DMAIC: Control and Sustained Performance

Introduction

The Control phase is the final phase of DMAIC, but it is not merely a closing activity. It is the management system that preserves the gains achieved during Define, Measure, Analyse and Improve. A process improvement is successful only when the improved method becomes the normal method, performance remains predictable, ownership is transferred to the process owner, and deterioration produces a timely response.

Control combines statistical process control (SPC), documented standard work, visual management, auditing, capability monitoring, accountability and benefit realisation. Its central question is:

How will the organisation know that the improvement is still working, and what will it do when performance begins to deteriorate?

A technically excellent improvement can fail if employees revert to former practices, measurement systems change, control limits are misused, responsibilities are unclear, or management stops reviewing results. Effective control therefore links operational discipline with organisational learning.

Learning Objectives

By the end of this chapter, the learner should be able to:

1. Design a practical control plan for an improved process.
2. Distinguish specification limits from control limits.
3. Select and interpret appropriate control charts.
4. Develop standard operating procedures, ownership arrangements and escalation plans.
5. Use audits, dashboards and benefit-realisation reviews to sustain performance.
6. Recognise common causes, special causes and inappropriate reactions to variation.
7. Evaluate the limitations and governance requirements of control systems.

1. The Purpose and Logic of the Control Phase

The Improve phase identifies and implements changes that address verified causes of defects, delay, cost or variation. Control ensures that these changes are embedded. The phase normally includes:

• confirming that the improved process meets customer and business requirements;
• identifying critical-to-quality characteristics and process variables;
• establishing monitoring methods and sampling frequency;
• setting control limits and, where relevant, specification limits;
• documenting standard work and reaction procedures;
• assigning process ownership;
• training affected personnel;
• auditing adherence and effectiveness;
• transferring responsibility from the improvement team to line management; and
• reviewing whether benefits have actually been realised.

Control has two related dimensions. Process control prevents or detects deterioration during operation. Management control ensures that responsibilities, resources, reviews and decisions remain in place over time. SPC without management action becomes a charting exercise; management review without reliable process data becomes opinion-based control.

1.1 Control versus correction

Control is not the same as inspection. Inspection identifies nonconforming output after it has been produced. Control seeks to understand process behaviour and prevent nonconformity. Correction deals with an observed problem; corrective action addresses the cause of that problem. A reaction plan should therefore specify immediate containment as well as investigation and permanent corrective action where necessary.

1.2 Common-cause and special-cause variation

A stable process is affected by common-cause variation, the cumulative effect of many ordinary factors built into the system. Examples include normal machine-to-machine differences, routine environmental variation and ordinary operator variation. A special cause is an unusual, identifiable influence, such as a broken tool, incorrect material, failed sensor or untrained substitute operator.

A process may be stable but incapable of meeting specifications, or capable on average but unstable. Stability should normally be addressed before capability is interpreted. Adjusting a process after every ordinary fluctuation can increase variation rather than reduce it; this is known as tampering.

2. The Control Plan

A control plan translates the improved process design into operational instructions. It should be specific enough for an operator and process owner to understand what is controlled, how it is measured, how often it is checked, what constitutes an abnormal condition and who must respond.

A useful control-plan structure is shown below.

The control plan should include prevention controls where possible. For example, a mistake-proofing device may prevent incorrect assembly, whereas final inspection merely detects it. Both may be required, but prevention is generally more robust than detection.

2.1 Selecting what to control

Not every variable should be monitored continuously. Selection should consider:

• the relationship between the variable and the critical output;
• customer, legal and safety consequences;
• process sensitivity and speed of change;
• cost and reliability of measurement;
• historical failure modes;
• detectability before value is added; and
• the ability of the organisation to respond.

Controls should be connected to verified causes from the Analyse phase. Monitoring irrelevant variables creates administrative burden and can obscure genuinely important signals.

2.2 Measurement-system readiness

A control chart is only as credible as the measurement system supporting it. The organisation should confirm operational definitions, instrument calibration, resolution, repeatability, reproducibility and data integrity. If measurement error is large relative to process variation, apparent signals may be measurement noise. Measurement-system analysis, including gauge repeatability and reproducibility studies where appropriate, should therefore precede reliance on detailed control limits.

3. Statistical Process Control and Control Charts

SPC uses time-ordered data to distinguish routine process variation from evidence of an unusual change. A control chart normally contains:

• a centre line, often the process mean or average;
• an upper control limit (UCL);
• a lower control limit (LCL); and
• observations plotted in the order in which they occur.

