Assuring Quality in Operations

Need for Six Sigma

Six Sigma is a mechanism to deliver near zero defects in operations by applying process control principles. A defect is any unacceptable state of a product or service from a customer’s perspective. At Six Sigma quality levels, defects become extraordinarily rare – e.g., a few defects per million units produced or per million service opportunities.

Why 99% quality is not enough Two examples illustrate the tangible impact of seemingly high quality levels.

Example 1: Five-star hotel at 99% quality

Example 2: Insurance firm

28,965 defects per year – hardly acceptable. Quality levels must far exceed 99–99.5%.

Why high quality reduces total cost A superior quality management system leads to fewer disruptions, smoother output, less rework, higher quality finished goods, and lower indirect costs (inspection, correction). Net effects: lower inventory, higher productivity, less labour, and lower overall cost – contradicting the old belief that quality improvements are uneconomical.

Exam tip: The shift from “quality costs more” to “quality costs less” is a foundational insight of modern quality management.

Measurement Metrics for Six Sigma Quality

Traditional metrics (AQL, AOQL) are based on percentages and inadequate for near-zero defect targets. Six Sigma uses two equivalent metrics:

• Parts Per Million (PPM) – defects per million units produced (manufacturing).
• Defects Per Million Opportunities (DPMO) – defects per million defect opportunities (services).

DPMO formula Let

• $K$ = number of opportunities for a defect per unit of process execution
• $n$ = number of units observed
• $d$ = number of defects observed

$$\text{DPMO} = \frac{d}{n \times K} \times 1{,}000{,}000$$

Worked example – hotel check-in process

• $K = 11$ (e.g., wrong name, missing details)
• $n = 1{,}250$ guests handled
• $d = 357$ defects observed

$$\text{DPMO} = \frac{357}{1{,}250 \times 11} \times 1{,}000{,}000 = \frac{357}{13{,}750} \times 1{,}000{,}000 \approx 25{,}964$$

Premises of Six Sigma Quality

To achieve PPM/DPMO levels, quality management must rest on four premises:

1. Continuous and data-driven – not a one-time event.
2. Prevention and elimination, not detection and correction.
3. Performance standard is zero defects (practically, near-zero).
4. Primary responsibility lies with those who produce and deliver (not with a separate quality control department).

DMAIC Methodology

The Six Sigma program provides a structured, five-phase improvement cycle:

• Define – scope, project charter, goals.
• Measure – identify variables, data collection, baseline metrics.
• Analyze – graphical tools, identify variation sources, vital few root causes.
• Improve – generate and validate alternatives, new process maps.
• Control – control plan, revised standards, training, prevent backsliding.

Organizational Structure for Six Sigma

Sustainable improvement requires clear ownership, mandate, and support. Key roles:

Exam tip: Know the hierarchy: Sponsor -> Master Black Belt -> Black Belt -> Green Belt. The process owner is often the person whose area is being improved.

Key takeaways

• Six Sigma aims for near-zero defects, measured in PPM or DPMO.
• 99% quality is unacceptable – demonstrated with hotel and insurance examples.
• Higher quality reduces total cost via fewer disruptions and lower waste.
• DPMO = (defects) / (units × opportunities per unit) × 1,000,000.
• DMAIC (Define–Measure–Analyze–Improve–Control) is the core improvement cycle.
• A dedicated structure (process owner, belts, coach, sponsor) ensures ownership and sustainability.

Quality Gurus and their Teachings

The modern quality movement did not arise from a single insight. Instead, several thinkers – quality gurus – reshaped how organizations view defects, management’s role, and the tools needed for sustained improvement. Their combined teachings form the foundation of contemporary quality assurance.

The Major Gurus and Their Core Contributions

Four Overarching Prescriptions from the Gurus

1. New definitions of quality – e.g., conformance to standards, target-driven performance.
2. New methods to build quality in – prevention, mistake-proofing, robust design.
3. New tools to assess performance – control charts, cause-and-effect diagrams, quality costing.
4. Changed roles for managers – from control to facilitation; middle managers become enablers, not inspectors.

Exam tip: The gurus are often tested by matching the person to the concept (e.g., “Who introduced zero defects?” → Crosby; “Who developed the Quality Trilogy?” → Juran). Know the pairings cold.

Key takeaways

• Deming: top management leadership + PDCA + 14 points.
• Juran: Trilogy (planning, control, improvement).
• Crosby: five absolutes, zero defects, prevention.
• Ishikawa: fishbone diagram, CEDAC.
• Shingo: poka-yoke (mistake-proofing).
• Taguchi: loss function, design of experiments.
• Four broad directions: alternative definitions, build-in methods, assessment tools, facilitation roles.

