7. Six Sigma and Design for Quality

Introduction and Context

Six Sigma and Design for Six Sigma (DFSS) provide two complementary approaches to operational excellence. Six Sigma improves an existing process by reducing variation and defects; DFSS designs a new product, service, or process so that customer and business requirements are built into it from the beginning. Together, they connect customer expectations, process knowledge, statistical analysis, risk management, and disciplined project execution.

The distinction is important. DMAIC—Define, Measure, Analyse, Improve and Control—is appropriate when a process exists but is not achieving required performance. DFSS is more suitable when a new process is being created, a product is being redesigned, or an existing process cannot meet requirements through incremental improvement. Common DFSS road maps include DMADV (Define, Measure, Analyse, Design and Verify) and IDOV (Identify, Design, Optimise and Validate).

At postgraduate level, Six Sigma should not be viewed merely as a collection of statistical tools. It is a managerial system for selecting high-value projects, converting customer needs into measurable requirements, testing evidence before making decisions, and sustaining gains through process ownership and control. Its effectiveness depends on leadership, data quality, appropriate methods, and the integration of technical analysis with organisational change.

Learning Objectives

After studying this chapter, students should be able to:

1. Explain the logic, terminology and managerial principles of Six Sigma and DFSS.

2. Develop a project charter and translate the voice of the customer into critical-to-quality requirements.

3. Select and interpret appropriate measures of process performance and capability.

4. Apply regression, design of experiments, robust design and FMEA in improvement and design projects.

5. Distinguish between DMAIC and DFSS and select an appropriate project pathway.

6. Develop control plans that sustain process performance.

7. Evaluate risks, limitations, ethical issues and digital applications in Six Sigma projects.

Six Sigma as a Management and Improvement System

A process produces outputs through a combination of inputs, methods, people, equipment, materials, information and environmental conditions. Six Sigma expresses this relationship as Y = f(X), where Y represents an output or customer result and X represents the factors that influence it. Improvement consists of understanding which X variables materially affect Y, controlling them, and reducing unwanted variation.

Variation may be common-cause or special-cause. Common-cause variation is inherent in the current system and generally requires changes to process design, technology, policy or operating conditions. Special-cause variation arises from an unusual and identifiable event, such as a damaged tool, incorrect setting or data-entry error. A process should not be improved by tampering with stable common-cause variation, nor should special causes be treated as if they were permanent characteristics of the system.

The familiar Six Sigma metric is associated with a defect rate of approximately 3.4 defects per million opportunities under a conventional long-term shift assumption. This figure is a benchmark, not a universal guarantee. Capability indices, defect opportunities, measurement definitions and the stability of the process must be examined before performance claims are made. A process can have a favourable average while producing unacceptable variation, or it can be statistically capable but fail because the customer requirement was defined incorrectly.

Six Sigma projects normally have an executive sponsor, process owner, Master Black Belt or programme leader, Black Belts who lead complex projects, and Green Belts who lead or support smaller projects. These roles clarify accountability, but statistical expertise does not replace operational knowledge. The people who perform the work often understand sources of variation that are invisible in formal data.

From Customer Voice to CTQs

The starting point for a credible project is not a preferred solution but a customer or business problem. The voice of the customer may be obtained from interviews, complaints, service records, observations, surveys, market research, regulatory requirements or internal users. These statements are often qualitative—for example, “the delivery should be reliable” or “the application must be easy to use.” They must be translated into measurable critical-to-quality (CTQ) characteristics.

A CTQ tree typically moves from a broad need to a driver and then to an operational measure. For example:

• Need: reliable delivery.
• Driver: delivery consistency.
• CTQ: percentage of orders delivered within the promised time window.
• Specification: at least 98% within the agreed window.

A CTQ should have an operational definition. It must state what is measured, how it is measured, the unit, the sampling rule, the data source and the acceptance criterion. Poorly defined CTQs create disputes during analysis and can encourage teams to optimise an internal measure that does not represent customer value.

Project Charters and Project Selection

A project charter is the governance document that establishes why a project exists and how it will be managed. A sound charter normally includes the business case, problem statement, goal statement, scope, customers, CTQs, baseline performance, anticipated benefits, milestones, team roles, assumptions and constraints.

The problem statement should describe the gap without prematurely asserting its cause or solution. “Invoice cycle time averaged 11 days against a customer requirement of five days during the previous quarter” is stronger than “the invoicing software is inefficient.” The goal statement should be specific and time-bound, while remaining achievable and analytically defensible.