Control limits are calculated from process behaviour. They are not customer requirements. Specification limits are externally or internally defined acceptance boundaries, such as a maximum defect rate or a dimensional tolerance.

3.1 The three-sigma principle

For a stable, approximately normal process, observations have a high probability of falling within approximately three standard deviations of the process mean. The general form is:

Here, μ is the process mean and σ is the standard deviation of individual observations. The units of all three quantities are the same as the measured characteristic. These limits are appropriate only when the chart type, subgrouping and estimation method support the assumptions. For attribute charts and range-based variable charts, limits use distributions and estimators appropriate to those charts rather than simply applying this formula to raw observations.

For subgroup means, the standard error is smaller than individual-observation variation. If subgroup size is n and σ is known or reliably estimated, the approximate limits for means are:

where x̄ is a subgroup mean. The formula assumes independent observations within the relevant structure and a suitable estimate of σ. In practice, constants or within-subgroup estimates are commonly used for X̄–R and X̄–S charts.

Limits should be truncated at a meaningful physical boundary when the chart statistic cannot take negative values, as with counts or proportions. Such limits should not be confused with specification limits.

3.2 Choosing a chart

The word defective refers to a unit failing one or more requirements; a defect is an individual nonconformity. One defective unit can contain several defects, so p and np charts are not interchangeable with c and u charts.

3.3 Attribute-chart relationships

For a p chart with sample size n and estimated fraction defective p̄, the approximate limits are:

If the calculated lower limit is negative, it is set to zero because a proportion cannot be negative. The formula assumes independent Bernoulli outcomes and a sufficiently suitable approximation; very small expected counts may require an exact or alternative method.

For a c chart, where c̄ is the average number of defects per constant inspection unit:

The Poisson model and approximately constant opportunity must be reasonable. For varying inspection opportunity, a u chart is more appropriate, with centre line ū and limits based on the relevant unit size.

3.4 Interpreting chart signals

A point outside a control limit is strong evidence that the process may have changed, but it is not proof of a particular cause. Other evidence includes:

• a sustained run on one side of the centre line;
• a trend of successive increases or decreases;
• repeated points close to a control limit;
• unusual cycles or stratification; and
• patterns that conflict with the assumed distribution.

Rules for runs and trends should be defined before routine use. Excessive rules increase false alarms, while too few rules can delay detection. A signal should trigger investigation of time, material, machine, method, measurement and personnel conditions—not automatic adjustment of the process.

3.5 Worked example: an individuals chart

An order-processing team records the number of hours required for one order each day. After removing a verified special-cause period, the stable baseline has mean μ = 4.0 hours and standard deviation σ = 0.5 hours. For an individuals chart using the simplified three-sigma relationship:

A later observation of 6.1 hours is above the UCL and should be investigated. It does not automatically demonstrate that every future order will fail, nor does it identify the cause. Investigation might examine system downtime, order complexity, staffing and data-entry problems. If the cause is confirmed and removed, the baseline may be recalculated using an appropriate, documented method. Limits must not be repeatedly recalculated simply to hide deterioration.

4. Process Capability and Performance

Control charts address stability; capability indices compare stable process variation and location with specification requirements. For a two-sided specification with lower specification limit LSL, upper specification limit USL, process mean μ and within-process standard deviation σ, the potential capability index is:

The centred capability index is:

Both indices are dimensionless. They require meaningful, reliable specifications, a reasonably stable process and a defensible estimate of within-process variation. Cp describes the width of the specification interval relative to process spread but does not account for off-centre operation. Cpk accounts for centring. Capability indices should not be used to claim capability when the process is unstable or when the distributional assumptions are inappropriate.

For example, if USL = 10, LSL = 0, μ = 6 and σ = 1:

The process spread is relatively narrow, but the mean is closer to the upper specification. Organisational acceptance thresholds vary by industry and risk; no index should be interpreted without customer, regulatory and process context.

5. Standard Work, Procedures and Knowledge Transfer

A control system must convert the improved process into repeatable work. A standard operating procedure (SOP) should state:

1. purpose and scope;
2. roles and competence requirements;
3. required materials, equipment and software;
4. sequence of critical steps;
5. control points and acceptance criteria;
6. safety, environmental and compliance requirements;
7. abnormal-condition actions;
8. records and data-retention rules; and
9. revision history and approval authority.

Standard work is not intended to eliminate professional judgement. It defines the current best-known method while identifying where authorised judgement is necessary. Visual standards, checklists, photographs, digital prompts and mistake-proofing can improve adherence, but documents should be usable at the point of work rather than stored only in a remote repository.