---

Total Quality Management (TQM)

TQM is an organization-wide mechanism, backed by top management mandate, to systematically and sustainably solve problems. (Contrast with JIT, which exposes problems; TQM solves them.)

The Four Pillars of TQM

• Top management commitment: Lead by example; signal quality’s importance at every opportunity; help middle management resolve short-term vs. long-term trade-offs (e.g., temporary production loss for lasting improvement).
• Employee involvement: Build a culture of process ownership; train staff in simple tools; foster teamwork; implement project-by-project continuous improvement (structured, not ad hoc).
• Tools and techniques: Provide operational personnel with data-driven methods to identify, analyze, and solve problems.
• Quality system: A systematic way to capture and report quality issues – documentation, learning, and certification (e.g., ISO).

The “Total” in TQM: Three Dimensions

• Everyone – every employee participates.
• Everywhere – all functions and locations.
• Every time – continuous, not one-off.

Components of a Robust Quality Assurance System

1. Top management commitment to quality.
2. Mechanisms to understand customer needs.
3. Translation of needs into measurable operating targets.
4. Mechanisms to identify quality problems.
5. A set of tools and techniques for employees (root-cause tracking, corrective actions).
6. Employee involvement driving continuous improvement.
7. Methods for preventing recurrence of problems.
8. Documentation of all initiatives for learning.
9. Quality certification and benchmarking (e.g., ISO).

Exam tip: The contrast between JIT (“expose problems”) and TQM (“solve problems”) is a frequent short-answer question. Also remember that TQM requires mandate and top management support.

Key takeaways

• TQM = systematic, sustainable problem-solving across the whole organization.
• Four pillars: top management, employee involvement, tools, quality system.
• “Total” = everyone, everywhere, every time.
• Top management leads, signals, and resolves trade-offs.
• Employee involvement requires process ownership, training, teamwork, and structured projects.
• Quality system = documentation + certification (ISO).

---

Tools for Quality Management in Operations – Part I

Quality tools serve two overarching purposes:

1. Ensure data-based decisions – no quality initiative without measurement.
2. Engage operational personnel – tools empower frontline workers to participate.

Two Categories of Tools

• Quality at operations tools are used by frontline teams to spot issues, analyze causes, and drive day-to-day improvements.
• Quality planning and design tools are used during product/service development and for strategic alignment (e.g., translating customer needs into design targets, quantifying the cost of quality).

Key takeaways

• Two roles of tools: ensure data-driven action and involve operational staff.
• Category 1 – operations: control charts, fishbone, CEDAC, and general data plotting.
• Category 2 – planning/design: tree diagrams, poka-yoke, QFD, quality costing.
• Root-cause analysis can be done at operations level (fishbone) or planning level (affinity/relationship diagrams).

Operational Tools

Histogram and Pareto Diagram

A histogram is a bar chart of frequency of occurrence. It gives an immediate visual sense of which problems occur most often. Example: In an earth-moving equipment manufacturer, data collected over a month showed frequencies of quality issues:

A Pareto diagram reorders the histogram bars in descending order and adds a cumulative percentage line (in blue). This makes the vital few stand out. In the example, the first three categories together contributed ~85% of all problems. A deeper rework analysis revealed:

Exam tip: The Pareto principle (80/20 rule) is embedded here: focus on the tall bars first to achieve the largest improvement with the least effort.

Cause-and-Effect (Fishbone) Diagram

Also called Ishikawa diagram or fishbone diagram. The effect (the quality problem) is placed at the right end of a horizontal arrow. Bones (branches) represent major cause categories — typically methods, men, machines, materials (or process, equipment, labour). Sub-causes branch off each bone.

Example: Head breakage of a screw.

• Material: less grade, hardness variation
• Method: less strength due to thread runout not provided or specification missing
• Machines and Men also had sub-causes.

The diagram is built through group brainstorming.

CEDAC (Cause-and-Effect Diagram with Addition of Cards)

A variation of the fishbone diagram. Instead of a fixed group, an open board is used with two sets of blank cards: problem cards and solution cards. Any employee can contribute, tapping the expertise of the entire workforce. Kaoru Ishikawa proposed both the fishbone and CEDAC.

Key takeaways – Operational Tools

• Histograms display frequency; Pareto diagrams sort frequencies descending with cumulative line to highlight vital few.
• Cause-and-effect diagrams systematically organize potential causes under major categories.
• CEDAC opens problem-solving to all employees via cards.
• These tools improve visibility of quality problems, fuel imagination, and motivate thinking about causes and solutions.

