, protect the customer if necessary, identify what changed, determine whether the cause is special or common, and document the decision. Frequent tampering—making adjustments in response to ordinary random variation—can increase instability. Deming’s funnel experiment illustrates how well-intentioned adjustment to every individual observation can make a process worse.

Control limits and specification limits

Control limits describe the voice of the process; specification limits describe the voice of the customer, regulator or designer. A process can be statistically stable but fail specifications. It can also meet specifications temporarily while being statistically unstable. Confusing these concepts is one of the most serious errors in capability analysis.

Process Capability After Improvement

Capability analysis asks whether a stable process can meet defined requirements. For approximately normal continuous data, common indices include:

Cp = (USL − LSL) / 6σ

Cpk = minimum[(USL − μ) / 3σ, (μ − LSL) / 3σ]

Cp reflects potential capability if the process is centered. Cpk reflects actual capability, including displacement of the mean from the target. A large difference between Cp and Cpk indicates that centering may be an important opportunity.

For a one-sided requirement, indices such as Cpu or Cpl may be appropriate. Attribute data require different approaches, often involving defect rates, yield, DPMO or confidence intervals. Capability indices should not be calculated mechanically when the data are autocorrelated, heavily skewed, mixed across distinct populations or collected from an unstable process. Transformation, distribution-free methods or stratified analysis may be needed.

Capability after improvement should be compared with the baseline using consistent definitions, measurement methods and operating conditions. The comparison should consider more than an index. Management should examine customer complaints, rework, throughput, capacity, safety, compliance and cost. A statistically impressive improvement that creates overtime or transfers defects downstream is not a complete success.

Response Plans and Escalation

A response plan converts a signal into disciplined action. It should specify what constitutes an abnormal condition, who is authorized to respond, what immediate containment is required, how the cause will be investigated and when the issue must be escalated.

A practical response sequence is:

1. Detect: identify the signal through a chart, audit, alarm or dashboard.

2. Validate: confirm the measurement, data integrity and operating context.

3. Contain: protect customers, employees and downstream operations.

4. Diagnose: investigate the assignable cause using evidence rather than assumption.

5. Correct: remove the cause or restore the standard condition.

6. Verify: confirm that performance has returned to a stable and acceptable state.

7. Learn: update the standard, control plan, training or risk analysis if necessary.

Escalation thresholds should be proportionate to risk. A critical safety or regulatory deviation may require immediate stoppage and formal reporting. A minor fluctuation may require local review at the next shift meeting. Escalation should never be designed to punish people for reporting problems; otherwise the system will generate hidden failure rather than reliable information.

Dashboards, Audits and Visual Management

A dashboard is useful when it supports a decision. It should not be a collection of attractive graphs without defined users, targets and actions. A robust dashboard links different levels of performance:

strategic outcomes, such as margin, customer retention or regulatory performance;
customer outcomes, such as complaints, service level and defect experience;
process outcomes, such as cycle time, yield and first-pass accuracy;
leading indicators, such as adherence to standard work, preventive maintenance completion or queue health.

The dashboard should display current status, trend, target, owner and required action. Definitions must be stable. For example, “on-time delivery” should specify the promised date, exclusions, time zone and treatment of partial shipments. Otherwise, apparent improvement may result from changing the metric rather than improving performance.

Audits provide a different form of control. Statistical monitoring asks whether process results are behaving as expected; audits ask whether the management system and standard work are being followed. Layered process audits may involve operators, supervisors, managers and senior leaders at different frequencies. Effective audits focus on critical controls, observe the work directly, record evidence and close actions. They should not become ceremonial inspections or occasions for blame.

Standard Work and Transfer of Ownership

Standard work is the documented best-known method for performing a task safely, consistently and efficiently under defined conditions. It may include sequence, takt or expected time, work-in-process limits, quality checks, visual aids and abnormal-condition instructions. Standard work is not intended to prevent improvement. It establishes a stable baseline from which further improvement can be made.

Ownership transfer should occur before formal project closure. The process owner must understand the improved process, control plan, measures, response rules, risks, open actions and expected benefits. Training should be demonstrated through observed performance, not merely attendance. Documents, software settings, forms and maintenance responsibilities must be updated in the systems where work is actually performed.