Training should verify demonstrated competence, not merely attendance. Changes to equipment, materials, software or staffing may require retraining and controlled document revision. Obsolete versions must be removed from points of use.

6. Dashboards and Performance Reviews

A dashboard should support decisions, not simply display large quantities of data. A balanced operational dashboard commonly includes:

• outcome measures: defects, complaints, on-time delivery or cost;
• process measures: cycle time, first-pass yield or adherence to standard work;
• leading indicators: overdue maintenance, training completion or audit findings;
• customer measures: satisfaction, returns or service-level performance; and
• benefit measures: verified savings, capacity released or avoided cost.

Every measure requires a clear definition, owner, data source, frequency, target and escalation rule. Red, amber and green displays are useful only when thresholds are defined and colour is not the sole means of communication. Dashboards should show denominators and time periods; a count of five defects is difficult to interpret without knowing whether it arose from 50 or 50,000 units.

7. Reaction Plans and Escalation

A reaction plan specifies what happens when a control signal, audit failure or performance threshold is observed. A robust plan distinguishes four levels:

The response must be proportionate to risk. Safety, legal and customer-critical conditions may require immediate stop-and-escalate rules. In lower-risk situations, continued operation with increased sampling may be appropriate if authorised. The plan should define who can stop the process, who must be notified, maximum response time and the records required.

A common failure is to react to every fluctuation by changing machine settings. This tampering can increase variation. Conversely, ignoring a sustained shift because no individual point exceeds a limit allows deterioration to become normal. The correct response depends on the type and strength of evidence.

8. Audits, Ownership Transfer and Benefit Realisation

8.1 Layered process audits

Audits should test both compliance and effectiveness. A layered process audit may involve operators, supervisors, process engineers and managers at different frequencies. Questions should be observational and evidence-based: Is the current standard available? Is the measurement device valid? Are reaction records complete? Does the process produce the intended result?

Audits should not become punitive inspections. Their purpose is to expose system weaknesses, remove barriers and verify that controls remain suitable. Repeated findings indicate that the management system, not merely individual behaviour, requires attention.

8.2 Ownership transfer

The improvement team should not retain permanent responsibility for routine performance. Transfer should occur only when:

• the process owner accepts the control plan;
• baseline and improved performance are documented;
• procedures and training are complete;
• data ownership and access are clear;
• reaction and escalation responsibilities are tested;
• open risks and unresolved actions are recorded; and
• the owner agrees to review frequency and benefit measures.

A formal tollgate or handover record creates accountability. The improvement team may remain available for a defined stabilisation period, after which unresolved structural issues should enter the normal management system or a new improvement project.

8.3 Benefit-realisation review

Benefits must be distinguished from activity. Installing a new system or completing training is an output; reduced defects, released capacity or lower cost is an outcome. A benefit-realisation review should compare an agreed baseline with post-implementation performance, adjust for volume and mix, and identify whether gains are financial, operational, customer-related, risk-related or intangible.

Financial claims require transparent assumptions. For example, annualised gross savings may be estimated as:

This relationship assumes that the cost definitions are comparable and that expected volume is realistic. It does not automatically equal cash released. Savings may be avoided cost, capacity value or actual budget reduction; finance and operational owners should agree the classification. Benefits should be reviewed after sufficient time to distinguish temporary improvement from sustained performance.

9. Illustrative Case: Sustaining Improvement in a Medical-Device Packaging Cell

This is a fictional illustrative case.

Background and problem

A medical-device manufacturer experienced a first-pass yield of 92% in a packaging cell. The main failures were incorrect label position and incomplete seal verification. A DMAIC team identified variation in changeover settings, inconsistent verification technique and delayed detection.

Improve actions

The team introduced a keyed fixture, a standard changeover checklist, a validated label-position gauge and an automatic seal-verification interlock. A pilot demonstrated first-pass yield of 98.5% over four weeks, with no evidence that the improvement depended on one exceptional operator.

Control design

The control plan assigned the operator a start-up verification, the supervisor a shift review and the quality engineer a weekly trend review. Label position was recorded as a continuous characteristic using an approved gauge. Seal verification was treated as a critical attribute with an interlock and documented challenge test. The SOP defined changeover sequence, acceptable conditions, equipment checks and actions for failed verification.