Project selection should consider strategic importance, customer impact, financial or social value, feasibility, urgency, data availability and the authority of the team to change the process. A project with a large theoretical benefit may fail if its scope crosses several departments without sponsorship. Conversely, a small project may be strategically valuable if it removes a major source of customer dissatisfaction or regulatory risk.

DMAIC: Improving an Existing Process

Define

The Define phase establishes the problem, customer, scope and expected outcome. The team develops the charter, high-level process map and initial CTQs. SIPOC—suppliers, inputs, process, outputs and customers—is useful for defining boundaries without prematurely mapping every activity. A tollgate review should confirm that the problem is meaningful, the team has authority, and the project is not simply a solution seeking justification.

Measure

Measure establishes baseline performance and tests whether the measurement system is adequate. The team defines operational measures, develops a data-collection plan, assesses sampling bias and verifies the measurement process. For continuous measurements, gauge repeatability and reproducibility studies examine variation attributable to appraisers and equipment. For attribute judgments, agreement analysis may be required.

Baseline measures may include defect proportion, defects per unit, defects per opportunity, cycle time, yield, first-pass yield, mean, standard deviation and process capability. Data should be stratified by relevant factors such as product family, shift, location, supplier or transaction type. Aggregated averages can conceal important patterns.

Analyse

Analysis identifies and validates potential causes. The team may use process maps, cause-and-effect diagrams, Pareto analysis, failure data, scatter plots, hypothesis tests, regression and analysis of variance. Brainstormed causes are hypotheses, not conclusions. A statistically significant association may be operationally unimportant, and a non-significant result may reflect inadequate sample size or poor measurement.

Regression models describe the relationship between a response and one or more predictors. A model such as Y = β0 + β1X1 + β2X2 + ε estimates how the expected response changes with the predictors. Analysts should examine residuals, linearity, independence, constant variance, influential observations and multicollinearity. Regression supports prediction and prioritisation, but observational association does not by itself establish causation. Experimental confirmation is often needed.

Improve

In Improve, the team develops and tests changes that address validated causes. Pilot trials, simulation, mistake-proofing, standard-work redesign and design of experiments may be used. Improvements should be assessed against the CTQ and unintended consequences. A local reduction in cycle time is not an improvement if it increases defects, safety risk or downstream workload.

Control

Control institutionalises the improved process. The team updates standard work, training, documentation, visual controls, ownership and escalation rules. Statistical process control charts can distinguish routine variation from signals requiring investigation. The control plan specifies the CTQ, measurement method, sampling frequency, responsible person, reaction plan and record requirements. Benefits should be verified after implementation, and the process owner must accept accountability before project closure.

Process Capability and Statistical Control

Capability compares process variation and centring with specification limits. For a stable approximately normal process, Cp = (USL − LSL)/(6σ) reflects potential capability, while Cpk = minimum[(USL − μ)/(3σ), (μ − LSL)/(3σ)] also reflects centring. A high Cp with a low Cpk indicates that the process may be sufficiently narrow but poorly centred. Capability indices are misleading when the process is unstable, the distribution is markedly non-normal, the specification is inappropriate, or the measurement system is inadequate.

Control limits are calculated from process data and describe expected process behaviour; specification limits are established by customers, designers, regulators or business requirements. They are not interchangeable. A point outside a control limit indicates unusual process behaviour, whereas a point outside a specification indicates nonconformance. A process may be stable but incapable, or capable in a short-term study but unstable over time.

Design for Six Sigma

DFSS begins when the existing process is not the right foundation for improvement or when requirements must be built into a new design. DMADV provides a structured route:

• Define customer, business and regulatory requirements.
• Measure CTQs, design targets, constraints and risks.
• Analyse alternative concepts and the relationships between requirements and design parameters.
• Design the product or process using modelling, experiments, tolerance analysis and mistake-proofing.
• Verify performance through testing, pilot production, validation and transfer to operations.

Quality Function Deployment (QFD) can translate customer requirements into technical characteristics and expose competing priorities. Design requirements should be measurable, traceable and linked to verification methods. DFSS also considers manufacturability, serviceability, safety, accessibility, cybersecurity where relevant, lifecycle cost and environmental impact—not merely conformance at final inspection.

