Engineer Quality Management cheat sheet

ISO 9001:2015 is built on seven quality management principles: customer focus, leadership, engagement of people, process approach, improvement, evidence-based decision making, and relationship management.

ISO 9001:2015 has 10 clauses: 1 Scope, 2 Normative references, 3 Terms and definitions, 4 Context of the organization, 5 Leadership, 6 Planning, 7 Support, 8 Operation, 9 Performance evaluation, 10 Improvement.

Clauses 4-10 contain the auditable requirements; clauses 1-3 are informational and cannot be used for nonconformity findings.

The process approach requires identifying process inputs, outputs, sequence, interaction, owners, and performance indicators; the PDCA cycle and risk-based thinking are applied to each process.

Documented information replaces the old terms 'documents' and 'records': retain records as evidence of results, maintain documents as instructions or rules.

ISO 9001:2015 clause 6.1 requires planning actions to address risks and opportunities; there is no separate formal risk-management standard required, and no requirement for a documented risk register.

ISO 9001 certification is issued by an accredited certification body (third party); internal audits are first party and supplier audits are second party.

ISO 19011 provides guidance for auditing management systems: principles include integrity, fair presentation, due professional care, confidentiality, independence, and an evidence-based approach.

ISO 9000:2015 contains the fundamental terms and definitions; ISO 9004 gives guidance for sustained success and is not a certification standard.

Hoshin kanri (policy deployment) cascades annual objectives from top management to operational levels; the catchball step negotiates targets and means between levels before commitment.

A X-matrix links long-term objectives, annual objectives, improvement priorities, and targets with correlation symbols, giving one-page traceability of strategy to action.

QFD translates customer requirements (the 'whats') into technical requirements (the 'hows') using a relationship matrix; the roof matrix records interactions among technical requirements.

In QFD, importance ratings are computed as the sum over customer requirements of (customer importance x relationship weight); relationship weights are commonly 9 strong, 3 medium, 1 weak.

The house of quality sequence is: customer requirements, competitive assessment, technical requirements, relationship matrix, technical targets, and correlation roof.

A quality plan per ISO 10005 defines the specific procedures, resources, responsibilities, and timing for a product, project, or contract.

APQP is a five-phase product quality planning structure: plan and define, product design and development, process design and development, product and process validation, feedback and corrective action.

Strategic deployment must be measurable: each objective needs a metric, baseline, target, owner, and review cadence; unassigned targets are the most common deployment failure point.

Cost of poor quality splits into internal failure, external failure, appraisal, and prevention; prevention spending typically reduces total quality cost more than appraisal spending.

Control chart selection: variables data uses X-bar/R (n=2-9) or X-bar/S (n>=10); individuals data uses I-MR; attributes use p, np, c, or u charts.

p chart = proportion defective, variable subgroup size; np chart = number defective, constant subgroup size; c chart = count of defects per unit, constant opportunity; u chart = defects per unit, variable opportunity.

X-bar chart limits use A2 x R-bar; R chart limits use D3 x R-bar and D4 x R-bar; constants depend on subgroup size n.

Western Electric rules: 1 point beyond 3 sigma; 2 of 3 consecutive points beyond 2 sigma; 4 of 5 beyond 1 sigma; 8 consecutive points on one side of the centerline.

Cp = (USL - LSL) / (6 sigma); Cpk = min[(USL - mean), (mean - LSL)] / (3 sigma). Cpk accounts for centering, Cp does not.

Pp and Ppk use the overall sample standard deviation; Cp and Cpk use within-subgroup sigma estimated from average range or average standard deviation.

Process capability requires a stable process; calculating Cpk on an out-of-control chart produces a meaningless index.

A Cpk of 1.33 corresponds to roughly 63 defects per million opportunities for a centered normal process; 1.67 corresponds to about 0.6 ppm.

Attribute capability uses defect rate p and the binomial or Poisson distribution; a Poisson model is valid when defects per unit are rare and independent.

GR&R studies decompose variation into repeatability (equipment variation) and reproducibility (appraiser variation); total gauge variation combines both.

Percent GR&R = 100 x (GR&R standard deviation / total study variation) or as a percentage of tolerance; AIAG guidance: under 10% acceptable, 10-30% conditionally acceptable, over 30% unacceptable.