---

Design and Planning Tools

Poka‑Yoke (Mistake‑Proofing)

Poka‑Yoke is a Japanese term meaning fool-proofing or mistake-proofing. Its premise: many defects are avoidable; errors and defects have cause–effect relationships, so removing the cause eliminates the defect. A product or process is modified so that errors cannot occur or are detected immediately.

Examples:

• SIM card, pen drive, RAM: shaped asymmetrically so they cannot be inserted wrongly.
• Radial drilling machine: Two sensors and a contact lever ensure a through‑hole is always drilled. The lever must touch the lower sensor before the wheel can be retracted upward. A counter automatically increments each time the lever touches the lower sensor; when it reaches 100 pieces, the batch is sent to the next process—improving both quality and productivity.

Matrix Diagram

A two‑dimensional matrix used for strategic analysis. Dimensions:

• Importance of attributes: Order winning, Order qualifying, Less important
• Company performance relative to competition: Worse, Same, Better

Attributes are plotted to create zones:

• Excess zone: attribute is unimportant but company performance is better than competition.
• Critical zone (urgent action): attribute is highly important (order winning) but company performance is worse.
• Appropriate zone: middle ground.

Example from an earth-moving equipment manufacturer (10 attributes):

• Attribute A (product cost): order qualifying, worse → moderate priority.
• Attribute F (delivery reliability): order winning, worse → high priority (consistent with high lead times found in the study).

Quality Function Deployment (QFD) — House of Quality

QFD (also called House of Quality) links customer needs all the way to process plans through four stages. The first stage builds a “house” with these elements:

1. Customer requirements (the WHATs) — e.g., steaming hot food, easy to carry home, quick order processing.
2. Importance ratings for each requirement (e.g., steaming hot = very important).
3. Product characteristics (the HOWs) — e.g., temperature of cooked item, time to cook, order processing time, number of tables.
4. Relationship matrix — symbols (e.g., double plus, minus) indicate how strongly each characteristic affects each requirement.
5. Trade‑off matrix (the roof) — shows correlations between characteristics (positive or negative).
6. Competitive benchmarking — compares own company with competitors on each requirement.

A restaurant example:

• Steaming hot is strongly positively related to temperature and cooking time.
• Quick order processing is negatively related to cooking time (longer cooking hurts speed).
• Order processing time is negatively correlated with number of service counters during peak time.

The output of the first house becomes the input to the second house, and so on, until customer needs drive the final process plan.

Key takeaways – Design and Planning Tools

• Poka‑Yoke prevents errors by design; examples include shaped slots and sensor‑controlled machines.
• Matrix diagram prioritizes improvement actions by plotting attribute importance vs. company performance.
• QFD (House of Quality) systematically translates customer requirements into product/process characteristics, capturing trade‑offs and competitive benchmarks.
• All these tools support the DMAIC methodology for continuous improvement.

Statistical Process Control Fundamentals

Statistical Process Control (SPC) is a set of tools and techniques to systematically analyze process variations and distinguish between harmless random noise and signals of real problems. The core question is not whether variation exists — it always does — but whether we can make meaningful sense of it.

Two Types of Process Variation

All business processes exhibit variation. The critical task is to classify it:

Exam tip: The goal of SPC is to detect when assignable causes have shifted the process, not to eliminate common cause variation. A process affected only by common causes is said to be in statistical control.

---

Voice of the Customer – Specifications

Customer requirements define acceptable limits. These are expressed as:

• Target: the ideal value the customer desires (center of the specification).
• Upper Specification Limit (USL) and Lower Specification Limit (LSL): the maximum and minimum acceptable values.
• Tolerance: the allowable range, $\text{Tolerance} = \text{USL} - \text{LSL}$.

Example 1 – Service (Hotel checkout time) Target = 90 s, tolerance ±20 s → USL = 110 s, LSL = 70 s, tolerance = 40 s.

Example 2 – Manufacturing (Pen diameter) Target = 8 mm, tolerance ±0.5 mm → USL = 8.5 mm, LSL = 7.5 mm, tolerance = 1 mm.

---

Voice of the Process – Control Limits

The process itself has its own natural behavior, described by:

• Process average (center of process distribution).
• Upper Control Limit (UCL) and Lower Control Limit (LCL), set at ±3 standard deviations from the process average.
• Spread: $\text{UCL} - \text{LCL}$, the range of natural process variation.