A useful tollgate for closure asks:

Is the process stable and capable under normal conditions?
Are controls embedded in daily management?
Is the owner formally identified and trained?
Are standard operating procedures and visual controls current?
Are unresolved risks assigned with due dates?
Are benefits measured using an agreed baseline and finance-approved method?
Is there a scheduled post-closure review?

Verifying Benefits and Financial Realization

Benefit verification separates operational change from financial realization. A reduction in defects may create theoretical savings, but those savings become financial benefits only when the organization avoids cost, releases capacity, increases contribution, reduces working capital or otherwise improves the economic position.

Benefits should be classified clearly. Hard savings may include eliminated external charges, reduced material consumption or cancelled temporary labor. Cost avoidance prevents a forecast increase but may not reduce current expenditure. Capacity release creates available time or resources, but its value depends on whether that capacity is used productively. Revenue or margin impact requires evidence that improved performance changes sales, retention or contribution.

Benefit validation should define the baseline period, calculation logic, volume assumptions, exclusions, owner and reporting frequency. Finance or an equivalent independent function should review major claims. Double counting must be avoided when several projects affect the same cost or performance measure.

A simple financial model might be:

Net annual benefit = avoided cost + realized productivity value + incremental contribution − recurring control cost − implementation cost.

The model should be tested against actual results after implementation. If defect reduction increases inspection cost, or if cycle-time improvement requires additional software licenses, those costs belong in the net benefit calculation. Benefits should also be disaggregated by site, product, customer segment or shift when performance is uneven.

Illustrative Case: Sustaining Improvement in a Medical-Device Packaging Process

Background

The following is an illustrative case. A medical-device manufacturer experienced a high rate of packaging rejects caused by seal defects. The defects led to rework, delayed release and occasional destruction of finished units. A DMAIC team identified variation in sealing temperature, inconsistent setup practices and delayed detection as major contributors.

Situation and analysis

After improvement, the process team introduced a validated temperature window, a setup checklist, preventive-maintenance triggers and an in-line seal-strength test. The baseline reject rate had been measured over several months, while post-improvement results were collected across different operators and shifts. Initial results showed a substantial reduction in rejects, but the team noticed that the night shift had greater variation.

The team constructed an I-MR chart for seal strength because production volume and sampling conditions made rational subgroups unsuitable. The chart showed a special-cause signal coinciding with a heater replacement. Investigation found that the replacement component had a different calibration requirement. The process was contained, affected lots were evaluated under the quality system, and the calibration procedure was revised.

Control strategy

The final control plan included automatic temperature recording, a startup verification, periodic seal-strength sampling, a daily review of chart signals and a weekly audit of setup adherence. The response plan required line stoppage for a critical seal failure, segregation of potentially affected product and notification of quality engineering. Operators were trained to respond to alarms rather than bypass them.

A dashboard displayed reject rate, first-pass yield, chart signals, preventive-maintenance completion and audit findings. The production supervisor owned daily control; quality engineering owned capability review; the maintenance manager owned equipment reliability.

Outcome and benefit verification

After three months, the process remained stable and capability relative to the seal-strength requirement improved. The organization did not claim the entire reduction in rework cost as immediate cash savings. Finance verified the portion associated with reduced scrap purchases and external testing. Released operator capacity was reported separately because it was used partly to absorb increased demand rather than to reduce headcount.

Learning points

The case demonstrates that control requires more than a lower defect rate. The process needed validated measurement, special-cause investigation, cross-shift ownership, documented escalation and cautious financial interpretation. It also illustrates why equipment changes must be treated as potential sources of variation and why quality controls in regulated industries must be integrated with formal change control and traceability.

Challenges, Risks and Common Mistakes

Several failure modes repeatedly weaken sustainment. The first is premature closure: a project is declared complete after a short period of favorable results. The second is selecting measures that are easy to collect but weakly related to customer or business performance. The third is using control charts without checking measurement quality, distribution, independence or subgrouping.

Other risks include recalculating limits after every change, confusing specifications with control limits, reacting to every random fluctuation, and allowing multiple owners to share accountability without assigning clear responsibility. A further danger is metric conflict. For example, a call center may reduce average handling time while increasing repeat calls and customer dissatisfaction.

Human and organizational factors are equally important. If standard work is unrealistic, employees will create informal workarounds. If audits are punitive, abnormalities will be hidden. If leaders reward output without recognizing quality and safety, the control system will be undermined by its own incentives.