Number of distinct categories (ndc) = 1.41 x (process variation / gauge variation); ndc of 5 or more is generally required to use a gauge for process control.

Bias is the difference between the average measured value and the reference value; linearity is the change in bias across the operating range.

Stability is the change in bias over time, evaluated with a control chart of repeated measurements on a reference standard.

Resolution should be at most one tenth of the tolerance or one tenth of the process variation to avoid rounding-induced capability errors.

Attribute agreement analysis reports within-appraiser agreement and agreement versus a known standard; kappa adjusts for chance agreement.

Calibration establishes the relationship between instrument readings and a reference traceable to national or international standards; verification checks that calibration remains valid.

Measurement uncertainty is reported as an interval about the result, typically using a coverage factor k=2 for approximately 95% confidence.

DOE objective hierarchy: screening (many factors, few runs), characterization (main effects and interactions), optimization (response surface), and robust design (tolerance to noise).

Full factorial with k factors at 2 levels requires 2^k runs; a fractional factorial 2^(k-p) uses a resolution equal to the smallest sum of generator words.

Resolution III designs alias main effects with two-factor interactions; Resolution IV alias two-factor interactions with each other; Resolution V alias main effects with four-factor interactions, so two-factor interactions are clear.

Randomization protects against nuisance factors; blocking removes known sources of variability; replication provides an independent estimate of pure error.

ANOVA in DOE partitions total sum of squares into model and residual; a factor is significant when its p-value is below alpha and its effect exceeds practical importance.

Taguchi methods use orthogonal arrays, signal-to-noise ratios, and loss functions; smaller-the-better, larger-the-better, and nominal-the-best SN ratios are the three standard forms.

The quality loss function L = k(y - T)^2 shows loss grows quadratically with deviation from target, so conformance to specification alone understates loss.

FMEA uses severity, occurrence, and detection ratings (typically 1-10) with RPN = S x O x D; action priority should consider severity even when RPN is moderate.

Poka-yoke prevents errors at the source through physical or design constraints rather than relying on operator vigilance.

Correction fixes the symptom; corrective action removes the cause of a detected nonconformity; preventive action removes the cause of a potential nonconformity.

The 5 Whys must follow the physical or causal chain, not opinions; each 'why' answer becomes the subject of the next question until the systemic cause is reached.

Fishbone (Ishikawa) categories in manufacturing are commonly Man, Machine, Material, Method, Measurement, and Environment (6M).

Fault tree analysis works top-down from an undesired event using AND/OR gates; FMEA works bottom-up from failure modes.

A Pareto chart identifies the vital few contributors; root cause analysis should start with the largest contributor unless safety or regulatory risk dictates otherwise.

A good problem statement includes what, where, when, how much, and the gap versus standard, and contains no cause or solution language.

Verification confirms the implemented action was done as planned; validation confirms the action actually eliminated the nonconformity or prevented recurrence.

Effectiveness checks require objective evidence over a defined period; closing a CAPA without an effectiveness check leaves recurrence risk unaddressed.

Is/is-not analysis narrows the problem by contrasting where the defect occurs versus where it does not, revealing distinguishing factors.

Supplier selection criteria typically include capability (Cpk), quality system certification, capacity, financial stability, and past performance.

Incoming inspection sampling plans follow ANSI/ASQ Z1.4 (attributes) or Z1.9 (variables); the plan is defined by lot size, inspection level, and AQL.

In the OC curve, producer's risk (alpha) is the probability of rejecting a good lot; consumer's risk (beta) is the probability of accepting a bad lot.

AQL is the quality level that corresponds to a specified low probability of acceptance; it is not a defect target for the supplier.

Skip-lot inspection reduces inspection frequency after a defined number of consecutive accepted lots; it must be suspended immediately after a rejected lot.

A supplier quality agreement defines specifications, change notification, traceability, audit rights, and nonconformity response times.

PPAP requires submission of design records, engineering change documents, process flow, PFMEA, control plan, MSA, dimensional results, material tests, and a part submission warrant.

First article inspection verifies the first production parts against all drawing characteristics before full production release.

Supplier scorecards combine quality (PPM, nonconformities), delivery, and responsiveness; scores should drive escalation or development actions.

Audit types: first-party (internal), second-party (supplier), third-party (certification or regulatory); each has different independence requirements.

An audit finding must cite the requirement, the objective evidence, and the gap; opinions or undocumented observations are not nonconformities.