Control limits are not specification limits – they reflect what the process actually does, not what the customer wants.

---

How to Measure Process Quality

Before launching an SPC program, decide what to measure and how. Two fundamental measurement approaches:

Choosing the measurement method determines which control chart to use later.

---

The Statistical Basis of Control Charts

The foundation is the normal distribution and the empirical rule:

$P(\mu - 3\sigma \leq X \leq \mu + 3\sigma) \approx 0.9973$

• Observations falling within ±3σ almost certainly (99.73 %) arise from common causes (random variation).
• An observation outside ±3σ is so unlikely under random variation that we suspect an assignable cause – the process may be out of control.

Hence control limits are set at $\mu \pm 3\sigma$:

$\text{UCL} = \mu + 3\sigma, \quad \text{LCL} = \mu - 3\sigma$

When all plotted points lie within the control limits, the process is said to be in a state of statistical control.

---

Types of Control Charts

---

Steps to Set Up a Control Chart

1. Choose the characteristic to measure (e.g., checkout time, pen diameter).
2. Select the measurement method (attribute or variable).
3. Choose the appropriate control chart (P, C, X‑bar+R).
4. Decide on a sampling plan – how many samples, how frequently.
5. Collect data and calculate control limits using statistical formulas.
6. Plot the data and analyze – are all points within limits? Any patterns?

A process with points outside the limits or with non‑random patterns signals the need for root‑cause investigation and corrective action.

---

Key Takeaways

• Two sources of variation: common causes (random, always present) and assignable causes (traceable, fixable).
• Specification limits (USL, LSL) reflect customer tolerance; control limits (UCL, LCL) reflect process capability.
• Measurement can be attribute‑based (good/bad count) or variable‑based (numerical values); choose based on cost and information needed.
• Control limits are set at $\mu \pm 3\sigma$; a point outside signals an assignable cause.
• The six‑step procedure: choose characteristic → choose measurement → choose chart → plan sampling → collect data & compute limits → plot & analyze.
• A process with only common causes is in statistical control; assignable causes indicate the process is out of control and needs investigation.

Variable Control Charts

Control charts monitor process stability over time. For variable data (continuous measurements like diameter in cm), the paired $\bar{X}$-chart (tracks the mean) and R-chart (tracks the range) are used. Their setup follows the same six-step methodology as attribute charts, differing only in parameter extraction.

Six-Step Setup (Variable Example: Cylinder Diameter)

1. Measurement characteristic – e.g., diameter of a cylindrical component (cm).
2. Measurement method – exact measurement (continuous).
3. Control chart type – $\bar{X}$-chart and R-chart.
4. Sampling plan – e.g., sample 5 consecutive pieces every 20 minutes; collect 15 such samples.
5. Collect data and establish control limits (detailed below).
6. Plot and analyse – check if all points fall within control limits.

Extracting Process Parameters from Sample Data

For each sample $i$ (size $n=5$), compute:

• $\bar{X}_i = \frac{1}{n}\sum_{j=1}^{n} X_{ij}$ (sample mean)
• $R_i = \max(X_i) - \min(X_i)$ (sample range)

Then the grand averages:

• $\bar{\bar{X}} = \frac{1}{k}\sum_{i=1}^{k} \bar{X}_i$ (overall mean)
• $\bar{R} = \frac{1}{k}\sum_{i=1}^{k} R_i$ (average range)

In the example, $\bar{\bar{X}} = 12.417$ and $\bar{R} = 0.119$.

Control Limits Using Standard Constants

Constants $A_2$, $D_3$, $D_4$ are read from a standard table based on sample size $n$. For $n=5$:

$\bar{X}$-chart limits $\text{UCL}_{\bar{X}} = \bar{\bar{X}} + A_2 \bar{R} = 12.417 + 0.577 \times 0.119 = 12.486$ $\text{LCL}_{\bar{X}} = \bar{\bar{X}} - A_2 \bar{R} = 12.417 - 0.577 \times 0.119 = 12.348$

R-chart limits $\text{UCL}_{R} = D_4 \bar{R} = 2.114 \times 0.119 = 0.252$ $\text{LCL}_{R} = D_3 \bar{R} = 0 \times 0.119 = 0$

All sample $\bar{X}_i$ and $R_i$ plotted within these limits → process is in statistical control.

Exam tip: The constants depend only on sample size $n$. Memorise common values (e.g., $n=5$: $A_2=0.577$, $D_3=0$, $D_4=2.114$) or know how to read the table.

---

Attribute Control Charts (P and C Charts)

When measurement is attribute (pass/fail, defect count), two common charts are used.