The quality loss perspective associated with robust design emphasises that loss can increase as performance moves away from the target, even when the result remains inside specification. This differs from a narrow pass/fail view. Robust design seeks settings that make performance less sensitive to uncontrollable noise factors. In a robust experiment, control factors are deliberately selected while noise factors are varied or represented so that the design performs consistently under realistic conditions.

Design of Experiments

Design of experiments (DOE) changes selected factors systematically to estimate main effects and interactions efficiently. A factorial design can reveal that the effect of temperature depends on pressure—an interaction that one-factor-at-a-time experimentation may miss. Screening designs identify a smaller set of influential factors; response-surface methods then model and optimise the response within a defined region.

A responsible DOE begins with a clear response, factor definitions, ranges, practical constraints, randomisation, replication and, where appropriate, blocking. Randomisation reduces the influence of time-related or uncontrolled effects. Replication estimates experimental error, while blocking controls known nuisance sources. The experiment must be feasible and safe; statistical efficiency cannot justify exposing employees, customers or equipment to unacceptable risk.

Results should be interpreted using effect estimates, confidence intervals, practical significance, residual diagnostics and confirmation runs. Optimising a mathematical model outside the tested region is unsafe. Confirmation at the proposed settings is essential before full-scale implementation.

FMEA and Control Planning

Failure Modes and Effects Analysis (FMEA) is a structured method for anticipating how a product or process can fail, what effects may result, what causes are plausible, and how current controls address the risk. Typical fields include process step, failure mode, effect, cause, prevention controls, detection controls, severity, occurrence and detectability. Traditional risk-priority-number methods multiply ratings, but multiplication can hide an extremely severe failure behind moderate scores. Many organisations therefore use action-priority approaches and require explicit review of high-severity risks.

FMEA is most valuable when performed cross-functionally and early enough to influence design. It is not a one-time form-filling exercise. It should be updated after failures, process changes, customer complaints, audit findings and control-plan revisions. Prevention is generally stronger than detection: eliminating an incorrect connection through keyed design is preferable to relying only on final inspection.

A control plan converts risk and process knowledge into operating discipline. It identifies process steps, characteristics to control, specifications, measurement methods, sampling, reaction plans, records and responsible roles. Effective control plans are usable at the workplace, linked to visual or digital instructions, and reviewed when process conditions change.

An Illustrative Case: Reducing Medication-Order Processing Errors

Background

The following is a fictional illustrative case. A hospital pharmacy received electronic medication orders from several clinical departments. The pharmacy’s strategic objective was to reduce delays and transcription-related errors while maintaining patient safety.

Situation and problem

A preliminary review showed that approximately 7% of orders required clarification before dispensing. The project team initially suspected that the electronic interface was the primary cause. The charter defined the problem as excessive clarification and rework in the order-to-dispense process, with scope from order receipt to pharmacist verification. Patient safety was treated as a non-negotiable constraint.

Analysis

The team created a SIPOC and stratified data by department, order type, shift and medication category. A measurement review found inconsistent definitions of “clarification”; the team established an operational definition and trained reviewers. Pareto analysis showed that missing dose units, ambiguous frequency instructions and incomplete allergy information accounted for most cases. A regression model suggested that order complexity and department were associated with clarification, but the team did not treat the association as proof of cause. Direct observation showed that the highest-risk fields were displayed on separate screens.

An FMEA rated ambiguous frequency as high severity because of its potential patient impact. The team therefore prioritised prevention rather than merely adding a final check. A DOE in a test environment evaluated screen layout, mandatory prompts and terminology choices, with completion time and error-free entry as responses.

Approach and solution

The selected design placed dose, unit, route and frequency together; used standard terminology; required completion of critical fields; and provided context-sensitive prompts. A pilot was run in one department with pharmacist review, safety monitoring and a rollback plan. The control plan specified daily monitoring during the pilot, weekly review thereafter, an escalation rule for any serious near miss, and periodic audit of override behaviour.

Outcome

The pilot demonstrated lower clarification and rework, but the team also detected an increase in harmless alert overrides. Rather than declaring success solely from the original metric, the team refined the prompts and added an override-reason review. The redesigned process was then transferred gradually to other departments. The case illustrates that CTQs, risk analysis, experimentation and control planning must be integrated; a faster process is not acceptable if it weakens safety.