Nonconformity grading commonly distinguishes major (system failure or risk of escape) from minor (isolated lapse); repeat findings are usually escalated.

The audit process follows ISO 19011: objectives, scope, criteria, team selection, document review, opening meeting, evidence collection, findings, closing meeting, report, and follow-up.

Sampling in audits must be justified by risk and process volume; convenience sampling alone is weak evidence of system effectiveness.

Inspection can be receiving, in-process, or final; inspection alone cannot create quality and should be reduced as process capability improves.

Gauge calibration records, control plans, and traceability records are the highest-frequency audit evidence requests.

Regulatory compliance differs from voluntary certification: noncompliance can carry legal penalties, while ISO 9001 certification is contractual and market driven.

Auditor competence includes sector knowledge, audit technique, and communication; impartiality requires excluding auditors from auditing their own work.

DMAIC phases: Define, Measure, Analyze, Improve, Control; DMADV (Design for Six Sigma) applies to new processes and products.

Lean waste categories (TIMWOODS): Transport, Inventory, Motion, Waiting, Overproduction, Overprocessing, Defects, and Skills underutilization.

Value stream mapping distinguishes value-added time from lead time; the ratio of value-added time to total lead time is the process cycle efficiency.

Kaizen is continuous incremental improvement by the people doing the work; kaizen events are time-boxed and target a specific process area.

5S steps: Sort, Set in order, Shine, Standardize, Sustain; the fifth step is the one most often missing and determines durability.

OEE = Availability x Performance x Quality; it measures equipment effectiveness against ideal cycle time and is not a substitute for capacity planning.

Takt time = available production time / customer demand; it sets the pace, not the actual cycle time.

Kanban limits work in process and pulls material only when downstream consumption signals demand.

A control plan in the Control phase specifies the metric, sample size, frequency, owner, and reaction plan for each key process characteristic.

Six Sigma corresponds to 3.4 defects per million opportunities when the process mean shifts 1.5 sigma over time.

A SIPOC diagram defines Suppliers, Inputs, Process, Outputs, and Customers at the Define phase boundary; it is a scope tool, not a detailed process map.

Hypothesis testing in Analyze uses a p-value compared to alpha; failing to reject the null is not proof that the null is true.

Measurement system validation must precede capability analysis in the Measure phase; a poor gauge invalidates both baseline and improvement claims.

Standard work documents the current best method with cycle time, work sequence, and standard inventory, and is the baseline for further improvement.

Poka-yoke and jidoka (stop at defect) support built-in quality rather than inspection-based quality.

A3 problem solving captures background, current condition, target, root cause, countermeasures, and follow-up on one page to force concise thinking.

ISO 9001:2015 clause 9.2 requires internal audits at planned intervals considering process importance, changes, and prior audit results.

Clause 8.5.1 requires controlled conditions for production, including availability of work instructions and monitoring equipment.

Clause 10.2 requires retaining documented information on the nature of nonconformities, actions taken, and results of corrective action.

Clause 7.5 requires documented information to be controlled for availability, protection, and retention; version control is the most common audit finding.

The certification cycle is a three-year cycle with annual surveillance audits and a recertification audit before expiry.

A nonconformity cited in a certification audit must be corrected within a defined period, and the corrective action must address root cause, not only the finding.

The control plan is a planning deliverable that lists product and process characteristics, specifications, measurement methods, sample sizes, frequencies, and reaction plans.

A feasibility commitment in APQP records whether the supplier can meet capacity, tolerance, and timing requirements before design release.

A design review is a planned, documented examination of design outputs against requirements at defined stages; it is not a sign-off formality.

Quality function deployment should be re-run when customer requirements change materially; a stale matrix misdirects resources.

Balanced scorecard links quality, delivery, cost, and people metrics so that quality objectives are not traded away for output.

Annual quality objectives must be consistent with the quality policy and measurable; clause 6.2 requires planning who, what, resources, timing, and evaluation.

A process in statistical control has variation from common causes only; special-cause variation appears as a signal on a control chart.

Control limits are computed from process data and are not specification limits; specification limits come from design requirements.

Rational subgrouping minimizes variation within a subgroup and maximizes variation between subgroups so that shifts are detected.

The average run length (ARL) describes how many points are plotted before a signal; ARL0 applies to a stable process, ARL1 to a shifted process.