Governance, Ethics and Technology

Control systems must respect confidentiality, safety, regulatory obligations and the rights of employees and customers. Data used for dashboards should be accurate, access-controlled and traceable. Automated decisions require appropriate validation, especially in healthcare, finance, aviation and other high-consequence settings. Performance data should not be used to discipline individuals when the underlying process is unstable or when measurement conditions are inequitable.

Digital technologies can strengthen sustainment through real-time data capture, automated control charts, statistical alerts, electronic standard work, sensor-based condition monitoring and process mining. Machine-learning models may detect complex patterns, but they should complement—not replace—process knowledge and responsible human judgment. False alarms, model drift, poor data quality and opaque recommendations can create new risks. Any digital control should have a defined owner, validation method, fallback procedure and cybersecurity protection.

Current practice is moving toward integrated operational-management systems in which SPC, digital work instructions, maintenance data, quality records and financial measures are connected. The value of integration lies not in technology alone but in reducing the delay between abnormality, decision and verified response. Organizations are also giving greater attention to resilience: control plans increasingly consider supply disruption, demand volatility, cyber incidents, climate-related events and workforce capability rather than assuming that historical conditions will continue indefinitely.

Best-Practice Principles

Effective sustainment generally follows several principles:

1. Design controls from verified process risks and critical requirements.

2. Measure both results and leading conditions that influence results.

3. Use the simplest valid chart and the clearest useful dashboard.

4. Separate common-cause improvement from special-cause correction.

5. Define reaction plans before signals occur.

6. Make ownership visible at the point of work.

7. Treat standard work as a living baseline, not a static document.

8. Validate capability only after establishing stability.

9. Verify benefits with transparent assumptions and independent financial review.

10. Review controls after major changes, incidents, technology updates or shifts in demand.

Chapter Summary

The Control phase converts DMAIC improvement into sustained organizational performance. Its essential elements are a risk-based control plan, appropriate SPC, disciplined response plans, meaningful dashboards, layered audits, standard work and explicit process ownership. Stability and capability are different judgments: a process must first be shown to be statistically stable and then evaluated against customer or specification requirements. Capability indices are useful only when the data and assumptions support them.

Sustainment also requires benefit verification. Operational improvements should be translated into financial claims cautiously, distinguishing realized savings, cost avoidance, capacity release and revenue effects. Governance, ethics, data quality and regulatory requirements must be integrated into the control system. Digital tools can improve visibility and speed, but they require validation, ownership and safeguards. Ultimately, control is successful when improved performance becomes the normal way the organization operates and learns.

Key Terms

Control plan: A documented method for maintaining process performance through measures, responsibilities and reactions.

Statistical process control (SPC): The use of statistical methods to monitor and manage process variation over time.

Control limit: A statistically derived boundary indicating expected process variation under stated conditions.

Specification limit: A customer, regulatory or design requirement defining acceptable output.

Common-cause variation: Inherent variation produced by the overall process system.

Special-cause variation: Variation attributable to a specific, identifiable and non-routine cause.

Process capability: The ability of a stable process to meet specified requirements.

Cp: An index of potential capability based on specification width and process spread.

Cpk: An index reflecting both process spread and centering relative to specifications.

Rational subgroup: A group of observations collected under sufficiently similar conditions for meaningful SPC analysis.

Response plan: Predetermined actions to follow when a process signal or abnormality occurs.

Standard work: The documented best-known method for performing work under defined conditions.

Process owner: The person accountable for ongoing performance and control of a process.

Layered process audit: Auditing performed at several organizational levels and frequencies.

Leading indicator: A measure that signals conditions likely to influence future performance.

Lagging indicator: A measure of results achieved after process activity has occurred.

Benefit realization: Confirmation that expected operational or financial improvements have occurred in practice.

Capacity release: Productive time or resource availability created by improvement.

Measurement-system adequacy: The degree to which a measurement method produces reliable information for decisions.

Control-system tampering: Unnecessary adjustment of a stable process in response to ordinary random variation.