Learning points

The project succeeded because it corrected the measurement definition, analysed the process rather than blaming individuals, used design evidence to address a high-severity risk, and treated unintended effects as part of the improvement review. It also shows why the process owner and frontline professionals must remain involved after the statistical analysis is complete.

Common Failure Modes and Limitations

Six Sigma programmes can become bureaucratic when teams mistake completed templates for demonstrated improvement. Other common errors include selecting projects without strategic relevance, beginning with a solution, using unreliable data, confusing correlation with causation, ignoring non-normal data, applying capability indices to unstable processes, overfitting regression models, treating p-values as managerial decisions, and closing a project before benefits are sustained.

A further limitation is that not every problem is primarily a variation problem. A legal prohibition, capacity shortage, ethical conflict or fundamentally wrong business model may require policy, investment or strategic redesign rather than DMAIC. Statistical methods also require judgement about assumptions, sampling and practical significance. Six Sigma should therefore complement, not replace, systems thinking, lean flow improvement, human-factors engineering and professional expertise.

Governance, Ethics and Digital Practice

Project teams must protect confidential customer, employee and patient information. Data collection should have a legitimate purpose, appropriate access controls, retention rules and transparent definitions. Analysts should not manipulate inclusion criteria, remove inconvenient observations without justification, or report only favourable measures. Automation can accelerate data capture, anomaly detection and control-chart monitoring, but algorithmic alerts may reproduce biased historical decisions or create excessive false alarms.

Digital twins, process mining, Internet-connected equipment and machine-learning models increasingly support process discovery and prediction. These tools are useful when data lineage, model validation, cybersecurity, human oversight and change control are established. A predictive model should be monitored for drift and should not be treated as an unquestionable causal explanation. In regulated or safety-critical environments, validation, traceability and documented approval are essential.

Best Practices for Professional Application

Effective projects generally follow several principles:

• begin with a verified customer or business problem;
• define CTQs and measures operationally;
• establish measurement reliability before analysing causes;
• distinguish stability from capability;
• combine statistical evidence with process observation and expert knowledge;
• use experiments when causal claims matter;
• prioritise prevention and robust design over inspection;
• involve process owners and affected employees;
• test changes on a controlled scale and examine unintended effects;
• assign clear control ownership and reaction rules; and
• review benefits, risks and assumptions after implementation.

Chapter Summary

Six Sigma is a disciplined system for reducing defects and variation, while DFSS builds quality and reliability into new products and processes. DMAIC guides improvement of an existing process: Define the problem, Measure the baseline and measurement system, Analyse validated causes, Improve through tested changes, and Control the gains. Project charters, CTQs, capability analysis, regression, DOE, FMEA and control plans provide the evidence and governance needed at each stage.

Process stability must be distinguished from capability, and specification limits from control limits. Regression identifies relationships but does not automatically prove causation; DOE provides stronger evidence when factors can be deliberately varied. DFSS methods such as DMADV, QFD and robust design are appropriate when quality must be designed rather than repaired. Successful application requires ethical data use, cross-functional participation, risk-based thinking, digital governance and sustained process ownership.

Key Terms

Six Sigma: A management and analytical approach for reducing process variation and defects.

DMAIC: Define, Measure, Analyse, Improve and Control, the principal road map for existing-process improvement.

DFSS: Design for Six Sigma, an approach that embeds quality and reliability into new designs.

DMADV: Define, Measure, Analyse, Design and Verify, a common DFSS road map.

CTQ: Critical-to-quality characteristic that translates customer or regulatory needs into measurable requirements.

Project charter: A document defining a project’s business case, scope, objectives, roles and governance.

SIPOC: A high-level view of suppliers, inputs, process, outputs and customers.

Common-cause variation: Routine variation inherent in the current process system.

Special-cause variation: Unusual variation attributable to an identifiable event or condition.

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

Cp: An index comparing specification width with six standard deviations of process variation.

Cpk: A capability index that considers both variation and process centring.

Regression: A method for modelling the relationship between a response and predictor variables.

Design of experiments: A planned method for studying factor effects and interactions.

Robust design: Designing settings that reduce sensitivity to uncontrollable noise factors.

FMEA: A structured analysis of potential failure modes, effects, causes and controls.

Control plan: A documented scheme for monitoring process characteristics and responding to problems.

Measurement-system analysis: Evaluation of whether a measurement process is reliable for its intended use.

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

Quality Function Deployment: A method for translating customer requirements into technical design characteristics.