For a centered process, Cpk equals Cp; any off-centering makes Cpk smaller than Cp.

A Cpk below 1.0 means the process produces nonconforming output even when stable; process improvement is required, not tighter inspection.

A Type 1 study evaluates repeatability of one gauge against a reference standard with one appraiser.

A linearity study regresses bias against reference values; a significant slope indicates the gauge reads differently across its range.

Destructive testing uses nested GR&R designs because parts cannot be measured repeatedly by the same appraiser.

Calibration intervals should be set from drift history and criticality, not copied from a default calendar.

Traceability chain: working standard, reference standard, national standard, and SI definition; each link must be documented.

Attribute gauges require agreement studies because GR&R in the variables sense does not apply to pass/fail decisions.

A screening design should not be used to estimate interactions unless resolution permits; follow with a characterization design.

Center points in a 2-level design test for curvature; a significant curvature term means the response is not linear and a response surface design is needed.

Confounding means effects cannot be separated; aliasing structure is read from the design generators.

Robust design places the design point where the response is least sensitive to noise factors, not necessarily at the response maximum.

A confirmation run at the predicted optimum is required before adopting DOE results into production.

Sample size in DOE depends on the effect size to detect, the noise standard deviation, alpha, and power; underpowered experiments produce false negatives.

Human error is usually a symptom; look for the process conditions that made the error possible or undetectable.

Containment actions protect the customer immediately and are temporary; they must be replaced by permanent corrective action.

A CAPA record should link the nonconformity, root cause, action, owner, due date, verification, and effectiveness evidence.

Recurrence of the same nonconformity indicates the root cause was not addressed or the effectiveness check was inadequate.

Brainstorming and cause-and-effect diagrams generate candidate causes; data analysis is required to confirm which cause is real.

Change control must be applied to corrective actions that alter a validated process, drawing, or specification.

Critical characteristics should be identified on the drawing and control plan and verified at receiving when supplier capability is unproven.

A supplier corrective action request (SCAR) requires containment, root cause, corrective action, and verification of effectiveness from the supplier.

Incoming inspection records must include lot identity, sample size, acceptance criteria, results, and disposition.

Supplier development differs from supplier selection: development raises capability through training, audits, and joint improvement projects.

A certificate of conformance is a supplier declaration, not independent evidence; it should be supplemented by periodic verification.

Approved supplier list maintenance requires periodic re-evaluation based on performance data, not only initial qualification.

The audit program must consider risk and prior results when setting frequency and scope.

Audit evidence includes records, interviews, and observation; interviews alone are weaker than records plus observation.

The closing meeting presents findings and agrees on corrective action timing; the audit report is issued afterward.

A surveillance audit focuses on changes, previous findings, and continued conformity rather than a full system review.

Inspection planning should define the characteristic, method, sample, acceptance criterion, and reaction to nonconformance before inspection begins.

Compliance to customer-specific requirements is audited separately from ISO 9001 and often carries stricter evidence expectations.

A root cause verified in Analyze must be linked to a countermeasure in Improve; unrelated improvements break the DMAIC logic.

The Control phase must include a response plan for out-of-control signals, otherwise gains decay after the project closes.

Statistical significance does not guarantee practical significance; effect size and cost determine whether a change is worth implementing.

A pilot run validates the improvement before full deployment and exposes unintended side effects.

Project charters define the problem, goal, scope, team, and business case, and prevent scope creep during DMAIC.

Standardizing the improved method and retraining are required to sustain gains after the project team disbands.

Risk-based thinking in ISO 9001 does not mandate a formal risk methodology; the organization defines its approach and retains evidence of planning.

Outsourced processes must be controlled and their conformity ensured; the control method depends on the risk the process presents.

The quality policy must be appropriate to purpose, include commitment to requirements and improvement, and be communicated and available.

Top management must demonstrate leadership by ensuring resources and integrating quality requirements into business processes, not only by delegating to a quality manager.

Clause 8.7 requires control of nonconforming outputs with documented information on the nonconformity, actions, concessions, and identification of the authority deciding the action.

A concession is permission to use or release nonconforming product under specified limits, usually with customer agreement and limited duration.

Deployment matrices translate top-level objectives into department and team targets; gaps between rows indicate unowned objectives.

