Six Sigma DMAIC and Work Sampling: When to Use Each

DMAIC structures a project as Define, Measure, Analyse, Improve and Control. Sampling is a measurement technique usable within Measure or exploratory work. They are not mutually exclusive alternatives and combining them does not guarantee returns.

Five design questions

1. Is there a defined metric or only a suspicion? Establish unit, period and denominator.
2. Is the cause confirmed? Activity shares indicate where to investigate, not root-cause proof.
3. Which positions and conditions are comparable? Separate or weight different areas where appropriate.
4. What decision is needed and when? A short window limits coverage rather than making a small sample representative.
5. Who can implement and sustain changes? Establish owners, resources, quality and safety.

No universal two-week diagnosis or ROI rule requires starting with one technique.

Sampling in Measure

Define exclusive observable states, trial agreement and select times with suitable coverage. Retain omissions and delays. External observers do not eliminate reactivity; explanation and piloting may help without guaranteeing unbiased data.

Independent simple planning uses n₀=ceiling(z²p₀(1−p₀)/e²). At two-sided 95%, p₀=0.5 and five points, it gives 385; at three points, 1,068. z=2 is a different confidence approximation. These are not automatically per-position minima or requirements for detecting every difference.

Useful categories

Separate execution, flow waiting, preparation, confirmed breakdown, planned maintenance, logistics and breaks/training. No seven-category framework is verified to cover 95% of plants. Support and breaks are not automatically waste.

Waiting state and missing-reel cause are different levels. Avoid overlapping logistics and flow-waiting codes. MTBF and MTTR are event/duration indicators, not equivalent category names.

Simulated packaging case

A fictional plant has three lines, about twenty-eight direct positions and two shifts. Assume 1,576 observations across nine observed shifts; do not describe this as three days times three shifts when there are two.

Suppose 27% of stopped-state records have confirmed breakdown causes and 42% logistics causes, with the same stopped-state denominator. These are not whole-shift percentages. Assume additional records confirm reel waiting and information release; observations alone do not prove those causes.

A charter might target logistics below 15% of stops within twelve weeks, alongside absolute downtime and output. Reducing a share within stops does not guarantee total downtime falls.

Improve and Control

Assume 1,620 readings later and logistics at 13% of stops. Evaluation needs stopped-state counts, exposure, precision, mix and comparable conditions. Totals and conditional percentages alone cannot establish significance.

The previous scenario mentioned OEE rising from 58% to 71%. That is a separate hypothesis requiring planned time, run time, ideal cycle and good/total counts, not a result derived from cause shares. It is not presented as an actual Worksamp outcome.

Validate changes without removing necessary controls. Retain monitoring and review criteria after project closure.

Tools and sources

Worksamp supports random plans, capture and distribution analysis. It is not Six Sigma certification or a guarantee of audit acceptance or root-cause identification. Cronometras can complement specific timings; attendance records supply a different exposure source, not observation substitutes.

NIST sampling, OEE definitions. See Worksamp for supported functions.