An individuals chart was used for the low-volume label-position measurement. A p chart monitored the fraction of defective packaged units by shift. Control limits were calculated from a stable post-improvement baseline and were not treated as product specifications. The dashboard displayed first-pass yield, repeat defects, overdue calibration, open containment actions and verified monthly benefit.

Reaction plan

A point outside a control limit required confirmation of the measurement, segregation of potentially affected product and notification of the supervisor. A failed seal interlock required stopping the cell and quality escalation. The supervisor reviewed material lot, changeover records, equipment status and operator training before authorising restart. Repeated signals initiated a structured root-cause investigation rather than repeated adjustment of settings.

Ownership transfer and outcome

After six weeks, the process owner accepted the control plan. Audits showed 100% availability of current work instructions and 97% adherence to the changeover checklist; the remaining gaps were corrected through improved visual placement and refresher training. Over the next three months, first-pass yield averaged 98.2%, repeat defects declined and the organisation verified a reduction in rework hours. Finance classified the result as a combination of avoided rework cost and released capacity, rather than claiming the entire amount as direct cash saving.

Learning points

The case demonstrates that sustained performance required more than a successful pilot. The improvement was protected by measurement-system discipline, prevention controls, defined escalation, ownership transfer, auditing and cautious financial interpretation.

10. Challenges, Limitations and Governance

Control systems can fail for technical, human and organisational reasons. Important limitations include:

• Autocorrelation: Time-dependent data may violate independence assumptions and make conventional limits misleading.
• Non-normal or bounded data: Alternative charts or transformations may be needed; normality should not be assumed automatically.
• Changing processes: A major redesign may make historical limits irrelevant, requiring a documented baseline transition.
• Small samples: Sparse data can produce unstable estimates and delayed signals.
• Over-monitoring: Excessive charts and alarms create fatigue and false reactions.
• Data integrity: Missing, altered or selectively reported data undermine decisions.
• Metric gaming: If targets are punitive, employees may hide defects, delay recording or optimise one measure at the expense of the system.
• Confusing stability with quality: A stable process can consistently produce unacceptable output.
• Confidentiality and privacy: Dashboards and audit records must restrict access to personal, customer and regulated information.

Governance should define data ownership, access permissions, retention, change control, auditability and approval of limit recalculation. In regulated sectors, electronic records may require traceability, validated systems and controlled signatures. Artificial intelligence can assist with anomaly detection, forecasting and prioritising investigations, but it should not silently change limits or replace accountable process ownership. AI outputs require validated data, explainable thresholds, human review and monitoring for model drift and bias.

11. Best-Practice Checklist

Before closing a DMAIC project, confirm that:

• critical outputs and verified input controls are identified;
• the measurement system is suitable and calibrated;
• the process is stable enough for capability interpretation;
• chart type, subgrouping and limits are justified;
• specifications are clearly distinguished from control limits;
• standard work is current, accessible and understood;
• training and competence have been verified;
• reaction and escalation plans have been tested;
• dashboards contain owners, definitions and decision thresholds;
• audit findings lead to systemic action;
• process ownership has been formally transferred; and
• benefits are reviewed against an agreed baseline.

Chapter Summary

The Control phase converts a DMAIC improvement into sustained organisational performance. Its essential elements are a risk-based control plan, reliable measurement, appropriate control charts, standard work, visual and digital monitoring, defined reaction plans, audits, ownership transfer and benefit-realisation reviews. Control limits describe expected process behaviour; specification limits describe requirements. Stability must generally be established before capability indices are interpreted. Effective responses distinguish common-cause variation from special causes and avoid tampering. Sustainable control also depends on governance: data integrity, document control, competence, escalation, privacy and accountable ownership. The final test of DMAIC is not whether an improvement worked during a project, but whether the process continues to deliver value after the project team has withdrawn.

Key Terms

Control phase: DMAIC phase that sustains and manages an improved process.

Control plan: Document specifying what is controlled, how it is measured and how abnormal conditions are handled.

Statistical process control (SPC): Use of statistical methods to monitor and improve process behaviour.

Common-cause variation: Routine variation arising from the process system.

Special cause: An unusual, identifiable source of process variation.

Control limit: Statistically derived boundary used to assess process behaviour.

Specification limit: Requirement boundary defining acceptable product or service performance.

Centre line: Central reference value on a control chart.

Individuals chart: Chart for individual continuous observations, often paired with a moving-range chart.

X̄–R chart: Chart monitoring subgroup averages and ranges.

p chart: Attribute chart for the fraction of defective units.

c chart: Attribute chart for defect counts with constant inspection opportunity.

