EXIN LSSA Green Belt Practical: Lean Six Sigma Fundamentals and Project Organisation

This domain covers the foundations the exam expects you to apply rather than recite: the difference between Lean and Six Sigma, how Y = f(x) frames a project, where DMAIC stops and design starts, and how belts, sponsors and tollgates keep a project alive.

Lean and Six Sigma solve different problems

Lean attacks waste and flow: value is defined by the customer; the classic eight wastes are transport, inventory, motion, waiting, overproduction, overprocessing, defects and unused skills (TIMWOODS). Typical tools: value stream mapping, takt time, pull/kanban, 5S, SMED, standard work.
Six Sigma attacks variation: output Y is a function of process inputs X (Y = f(x)); the method is DMAIC. The 1.5-sigma shift convention converts sigma level to long-term DPMO: 6 sigma = 3.4 DPMO, 4 sigma = 6,210, 3 sigma = 66,807.

Roles and governance

Champion/sponsor: owns the business case, resources and barrier removal; approves charter and tollgates.
Master Black Belt: technical coach, cross-project deployment, training.
Black Belt: full-time, complex cross-functional projects. Green Belt: part-time, leads a project in the own work area alongside the day job.
Process owner: receives the improved process at closure and sustains the control plan.
Tollgates: phase-end reviews where deliverables (charter, data collection plan, verified causes, implemented solution, control plan) are accepted or the project is returned for more work.

Worked example

An invoicing process produces 60 defects on 20,000 invoices. DPMO = 60 ÷ 20,000 × 1,000,000 = 3,000. From the sigma table, 3,000 DPMO lies between 4.0 sigma (6,210) and 4.5 sigma (1,350), so roughly 4.2 sigma short-term. Reasoning: convert to a common normalised metric first, then read the table; raw defect counts cannot be compared across processes of different volume.

Traps that cost marks

Quoting "3.4 defects per million" as a percentage, or mixing defects (counted per unit) with defective units.
Prescribing DMADV for fixing an existing process, or DMAIC for a product that does not exist yet.
Selecting a project with no measurable Y and no link to cost of poor quality or strategy.
Treating the Green Belt as the sponsor: the belt leads the work, the champion clears organisational barriers.

Define Phase: Charter, VOC and Scope

This domain tests whether you can convert a vague complaint into a chartered, measurable project: voice of the customer into CTQs, SIPOC boundaries, a fact-based problem statement, a SMART goal and an agreed scope, all signed off at the Define tollgate.

The charter is the project contract

Business case: why now, linked to strategy and cost of poor quality.
Problem statement: metric, baseline, time frame, location and gap versus requirement, with no cause and no solution inside it.
Goal statement: SMART, with direction (reduce/increase), baseline, target and deadline.
Scope: in/out, process start and stop points, excluded sites, shifts or products.
Team and milestones, sponsor, process owner, and expected benefit validated with finance.

Voice of the customer

Collect VOC from complaints, interviews, surveys, warranty and service data; translate into CTQs with operational specifications using a CTQ tree (need → driver → measurable requirement).
Kano separates must-be, performance and delighter requirements; must-be failures drive dissatisfaction regardless of delighters.
SIPOC (suppliers, inputs, process, outputs, customers) fixes boundaries and shows where Y is measured. It is a high-level map, not a detailed flowchart.

Worked example

Original statement: "Our service desk is too slow because we are understaffed; improve customer satisfaction." Rewrite: "Between January and March 2024 the mean speed of answer on the retail line was 96 s against a customer CTQ of ≤30 s, and 18% of calls were abandoned. Target: mean speed of answer ≤30 s and abandon rate ≤5% by 31 December 2024." Reasoning: the rewrite carries metric, baseline, time frame, requirement and target; "understaffed" is a suspected cause and belongs in Analyse, not in the charter.

Traps

Problem statements containing a cause ("due to poor training") or a solution ("install a new system").
Goals such as "improve quality by 30%" with no baseline or operational definition.
VOC taken from one loud customer instead of a segmented, verified sample.
Scope so wide the project cannot close, or so narrow that the dominant variation sits outside it.
Skipping the tollgate: an unsigned charter is a common reason projects drift later.