Review Questions

Short Answer Questions

1. Why must a CTQ have an operational definition?

2. Distinguish common-cause from special-cause variation.

3. What is the difference between Cp and Cpk?

4. Why is a measurement-system study important before process analysis?

5. State two differences between DMAIC and DFSS.

Long Answer Questions

1. Explain each phase of DMAIC and describe the principal decisions made in each phase.

2. Discuss how a project charter supports governance and project selection.

3. Explain the role of regression and DOE in identifying and validating process causes.

4. Describe how FMEA and a control plan work together to prevent recurrence of failures.

5. Compare traditional inspection-based quality with robust design and DFSS.

Analytical / Application Questions

1. A stable filling process has a high Cp but a substantially lower Cpk. Interpret this result and recommend an improvement approach.

2. A service team reports that average response time has improved, but customer complaints have increased. Design a DMAIC investigation, including CTQs, stratification variables and possible control measures.

3. Using the illustrative pharmacy case, explain why a mandatory prompt alone would not be sufficient evidence of a successful improvement. Propose measures for evaluating safety, efficiency and unintended effects.

22. Multiple Choice Questions with Answers

Question 1

Which situation most appropriately calls for DMAIC rather than DFSS?

A. A new medical device is being designed without an existing production process

B. An existing billing process has excessive defects and can be measured reliably

C. A company is selecting a new market strategy

D. A process must be created to meet an entirely new regulatory requirement

Correct Answer: B. An existing billing process has excessive defects and can be measured reliably

Question 2

A project team discovers that its defect rate changes substantially depending on who performs the inspection. What should it do first?

A. Calculate Cpk immediately

B. Increase the specification limits

C. Investigate the measurement system

D. Run a response-surface experiment

Correct Answer: C. Investigate the measurement system

Question 3

A process has a high Cp but a much lower Cpk. What is the most likely interpretation?

A. The process is unstable because Cp is high

B. The process variation is narrow but the mean is poorly centred

C. The specification limits are identical to control limits

D. The measurement system has been proven inadequate

Correct Answer: B. The process variation is narrow but the mean is poorly centred

Question 4

Why is randomisation important in a design of experiments?

A. It guarantees that every factor has a significant effect

B. It removes the need for replication

C. It reduces the influence of time-related and uncontrolled effects

D. It allows experimentation outside safe operating limits

Correct Answer: C. It reduces the influence of time-related and uncontrolled effects

Question 5

Which statement best describes the relationship between control limits and specification limits?

A. Both are always calculated from the same data

B. Control limits describe process behaviour, while specifications define acceptability

C. Specification limits identify special causes, while control limits define customer needs

D. They are interchangeable when the process is normally distributed

Correct Answer: B. Control limits describe process behaviour, while specifications define acceptability

Question 6

A regression analysis finds that department is associated with order clarification. What is the most appropriate conclusion?

A. Department has been proven to be the direct cause

B. The relationship is a hypothesis requiring further investigation or experimentation

C. The process should be changed without a pilot

D. The result is irrelevant unless the coefficient is exactly one

Correct Answer: B. The relationship is a hypothesis requiring further investigation or experimentation

Question 7

Which FMEA response generally provides stronger risk reduction?

A. Adding a final inspection without changing the process

B. Increasing the number of forms completed by operators

C. Eliminating the possibility of an incorrect connection through keyed design

D. Reducing the severity score after training

Correct Answer: C. Eliminating the possibility of an incorrect connection through keyed design

Question 8

What is the primary purpose of a control plan?

A. To identify the original business case for a project

B. To specify ongoing monitoring, responsibility and reaction to process problems

C. To replace all standard operating procedures

D. To calculate the long-term Six Sigma benchmark

Correct Answer: B. To specify ongoing monitoring, responsibility and reaction to process problems

Question 9

In robust design, noise factors are considered because they are factors that:

A. Are always statistically insignificant

B. Cannot or should not be tightly controlled during normal operation

C. Must be excluded from all experiments

D. Represent only measurement errors

Correct Answer: B. Cannot or should not be tightly controlled during normal operation

Question 10

In the pharmacy case, why was a lower clarification rate alone insufficient as a success criterion?

A. Clarification is never a useful measure

B. The project had to consider patient safety and unintended alert overrides

C. Only financial measures can be used in healthcare projects

D. A pilot cannot be used for process improvement

Correct Answer: B. The project had to consider patient safety and unintended alert overrides