P-Chart (Proportion of Defects)

• Characteristic: proportion of defective items in a sample.
• Sampling plan: e.g., sample 100 pieces every 30 minutes; 12 samples.
• Data: count of defects per sample → proportion $p_i = \text{defects} / n$.
• Centre line: $\bar{p} = \frac{\sum p_i}{k}$ (e.g., $\bar{p}=0.105$).
• Standard deviation: $\sigma_p = \sqrt{\frac{\bar{p}(1-\bar{p})}{n}}$.
• Control limits (using $3\sigma$): $\text{UCL}_p = \bar{p} + 3\sigma_p,\quad \text{LCL}_p = \bar{p} - 3\sigma_p$

Example with $n=100$, $\bar{p}=0.105$: $\sigma_p = \sqrt{\frac{0.105 \times 0.895}{100}} \approx 0.031$ $\text{UCL}_p = 0.105 + 3 \times 0.031 = 0.198$ $\text{LCL}_p = 0.105 - 3 \times 0.031 = 0.012$ All plotted points inside → process in control.

C-Chart (Count of Defects per Unit)

• Use when: counting defects per constant unit (e.g., blemishes per $m^2$ of paint, thread runs per $m^2$ of cloth).
• Data: $c_i$ = number of defects in sample $i$; $m$ samples.
• Centre line: $\bar{c} = \frac{\sum c_i}{m}$.
• Standard deviation: $\sigma_c = \sqrt{\bar{c}}$.
• Control limits (3$\sigma$): $\text{UCL}_c = \bar{c} + 3\sqrt{\bar{c}},\quad \text{LCL}_c = \bar{c} - 3\sqrt{\bar{c}}$

The setup and interpretation mirror the P-chart; only the parameter formulae differ.

Worked C-Chart Example: Premium Fabric Rolls

A manufacturing plant producing premium fabric rolls tracks flaws such as broken threads, knots, holes, and uneven patterns. Each sampled roll is a constant inspection unit, so the quality team records the count of defects per roll and uses a C-chart to decide whether apparently high counts reflect random variation or a systematic problem.

• Sampling plan: $m=20$ rolls sampled at different times.
• Illustrative observations: roll 1 had $4$ defects, roll 6 had $3$, and roll 18 had $7$.
• Average count from all $20$ rolls: $\bar{c}=4.2$ defects per roll.

$\sigma_c = \sqrt{\bar{c}} = \sqrt{4.2} \approx 2.04$

$\text{UCL}_c = 4.2 + 3(2.04) \approx 10.34$

$\text{LCL}_c = \max\left(0,\;4.2 - 3(2.04)\right) = 0$

The centre line is $\bar{c}=4.2$. Although some rolls have $7$ defects, all plotted counts lie between $0$ and $10.34$ and show no systematic pattern. The process is therefore in statistical control: the observed variation is treated as inherent random variation, so production need not be stopped. A count above the UCL or a sustained non-random pattern—for example, counts repeatedly above $\bar{c}$—would instead require investigation and possibly a process halt.

---

Using Control Charts for Operational Decisions

Once control charts are established, two operational questions arise:

1. Process Out of Control? – Handling an Outlier

If a point falls outside the control limits (e.g., sample #6 in a C-chart):

• Step 1: Remove the outlier and recompute $\bar{\bar{X}}$, $\bar{R}$, limits.
• Step 2: Detailed root-cause investigation.
• Step 3: If no assignable cause → resume with revised chart. If cause found → implement countermeasures, then resume.
• Step 4: Re-establish limits, monitor to ensure control.

2. Early Detection of Impending Drift – Zone Rules

Even when all points are within limits, non-random patterns may signal a developing problem. Zones are defined relative to the centre line and $\pm 1\sigma$, $\pm 2\sigma$, $\pm 3\sigma$ lines:

Common rules to flag potential drift (stop and investigate):

All these rules share the logic: the pattern appears non-random, suggesting an assignable cause may be developing. Proactive investigation can avert a quality problem.

Exam tip: Always distinguish between “out of control” (points outside limits) and “unusual pattern” (zone rules). Both require action, but the rules allow earlier detection.

Key Takeaways

• Variable control uses $\bar{X}$-chart (mean) and R-chart (range) with constants $A_2$, $D_3$, $D_4$ from sample size $n$.
• Attribute control uses P-chart (proportion of defects) or C-chart (count of defects per unit); limits based on $\bar{p}$ or $\bar{c}$ and $3\sigma$.
• An outlier beyond limits demands removal, recomputation, and investigation for assignable causes.
• Zone rules (e.g., 9 points in a row, 6 increasing, 2 out of 3 in Zone A) detect non-random patterns before defects escalate.
• A process in statistical control can be used to predict future defect rates — a key managerial insight.