A quality plan should reference the applicable control plan, inspection plan, and acceptance criteria rather than restating them.

Design for manufacturability reviews reduce downstream variation by aligning tolerances with process capability during design.

Cost of quality reporting should separate prevention, appraisal, internal failure, and external failure so improvement priorities are visible.

Strategic quality objectives should be reviewed at management review with trend data, not only pass/fail status.

A project quality plan for a contract defines deliverables, acceptance criteria, and inspection points agreed with the customer.

Out-of-control signals must be investigated for assignable cause before any capability index is computed.

The moving range chart for individuals data uses the average moving range of consecutive points; limits use E2 and D4 constants.

A p chart with a small average number of defectives per subgroup (np under about 5) is unreliable; use a c or u chart for counts instead.

Process capability studies require at least 25 subgroups or 100 individual points to estimate sigma reliably.

Specification limits placed inside control limits indicate the process cannot meet requirements even when stable.

Capability improvement should target centering first when Cpk is much lower than Cp, then variation reduction.

GR&R as a percentage of tolerance is the correct metric when the gauge is used for acceptance decisions against specifications.

Reproducibility is influenced by appraiser technique and training; repeatability is influenced by the gauge, fixture, and part variation.

A gauge that passes GR&R but has poor stability over time will produce drifting capability results.

Calibration certificates should state the reference standard, uncertainty, environmental conditions, and the as-found and as-left readings.

Out-of-tolerance calibration results require an impact assessment on product measured since the last valid calibration.

Resolution finer than needed increases cost without improving capability; resolution coarser than one tenth of tolerance introduces measurement error into decisions.

Interaction plots show whether the effect of one factor depends on the level of another; parallel lines indicate no interaction.

Main effects plots show the average response at each factor level, ignoring interactions, and can mislead when interactions are strong.

A response surface design such as central composite or Box-Behnken estimates quadratic effects and locates an optimum.

Blocking should be randomized within blocks, and block effects should be tested before interpreting factor effects.

The power of a DOE increases with the number of replicates and decreases with noise; pilot runs estimate noise before the full experiment.

Factor levels should span the actual operating range; levels too close together produce effects indistinguishable from noise.

The 8D method proceeds through team formation, problem description, containment, root cause, corrective action selection, implementation, prevention, and recognition.

A root cause must be both necessary and sufficient to explain the observed nonconformity; a plausible cause that cannot reproduce the effect is not confirmed.

Detection failures are causes too: a defect that escapes inspection points to a control-plan or gauge weakness.

Corrective actions should be prioritized by risk reduction, not by ease of implementation alone.

Preventive action can be triggered by trend analysis, risk assessment, or FMEA, not only by an actual nonconformity.

Lessons learned should be fed back into FMEA, control plans, and standard work to prevent recurrence across similar processes.

Supplier audits should focus on the processes and characteristics that affect the supplied product, not only on documentation completeness.

Incoming sampling plans must state the lot definition; mixed lots invalidate the sampling assumption.

A rejected lot requires documented disposition: return, rework, sort, or concession with customer approval.

Supplier process capability data should be verified by the supplier's own MSA before being accepted as evidence.

Change notification requirements should cover process, material, sub-supplier, and facility changes.

Long-term supplier agreements should include quality targets, escalation paths, and audit access.

Audit criteria are the set of requirements used as a reference; findings must be traceable to a specific criterion.

Opening meetings confirm scope, criteria, schedule, and confidentiality, and give the auditee a chance to raise constraints.

Audit trails should be sampled from records back to the process, not only from the process forward to records.

Findings should be written so the auditee can verify the evidence independently; vague findings cannot be closed effectively.

Follow-up audits verify that corrective actions were implemented and effective, not merely planned.

Auditor independence is compromised if the auditor has operational responsibility for the area being audited.

The Define phase should quantify the business case and the gap between current and target performance.

The Measure phase establishes a validated baseline; without it, improvement claims cannot be substantiated.

The Analyze phase should confirm causes with data rather than accepting the team's first hypothesis.

The Improve phase should pilot and verify the effect with the same measurement system used for the baseline.

The Control phase transfers ownership to the process owner with a documented monitoring plan.

Lean and Six Sigma complement each other: Lean removes non-value-added flow waste, Six Sigma reduces variation in value-added steps.

