CLBB common mistakes
CLBB · Understand the traps and how to avoid them.
Treating Lean Six Sigma as a set of tools to memorise, so when a scenario asks which methodology fits a problem the candidate picks DMAIC or Kaizen by keyword rather than by problem characteristics.
Study material is often organised tool-by-tool, so tool recall is strong but the decision rule for choosing between DMAIC, DMADV, Kaizen, and rapid improvement events is never internalised.
What to do instead
Build a one-page decision table keyed on problem traits (new design vs. existing process, data availability, scope, urgency) and practise classifying 15 short scenarios before reading any answer choices.
Confusing the roles of Champion, Sponsor, Master Black Belt, Black Belt, Green Belt, and process owner, and assigning accountability for project selection or resource approval to the wrong role.
Role names overlap in everyday usage and candidates map them onto their own workplace hierarchy instead of the defined belt structure.
What to do instead
Write a role-by-responsibility matrix (project selection, resource approval, tool coaching, data collection, handoff to control) and drill scenario questions that hinge on who approves or who executes.
Writing a problem statement or business case that embeds a solution or an assumed root cause, then failing questions that test whether a charter is well formed.
Real-world habits push toward solution language, and candidates evaluate charters for completeness rather than for solution-neutrality and measurable baseline.
What to do instead
Apply a three-part test to every charter element: is it measurable, is it bounded in time and scope, and is it free of cause or remedy language. Reject any option that names a fix inside the problem statement.
Selecting the wrong tool for a given data type, for example applying a parametric test to ordinal or non-normal data, or using an I-MR chart for subgroup data.
Candidates memorise tool names and formulas but not the assumptions and data-type preconditions that gate each tool.
What to do instead
For each analysis tool, record its data type, distributional assumption, and minimum sample requirement, then practise matching a described dataset to the correct test before computing anything.
Confusing common-cause variation with special-cause variation when interpreting a control chart, and treating a stable process as out of control or vice versa.
Rule-based signals are learned in isolation from the concept of process stability, so candidates react to individual points instead of applying all detection rules and the stability definition.
What to do instead
Practise reading full chart patterns, not single points, and state explicitly whether the process is stable before recommending action. Tie every recommendation back to whether the signal is a rule violation.
Calculating process capability indices without checking stability first, or reporting Cp and Cpk interchangeably when the process is off-centre.
Formulas are easy to compute, so candidates skip the prerequisite that capability is only meaningful for a stable process, and they conflate spread with centring.
What to do instead
Adopt a fixed sequence: verify stability, verify normality or transform, then compute Cp for spread and Cpk for centring. Practise questions that ask which index is invalid and why.
Mistaking measurement system variation for process variation, or accepting a gauge study without interpreting the percentage of variation against decision thresholds.
Gauge R&R output is presented as numbers, and candidates recall that low is good without knowing how repeatability and reproducibility decompose or what threshold triggers action.
What to do instead
Learn the decomposition of repeatability versus reproducibility and the standard percentage-of-study-variation thresholds, then practise deciding whether to fix the measurement system before collecting process data.
Treating correlation as causation or stopping analysis at a significant correlation without running regression or designed experiments to confirm the relationship.
Scatterplots and correlation coefficients feel conclusive, and candidates under-weight confounding and the need for a designed study.
What to do instead
For every correlation question, ask what confounder could explain it and what follow-up design would isolate the effect. Practise rejecting answer options that assert causation from observational data.
Designing or interpreting an improvement pilot without a clear before-and-after comparison plan, so improvement claims rest on anecdote rather than on a valid test of the change.
Improve-phase content is often studied as a list of Lean tools, leaving the statistical validation of the improvement under-practised.
What to do instead
For each improvement scenario, specify the metric, the baseline period, the comparison method, and the sample size before evaluating any proposed solution. Reject options that claim improvement from a single post-change observation.
Building a control plan that documents the metric but omits the reaction plan, owner, and monitoring frequency, so the improvement is not sustained.
Control plans are treated as paperwork to complete rather than as the mechanism that transfers ownership, and candidates check for the presence of a chart rather than the completeness of the response protocol.
What to do instead
Require five elements in every control plan answer: metric and specification, monitoring method and frequency, responsible owner, reaction plan for out-of-control signals, and audit or review cadence. Eliminate any option missing one of these.