Process Capability and the Voice of the Customer

Once a process is statistically stable and in control (using control charts), the next step is to assess its performance against the voice of the customer – expressed as a target value, an upper specification limit (USL), and a lower specification limit (LSL). Superimposing these on the process’s own control limits (UCL, LCL) reveals whether the process can consistently meet customer requirements.

Spread vs. Centering: Two Dimensions of Process Capability

Even before comparing with specifications, the inherent capability of a process can be evaluated along two independent dimensions:

1. Spread (variation) – How wide the process distribution is.
2. Centering (location) – How close the process mean is to the target.

Consider two processes, A and B, both normally distributed and centered identically.

• Process A (blue): large spread, extends beyond USL and LSL → capable of producing defects.
• Process B (red): smaller spread, fits within specification limits → inherently better.

General rule: All else equal, smaller spread = more capable process.

Now consider two processes with identical spread but different centering.

• Process A (grey): centered exactly on the target → most output within specs.
• Process B (red): offset from the target → more output falls beyond USL or LSL → higher defect propensity.

General rule: All else equal, a process centered on the target is more capable than one that is off-centre.

Cp (Potential Capability)

The ratio of the specification range to the process's natural spread (6σ, representing ±3σ) measures the potential capability – what the process could achieve if perfectly centered.

$C_p = \frac{USL - LSL}{6\sigma}$

• Ignores centering – assumes process is positioned ideally.
• A higher $C_p$ indicates greater potential to meet specifications.

Cpk (Actual Capability)

Cpk adjusts $C_p$ for the offset (deviation) of the process mean $\mu$ from the target (or closer spec limit). It is the actual capability of the process as it currently runs.

$C_{pk} = \min\left( \frac{USL - \mu}{3\sigma}, \; \frac{\mu - LSL}{3\sigma} \right)$

• Captures both spread and centering.
• Used to predict defect rates and set improvement targets.

Cpk and Defect Rate

Exam tip: A Six Sigma organisation is defined as one that achieves Cpk = 2.0 – only 1.8 defects per billion opportunities (assuming a ±1.5σ shift). Memorise the Cpk-to-PPM relationship for Cpk = 1.0 and 2.0.

Practical Use of Cpk

• Monitor Cpk over time and set improvement targets (e.g., from 0.5 to 0.75 to 1.0).
• Supplier evaluation – require suppliers to report and maintain specified Cpk levels.
• Continuous improvement – reducing spread and improving centering pushes Cpk toward 2.0.

Key takeaways

• Process capability has two dimensions: spread (smaller = better) and centering (closer to target = better).
• $C_p$ measures potential; $C_{pk}$ measures actual capability accounting for off-centering.
• Cpk directly predicts defect rate (ppm).
• Six Sigma = Cpk = 2.0 = 1.8 defects per billion.
• Cpk is used for internal improvement targets and supplier performance monitoring.

---

Service Quality: Unique Challenges and the Gaps Model

Services are fundamentally different from products – they are performances, not objects. This makes quality assurance more complex.

Why Service Quality is Different

• Intangible & inseparable: Services cannot be counted, measured, inventoried, tested, or verified in advance.
• Heterogeneous: Vary across producers, consumers, and even day-to-day. Consistency of employee and customer behaviour is hard to ensure.
• Perishable & delivered in real time: Quality occurs during interaction with the customer. Consumer input is critical; managers have less real-time control.
• Difficult to recover: Failed services cannot be recalled like products. Poor service leads to customer exit.

Examples of poor service quality:

• Airline delay kept passengers in the dark.
• Weekend B-school programme far below participant expectations.
• E-tailers failed to deliver Christmas gifts on time (1999).

The Service Quality Gaps Model

Service quality is defined as the gap between perceived service and expected service (called Gap 5). The model proposes that Gap 5 can be explained by four other gaps (Gap 1–4).

Gap 1: Knowledge Gap – Management does not fully understand customer expectations.

• Cause: Lack of research, misinterpretation of customer needs.
• Fix: Better market research and customer listening.

Gap 2: Design Gap – Even if expectations are known, there is no adequate service design or standards.

• Cause: Absence of management commitment to quality, lack of know-how.
• Fix: Invest in service design, benchmarking, clear quality specifications.

Gap 3: Delivery Gap – The service performed does not meet the set specifications.