Management review inputs include audit results, nonconformities, performance trends, resource adequacy, and risks; outputs include improvement and resource decisions.

Competence requirements in clause 7.2 require determining necessary competence, ensuring it, and retaining evidence.

Customer property must be identified, protected, and safeguarded, with records of loss or damage reported to the customer.

Monitoring and measuring resources must be suitable and maintained; when found unfit, the organization must assess the validity of prior results.

Documented information from external providers must be controlled to the extent necessary for conformity.

The process approach requires performance indicators for processes, not only for final product.

Hoshin kanri reviews use monthly and quarterly checkpoints to adjust means while keeping annual targets stable.

A strategy map links learning and growth, process, customer, and financial perspectives so quality improvements connect to business outcomes.

Quality planning should identify special characteristics early, since late discovery forces costly design or process changes.

Deployment without resource allocation is a plan, not a commitment; budgets and staffing must follow the objectives.

Risk assessment during planning should consider both threats and opportunities, with actions proportional to impact.

Customer requirement reviews before acceptance of orders prevent later nonconformity and disputes.

A control chart distinguishes common-cause from special-cause variation; reacting to common-cause noise increases variation.

Control limit recalculation is justified only after a documented permanent process change.

A process can be in control and still incapable; control and capability answer different questions.

The Cpk target should be set by criticality: safety-critical characteristics often require 1.67 or higher.

Attribute control charts are less sensitive than variables charts and need larger samples to detect shifts.

Capability studies must use a measurement system with acceptable GR&R, or the index reflects gauge error.

GR&R study design: 10 parts spanning the process range, 3 appraisers, 3 trials, randomized order.

The range method of GR&R is quick but cannot separate repeatability from reproducibility.

ANOVA GR&R can detect appraiser-by-part interaction, which the average and range method cannot.

A gauge used only for sorting may tolerate higher GR&R than one used for process control.

Measurement error adds to observed variation; observed capability is worse than true capability when the gauge is poor.

Calibration traceability must be documented for every gauge used to accept product.

A 2^(k-p) design has p generators; the defining relation determines the alias structure and resolution.

Resolution IV designs are appropriate for screening when two-factor interactions may exist but main effects must stay clear.

Folding a Resolution III design over creates a Resolution IV design that separates main effects from two-factor interactions.

Randomization order must be recorded and followed; reordering runs invalidates the statistical assumptions.

Response surface methodology uses steepest ascent to move toward the optimum, then a quadratic model to locate it.

Tolerance design sets component tolerances from the loss function and the sensitivity of the response to each component.

A problem description should include the defect, location, time pattern, quantity, and the standard it violates.

Cause verification can use designed experiments, stratified data, or reproduction of the defect under suspected conditions.

Corrective action should address the systemic cause, such as a missing control or unclear standard, not just the individual error.

Effectiveness verification should run long enough to cover the conditions under which the problem originally occurred.

CAPA metrics such as open age, recurrence rate, and overdue actions indicate system health.

A corrective action that creates a new risk requires risk assessment before implementation.

Supplier qualification should include a capability assessment, not only a certificate review.

Incoming inspection levels should be adjusted based on supplier performance history and criticality.

A first article inspection report should cover every drawing characteristic with measured values, not pass/fail only.

Supplier nonconformity data should feed back into the approved supplier list and sourcing decisions.

Incoming material must be identified and segregated until inspection status is known.

Supplier process changes require revalidation of PPAP elements affected by the change.

Audit sampling should cover all shifts, lines, and shifts of staffing where the process operates.

Records must be legible, retrievable, and retained for the defined period; missing records are a common major finding.

Inspection results must be recorded against the characteristic and the acceptance criterion used.

Compliance audits against regulations require knowledge of the applicable legal requirements, not only the management system standard.

Findings should distinguish system nonconformity from product nonconformity because corrective action paths differ.

Audit reports should be distributed to those responsible for corrective action and tracked to closure.

Value stream maps should distinguish current state and future state with a dated implementation plan.

Setup reduction (SMED) separates internal from external setup activities to cut changeover time.

Total productive maintenance builds operator ownership of equipment care to prevent unplanned downtime.

Mistake-proofing should be designed into the process, not added as an inspection step at the end.

A kaizen event should have a defined scope, measurable target, and follow-up audit after the event.

Improvement projects should be prioritized by impact on customer, cost, and risk, not by ease alone.