Process capability: Extent to which a stable process can meet specifications.

Cp: Capability index comparing specification width with six standard deviations.

Cpk: Capability index accounting for process centring relative to the nearest specification limit.

Standard operating procedure: Controlled document defining an approved method of work.

Reaction plan: Predefined response to a signal, failure or abnormal condition.

Layered process audit: Audit system involving different organisational levels at planned frequencies.

Ownership transfer: Formal movement of routine accountability from the project team to the process owner.

Benefit realisation: Verification that planned operational, customer or financial gains have actually occurred.

Review Questions

Short Answer Questions

1. What is the primary purpose of the Control phase in DMAIC?
2. Distinguish a control limit from a specification limit.
3. What is the difference between a defect and a defective unit?
4. Why should a measurement system be evaluated before relying on a control chart?
5. What information should a reaction plan contain?

Long Answer Questions

1. Explain how a control plan, standard operating procedure and dashboard work together to sustain an improvement.
2. Compare X̄–R, I–MR, p, np, c and u charts, including the data conditions for selecting each.
3. Explain common-cause and special-cause variation and discuss why tampering is harmful.
4. Discuss the requirements for effective ownership transfer at the end of a DMAIC project.
5. Explain why process stability is important before calculating and interpreting Cp and Cpk.

Analytical / Application Questions

1. A service process has a stable mean completion time of 12 minutes and standard deviation of 2 minutes. Using a simplified individuals-chart approach, calculate the three-sigma control limits and explain the action required if an observation is 19 minutes.
2. A factory records 8 defective units in a sample of 200, then 18 defective units in a sample of 300. Explain why comparing raw counts is misleading and identify the more appropriate control-chart approach.
3. Design a control-plan outline for a process in your organisation. Include at least two critical characteristics, measurement methods, sampling frequencies, owners, control criteria and reaction actions.

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 statistically.

B. A control limit describes process behaviour, while a specification limit describes an acceptance requirement.

C. Both are calculated from the process mean and standard deviation.

D. A specification limit is used only for attribute data.

Correct Answer: B. A control limit describes process behaviour, while a specification limit describes an acceptance requirement.

Question 2

Which chart is generally appropriate for monitoring the fraction of defective units when sample size may vary?

A. c chart

B. X̄–R chart

C. p chart

D. I–MR chart

Correct Answer: C. p chart

Question 3

A stable process has μ = 20 units and σ = 2 units. Using a simplified three-sigma relationship, what is the UCL for individual observations?

A. 14 units

B. 18 units

C. 22 units

D. 26 units

Correct Answer: D. 26 units

Question 4

What should normally happen when a control chart produces a credible special-cause signal?

A. Recalculate the limits immediately to include the point

B. Investigate the cause and protect potentially affected output

C. Adjust the process after every fluctuation

D. Ignore the signal unless the customer complains

Correct Answer: B. Investigate the cause and protect potentially affected output

Question 5

Which index accounts for both process spread and displacement from the nearest specification limit?

A. Cp

B. Cpk

C. p̄

D. UCL

Correct Answer: B. Cpk

Question 6

A c chart is most appropriate when monitoring:

A. The proportion of defective units with varying sample sizes

B. The number of defects per constant inspection unit

C. Continuous measurements from subgroups

D. The time between rare events only

Correct Answer: B. The number of defects per constant inspection unit

Question 7

Which item is essential in a reaction plan?

A. A guarantee that no variation will occur

B. A predefined owner, containment action and escalation route

C. Automatic recalculation of specification limits

D. Removal of all audit records

Correct Answer: B. A predefined owner, containment action and escalation route

Question 8

Why is repeated adjustment in response to ordinary random fluctuation potentially harmful?

A. It can increase process variation through tampering.

B. It always lowers the process mean.

C. It converts specifications into control limits.

D. It eliminates the need for measurement.

Correct Answer: A. It can increase process variation through tampering.

Question 9

Which is an outcome measure rather than merely an improvement activity?

A. Installing a dashboard

B. Completing operator training

C. Reducing verified rework cost

D. Approving an SOP

Correct Answer: C. Reducing verified rework cost

Question 10

Before routine ownership is transferred to the process owner, which condition is most appropriate?

A. The project team retains all operational responsibility indefinitely.

B. The control plan, training, escalation duties and benefit measures are accepted and documented.

C. All historical data are deleted.

D. Control limits are changed to match the target specification.

Correct Answer: B. The control plan, training, escalation duties and benefit measures are accepted and documented.