• Cause: Employee performance variability, poor training, inadequate procedures.
• Fix: Training, standard operating procedures, employee empowerment.

Gap 4: Communication Gap – What is promised in external communication differs from what is actually delivered.

• Cause: Overpromising in ads/brochures; neglecting to inform customers about invisible quality efforts.
• Fix: Realistic external communications; manage expectations transparently.

Gap 5: Perceived Service Quality – The overall service quality experienced by the customer.

• Result of gaps 1–4. Reducing those gaps narrows Gap 5.

Closing the Gaps

Instead of treating service quality as unmeasurable, the gaps model pinpoints where and why quality fails. Organisations can then invest in:

• Employee training (reduces Gap 3).
• Better service design and standards (reduces Gap 2).
• Realistic external communications, revised brochures (reduces Gap 4).
• Improved customer insight (reduces Gap 1).

Exam tip: Gap 5 = f(Gap 1, Gap 2, Gap 3, Gap 4). Any question on service quality measurement will likely revolve around identifying which gap is present in a given scenario. Be able to match the example to the gap.

Key takeaways

• Service quality is intangible, heterogeneous, and produced in real-time – different from product quality.
• The gaps model decomposes service quality (Gap 5) into four underlying gaps.
• Gap 1: not understanding expectations; Gap 2: no proper specs; Gap 3: poor delivery; Gap 4: overpromising.
• Closing gaps requires targeted investments in training, design, and communication.
• The model provides a diagnostic framework for improving service quality at the planning stage.

X Bar & R Chart

X-bar and R charts are paired control charts used for continuous (variables) data—like weight, length, or time—to monitor both the central tendency (mean) and the dispersion (range) of a process over time. Intuitively, the process might produce pouches that average exactly 500 g (mean is fine) while individual pouches swing wildly (variation is out of control). Monitoring only one chart can miss the other problem.

Why Use Both Charts?

If either chart shows an out-of-control signal, the process is out of control. A process can be “on target” but have excessive variability – or have tight variability but drift off target.

Computing Control Limits

Steps (illustrated with Shakti Foods data: 16 subgroups, each of size $n=5$):

1. Collect subgroups – e.g., 5 sugar pouches sampled periodically.
2. For each subgroup $i$, compute:
   • Sample mean: $\bar{x}_i$
   • Range: $R_i = \max - \min$
3. Grand mean and average range: $\bar{\bar{x}} = \frac{1}{k}\sum_{i=1}^{k}\bar{x}_i \quad,\quad \bar{R} = \frac{1}{k}\sum_{i=1}^{k} R_i$
4. Look up constants from standard control chart tables (based on $n=5$):

X-bar chart: $UCL_{\bar{x}} = \bar{\bar{x}} + A_2\bar{R}$ $LCL_{\bar{x}} = \bar{\bar{x}} - A_2\bar{R}$

R chart: $UCL_R = D_4\bar{R}$ $LCL_R = D_3\bar{R} = 0 \quad (\text{since } D_3=0 \text{ for } n\le6)$

Worked Example: Shakti Foods Sugar Pouches

• Specification: 500 g per pouch.
• Data collected: 16 subgroups of 5 pouches each. Subgroup weights ranged from approximately 488 g to 531 g.
• Calculated: $\bar{\bar{x}}$ and $\bar{R}$ from the 16 means and 16 ranges.
• Constants (as above): $A_2=0.577$, $D_4=2.114$, $D_3=0$.
• X-bar chart limits: $\bar{\bar{x}} \pm 0.577\bar{R}$.
• R chart limits: $UCL_R = 2.114\bar{R}$, $LCL_R = 0$.

Interpretation:

• All 16 sample means fell between the upper and lower control limits on the X-bar chart → mean appears in control (random variation only).
• However, one subgroup range exceeded $UCL_R$ on the R chart → dispersion is out of control (a special cause has increased variability).

How to read the combined picture:

In the Shakti example: the mean is stable but variation is not → overall process is out of control. The quality engineer should pause and look for an assignable cause (e.g., worn filling nozzle, inconsistent raw material) that made one subgroup’s range spike.

Exam tip: A common mistake is to declare a process “in control” based only on the X-bar chart. Always review both charts. The X-bar chart can look fine while the R chart reveals a problem – or vice versa. Both must be in control for the process to be stable.

Key Takeaways

• X-bar chart monitors the process mean; R chart monitors the process dispersion.
• Both charts are needed for continuous data to get a complete stability picture.
• Control limits depend on subgroup size $n$ (constants $A_2, D_3, D_4$ from standard tables).
• A point above $UCL$ or below $LCL$ on either chart signals an out-of-control process.
• Shakti Foods example: means in control, but one large range → dispersion out of control.