Measure Phase: Data, MSA and Baseline Capability

This domain tests whether you can build a trustworthy baseline: operational definitions, a data collection plan, a measurement system you can defend, and baseline capability stated in the right metric.

Before any data is collected

Operational definition: exactly what counts as a defect, how it is counted, by whom, over what period. Ambiguity here invalidates everything downstream.
Data types: continuous (time, length, cost) versus discrete/attribute (pass-fail, counts, categories). Attribute data needs far more samples for the same confidence.
Data collection plan: metric, definition, type, source, sampling method, sample size, frequency, owner and data-checking approach.
Sampling: random where possible; stratified to cover shifts, lines and segments; rational subgroups for control charts (units made under the same conditions); avoid convenience samples from a single shift.

Measurement system analysis

GR&R for continuous data: %study variation below 10% acceptable, 10–30% conditionally acceptable depending on cost and risk, above 30% unacceptable; number of distinct categories ≥5 is desirable.
Attribute agreement / kappa for pass-fail judgement: check repeatability (same appraiser) and reproducibility (between appraisers) against a known standard.

Baseline capability

Continuous: Cp/Cpk (within-subgroup, short-term) and Pp/Ppk (overall, long-term). Capability is only meaningful on a stable process and, for Cp/Cpk, approximately normal data. Cp measures spread against specification; Cpk adds centring.
Attribute: DPMO, DPU, first-pass yield and rolled throughput yield.

Worked example

A GR&R on call-handling time reports equipment variation 9%, appraiser variation 6%, total GR&R 12.4% of study variation, ndc = 5. Reading: acceptable for improvement work. But check %tolerance before publishing capability - if GR&R is 28% of the tolerance band, the gauge consumes too much specification and capability results are unreliable. Action: tighten start/stop definitions, retrain appraisers, or improve the gauge.

Traps

Measuring first and arguing about definitions afterwards.
Reporting Cp/Cpk from an unstable process, or from non-normal data without transformation or a non-normal method.
Confusing %study variation with %tolerance.
Treating MSA as a formality: a 35% GR&R means the baseline is mostly noise.

Analyse Phase: Verifying Root Causes

This domain tests whether you can move from a list of opinions to verified root causes, using graphical tools first and statistical evidence where the risk justifies it.

Structure the search

Process level: detailed flowchart, value stream map, spaghetti diagram, value-added/non-value-added analysis and waste walk to locate delay, rework and handoffs.
Data level: Pareto (vital few), histogram, boxplot, run chart, scatterplot and multi-vari study to separate within-unit, between-unit and temporal variation.
Cause level: brainstorming plus Ishikawa (5M1E: man, machine, material, method, measurement, environment), 5 Whys, cause-and-effect matrix scoring Xs against Y, and FMEA for risk ranking.
Confirmation: hypothesis tests - two-sample or paired t for means, chi-square or two-proportion test for defect rates, ANOVA for more than two groups, regression for relationships. State H0/H1, set alpha (usually 0.05), check assumptions, and interpret statistical alongside practical significance.

FMEA

RPN = Severity × Occurrence × Detection. Severity is the anchor: a high-severity failure mode needs action even when RPN looks modest, because occurrence and detection scores can understate harm.

Worked example

Pick errors: 30% of orders come from the night shift but 62% of pick errors do. A two-proportion test gives p = 0.01, so the shift difference is unlikely to be chance. A Pareto of error types on that shift shows 71% are wrong-quantity picks on three fast-moving SKUs stored in adjacent, near-identical bins. 5 Whys: labels differ by one digit, pickers read the shelf label rather than the item, and both bins were re-slotted after a layout change with no verification step. The cause is verified with data, so the solution space is now specific: re-slot, add a check, change label format. Reasoning: statistics showed where, stratification and Pareto showed what, 5 Whys showed why.

Traps

Skipping verification and improving a merely plausible cause.
Reading a p-value as "the probability the null hypothesis is true", or treating p < 0.05 as proof of practical importance.
Presenting correlation as causation without a mechanism or a designed test.
Applying an RPN cut-off that overrides a high severity score.
Testing many hypotheses at once, inflating false positives.