Review Questions

Short Answer Questions

1. Why is a stable process not necessarily a capable process?

2. Distinguish control limits from specification limits.

3. What is the purpose of a response plan?

4. When is an I-MR chart appropriate?

5. Why should capacity release be reported separately from hard savings?

Long Answer Questions

1. Explain how a risk-based control plan should be designed and implemented.

2. Discuss the selection and interpretation of control charts for continuous and attribute data.

3. Explain the relationship between standard work, audits and process ownership in sustainment.

4. Describe how capability should be assessed after an improvement project.

5. Discuss the governance, ethical and technological issues involved in digital control systems.

Analytical / Application Questions

1. A process has Cpk below the required level but Cp substantially higher than Cpk. What does this suggest, and what actions should the process owner consider?

2. A control chart shows one point beyond the upper limit, followed by normal observations. Develop a disciplined response that avoids both neglect and unnecessary tampering.

3. A project reports large annual savings from reduced defects, but production volume has not changed and no budget has been reduced. How should the claimed benefit be examined and reported?

22. Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes a control limit from a specification limit?

A. A control limit is set by the customer, while a specification limit is calculated from process data.

B. A control limit describes expected process variation, while a specification limit describes acceptable requirements.

C. Both limits are calculated in exactly the same way from historical averages.

D. A specification limit is used only for attribute data, while a control limit is used only for continuous data.

Correct Answer: B. A control limit describes expected process variation, while a specification limit describes acceptable requirements.

Question 2

A process owner observes a control-chart point beyond the upper control limit. What should normally happen first?

A. Recalculate the control limits immediately.

B. Adjust the process setting without investigating the cause.

C. Validate the data and operating context, then contain risk if necessary.

D. Ignore the point because one unusual observation cannot matter.

Correct Answer: C. Validate the data and operating context, then contain risk if necessary.

Question 3

Which chart is generally suitable for monitoring the proportion defective when sample size varies?

A. p chart

B. np chart

C. c chart

D. X-bar chart

Correct Answer: A. p chart

Question 4

A process has a high Cp but a considerably lower Cpk. The most likely interpretation is that the process is:

A. Highly unstable with no measurable variation.

B. Potentially capable in spread but poorly centered relative to specifications.

C. Meeting specifications because Cp always reflects actual performance.

D. Unsuitable for capability analysis because Cpk is never useful.

Correct Answer: B. Potentially capable in spread but poorly centered relative to specifications.

Question 5

Which measure is most clearly a leading indicator of sustainment?

A. Annual profit after project closure

B. Customer complaints from the previous quarter

C. Adherence to critical standard-work checks

D. Total defects recorded at year-end

Correct Answer: C. Adherence to critical standard-work checks

Question 6

Why can adjusting a stable process after every individual observation be harmful?

A. It eliminates all special causes automatically.

B. It may respond to common-cause variation and increase process instability.

C. It makes specification limits wider.

D. It prevents the use of any dashboard.

Correct Answer: B. It may respond to common-cause variation and increase process instability.

Question 7

Which item is essential in a response plan?

A. A rule requiring every abnormality to be blamed on an operator

B. A predetermined definition of the signal and the actions to follow it

C. A requirement that all control limits be changed monthly

D. A financial forecast unrelated to process risk

Correct Answer: B. A predetermined definition of the signal and the actions to follow it

Question 8

Why should capability analysis generally follow confirmation of process stability?

A. Capability indices are meaningful only when variation is reasonably predictable.

B. Stable processes do not need specification limits.

C. Capability analysis removes the need for measurement-system analysis.

D. Unstable processes always have higher capability indices.

Correct Answer: A. Capability indices are meaningful only when variation is reasonably predictable.

Question 9

Which claim is most appropriately classified as capacity release rather than hard savings?

A. A cancelled external testing invoice

B. A reduction in purchased material consumption

C. Operator time made available and used to absorb additional demand

D. A direct reduction in monthly utility expenditure

Correct Answer: C. Operator time made available and used to absorb additional demand

Question 10

What is the primary purpose of transferring ownership before project closure?

A. To remove the need for future measurement

B. To ensure the process owner can maintain controls and respond to deterioration

C. To allow the project team to change the baseline retrospectively

D. To replace standard work with informal experience

Correct Answer: B. To ensure the process owner can maintain controls and respond to deterioration

Lean & Six Sigma Systems
Chapter 11 of 15

Chapter 11 DMAIC: Control and Benefit Sustainment