Tools, Techniques and Data
Choosing, not collecting
Technique questions rarely ask what a tool is; they ask which tool fits the situation. Build a mental index keyed by the question being asked, then eliminate.
Concepts to hold
- Modelling. Process models with swimlanes show who does what and where handoffs occur; the handoff, not the step, is usually where the defect lives.
- Root cause. Five whys for a single traceable chain; fishbone (cause and effect) for divergent causes across people, process, system, data; Pareto to find the vital few contributors; scatter plots for relationships.
- Traceability matrix. Links objective, requirement, design, test and outcome. Its absence is why scope and testing drift apart.
- Data quality dimensions. Accuracy, completeness, consistency, timeliness, validity, uniqueness. Name the dimension before proposing a fix; a completeness problem is not solved by cleansing accuracy.
- Sampling and measurement. A sample supports inference, not certainty; state confidence when reporting.
- Dashboards. Leading indicators predict, lagging indicators confirm. A dashboard of only lagging measures tells you about the past.
Traps
Correlation read as causation. Averaging a bimodal distribution and reporting a meaningless middle. Selecting a technique because it is familiar. Believing more data always improves a decision.
Rule of thumb: name the question first
"Why did it happen?" - root cause tools. "Where should we act first?" - Pareto. "Is there a relationship?" - scatter or correlation, with caution. "Is the process behaving as designed?" - control chart or audit.
Worked illustration
Complaints by category: late delivery 52%, damaged goods 23%, wrong item 15%, billing 10%. Pareto redirects effort to late delivery first; five whys then traces it to a single cut-off time. Both tools, used in that order, beat either alone.