Improve Phase: Solutions, Pilots and Lean Tools

This domain tests whether you can turn verified causes into solutions that are piloted, proven and handed over, not merely brainstormed.

From cause to solution

Generate options with structured brainstorming (silent or round-robin to avoid anchoring), benchmarking, mistake-proofing redesign, and Lean tools matched to the identified waste.
Screen options with a solution selection matrix using weighted criteria (impact on Y, cost, risk, time, acceptance), and re-check each against the verified root cause - if the cause disappears, the solution is on target.
Lean toolkit by waste: 5S and visual management for motion and waiting, standard work for variation, kanban/pull for inventory and overproduction, SMED for changeover, poka-yoke and jidoka for defects, cellular layout and line balancing for flow, TPM for downtime.
DOE: when factors interact, screening designs (fractional factorials) beat one-factor-at-a-time; a resolution IV design separates main effects from two-factor interactions. Confirm with a verification run.

Pilot before rollout

Define success criteria and the measurement plan before the pilot, run it in a representative area, and compare against baseline using the same operational definitions - with a run chart or hypothesis test, not eyeballing.
Update the FMEA (re-score severity, occurrence, detection), control plan and standard work; plan training, communication and rollback.

Worked example

A filling-line changeover takes 48 minutes against a 20-minute target. Observation splits activity into internal (machine stopped) and external (can run while producing). Eighteen minutes of preparation - tool retrieval, pre-heating, label staging - moves to external. Fittings are standardised to one size, removing nine minutes of adjustment; a shadow board and checklist with two-person verification cut search time. New mean is 19.5 minutes over 20 changeovers, with no upward drift on the run chart. Reasoning: externalise first, then simplify and standardise, then sustain with standard work.

Traps

Selecting the popular solution without linking it to the verified cause.
One-factor-at-a-time experiments that miss interactions.
Rolling out without a pilot, or piloting without pre-agreed success criteria.
Declaring victory before the control plan exists, or claiming benefits never validated by finance.
Treating 5S as tidying instead of a standard, audited system.

Control Phase: SPC, Standardisation and Closure

This domain tests whether you can prove the gain is stable and make it survive after the team leaves: control plan, SPC, reaction plan, standardisation and closure with handover.

Control plan

For each critical X and for Y, record: what is measured, the operational definition, method, sample size and frequency, owner, and the reaction plan - who does what, within what time, when a signal appears. Without a reaction plan a control plan is documentation, not control.

SPC essentials

Control limits are computed from the process (typically mean ± 3 sigma), not from the specification; they describe the voice of the process. Specification limits are the voice of the customer. Never plot specification limits as control limits, and never widen limits to hide a signal.
Chart choice: X̄-R for small subgroups (n ≈ 2–9), X̄-S for larger subgroups, I-MR for single measurements; p/np for defective units, c/u for defect counts.
Special-cause signals: one point beyond a control limit; a run of points on one side of the centre line (8 or 9 depending on the rule set used); 6 points steadily increasing or decreasing; 2 of 3 points beyond 2 sigma. Trends and stratification matter even when all points sit inside the limits.
Capability after improvement: Cp describes spread, Cpk adds centring. Cp = 1.33 with Cpk = 0.95 means variation is acceptable but the process is off target - shift the mean rather than shrink variation.

Worked example

The X̄ chart shows eight consecutive points above the centre line, all inside the limits. That is a special-cause signal, not "normal variation". First response is the reaction plan: check raw material lot change, a new operator, calibration due. Recalculate limits only after the cause is found and removed - never before.

Sustaining and closing

Standard work, visual management, training records, updated FMEA and control plan, audit schedule and error-proofing in place.
Handover to the process owner, benefits verified and signed off, lessons learned captured, documentation archived so the next belt does not repeat the analysis.

Traps

Confusing control limits with specification limits, or recalculating limits to erase a signal.
A control chart with no defined reaction plan or owner.
Closing the project before stability and capability are demonstrated.
Leaving the improvement dependent on the belt rather than the process owner.