Capability Problem

Process capability quantifies how well a process can consistently produce output that meets customer specifications. Intuitively: given the natural variation of a process (its "voice"), can it fit within the tolerance window set by the customer (the "voice of the customer")? Two indices measure this fit: Cp (for mean-centered processes) and Cpk (for any process, penalizing off-center means).

Definitions and Formulas

• Voice of the customer: the acceptable range $USL - LSL$ (Upper Specification Limit – Lower Specification Limit).
• Voice of the process: the natural spread of the process, assumed to be $\pm 3\sigma$ from the mean, i.e., $6\sigma$.

Cp – Process capability for a perfectly mean-centered process where the process mean $\mu_p$ equals the specification target $\mu_s$: $C_p = \frac{USL - LSL}{6\sigma}$ A value $<1$ indicates the process spread exceeds the tolerance; $>1$ indicates room to spare.

Cpk – Conservative capability index used when the process mean is not centered (or in all cases, as it is always ≤ Cp): $C_{pk} = \min \left( \frac{USL - \mu_p}{3\sigma},\ \frac{\mu_p - LSL}{3\sigma} \right)$ Cpk penalizes shift: even if the spread is small, an off-center mean reduces capability.

Decision Logic

Worked Example: Shakti Foods (Original Process)

Data

• Process: filling raw sugar pouches
• Target: 0.5 kg per pouch
• Process parameters: $\mu_p = 0.5\ \mathrm{kg}$, $\sigma = 0.006\ \mathrm{kg}$
• Customer specifications: $USL = 0.51\ \mathrm{kg}$, $LSL = 0.49\ \mathrm{kg}$
• Specification mean: $\mu_s = (0.51 + 0.49)/2 = 0.5\ \mathrm{kg}$

Step 1: Check centering – $\mu_p = \mu_s$ → process is mean-centered → use $C_p$.

Step 2: Compute $C_p$

$$C_p = \frac{0.51 - 0.49}{6 \times 0.006} = \frac{0.02}{0.036} \approx 0.56$$

Interpretation: $C_p = 0.56 < 1$ – the process spread (6σ = 0.036 kg) is wider than the tolerance (0.02 kg). A high defect rate is expected even when centered.

Changed Process: Mean Drift

After some weeks, the process mean shifts to $\mu_p' = 0.505\ \mathrm{kg}$ (σ unchanged at 0.006 kg). Specifications remain the same.

Check centering – $\mu_p' \neq \mu_s$ → use $C_{pk}$.

Compute $C_{pk}$

$$\frac{USL - \mu_p'}{3\sigma} = \frac{0.51 - 0.505}{3 \times 0.006} = \frac{0.005}{0.018} \approx 0.278$$ $$\frac{\mu_p' - LSL}{3\sigma} = \frac{0.505 - 0.49}{3 \times 0.006} = \frac{0.015}{0.018} \approx 0.833$$ $$C_{pk} = \min(0.278,\ 0.833) = 0.278$$

Capability dropped from 0.56 to 0.278. The shift increases the tail probability above the USL dramatically, more than offsetting the reduction below LSL.

Exam tip: $C_{pk} \leq C_p$ always. A large drop from $C_p$ to $C_{pk}$ signals mean shift. Here $C_p$ (if computed for the changed process ignoring shift) would still be 0.56, but $C_{pk}$ reveals the true performance.

Improvement Strategies

Two levers exist to restore (or improve) capability:

1. Recenter the process – bring $\mu_p$ back to 0.5 kg. This restores $C_p = 0.56$.
2. Reduce variability – if recentering is impossible (e.g., engineering constraint), shrink $\sigma$ so that even with the shifted mean, the tails fall within specifications. For example, if $\sigma$ were halved to 0.003 kg, $C_{pk}$ would double.

Combining both actions yields the greatest improvement.

Key Takeaways

• $C_p$ measures potential capability when the process is perfectly centered; $C_{pk}$ measures actual capability, penalizing off-center means.
• $C_p = (USL - LSL) / 6\sigma$; $C_{pk} = \min(\frac{USL - \mu}{3\sigma},\ \frac{\mu - LSL}{3\sigma})$.
• A capability index below 1 implies the process cannot consistently meet specs without defects.
• Mean drift reduces $C_{pk}$ even if spread remains the same.
• Improvement: (1) recenter the mean, (2) reduce variation, or both.