Pareto Law in Delays
Build and interpret a delay Pareto chart without treating the 80/20 rule as a law every dataset must follow.
Updated:
A Pareto chart orders categories by contribution and shows their cumulative share. The 80/20 rule is a management heuristic, not a law requiring every plant to distribute its losses that way, nor evidence that observations follow a mathematical Pareto distribution. A high contribution does not prove that the entire loss can be removed.
1. Define the measure before drawing the chart
Specify unit, period and metric. Do not mix incident counts with lost time: a hundred brief stops can occupy less time than one long failure. With suitably designed instantaneous sampling, frequencies estimate time shares for the defined resource; they are not counts of events occurring during the period.
A waiting observation is a state, not necessarily its root cause. Collect contextual evidence to distinguish materials, coordination, maintenance and other hypotheses. Do not classify training, safety or legitimate breaks as removable waste merely because of a label.
2. Sampling design and limits
For a proportion, n=ceil(z²p₀(1−p₀)/e²) is an initial approximation. With p₀=0.5, z=1.96 and e=0.03, it gives 1,068 independent visits. This is not a number per plant or per 50 operators. Recording several workers in one visit does not automatically create that many independent units; analysis must reflect selection and correlation.
Cover relevant shifts, days and conditions. Two or three weeks is a possible schedule, not a representation guarantee. Randomization does not remove misclassification or participant reactivity, nor justify a fixed 15–20% bias claim about stopwatch studies.
3. Build the chart
- Define categories, priority rules and an unclassifiable state with reasons. MECE does not require pretending uncertainty never occurs.
- Remove duplicates and document missed visits without inventing replacement records.
- Estimate contributions with weighting where the design requires it.
- Rank contributions and calculate cumulative percentages using an identified denominator.
- Review precision: similar ranks may change because of sampling uncertainty.
- Prioritize using contribution, risk, cost and feasibility. A small safety issue can outrank the largest time share.
The ASQ Pareto resource describes the prioritization tool. Drawing a chart does not establish causality.
4. Reproducible simulated example
Assume 1,000 equal-probability visits, including 200 delay states. The table is invented for teaching, not a sector benchmark or customer result.
| Delay state | Records | Share of delays | Share of all visits | Cumulative delay share |
|---|---|---|---|---|
| Waiting for material | 80 | 40% | 8% | 40% |
| Adjustment | 60 | 30% | 6% | 70% |
| Waiting for instructions | 30 | 15% | 3% | 85% |
| Equipment failure | 20 | 10% | 2% | 95% |
| Unclassifiable | 10 | 5% | 1% | 100% |
The first two categories are 40% of five categories and account for 70% of delays: not 20/80. The first three reach 85%. An 8% all-visit share waiting for material is not an 8% payroll saving or an 8% output increase.
5. Actions and follow-up
Investigate supply before implementing kanban; test whether adjustments can be shortened without harming quality; review authorization needs before removing controls. Compare follow-up data with matching definitions and coverage, considering product mix and demand. An uncontrolled before/after comparison does not establish the causal effect.
Delay shares depend on the process and period. To turn a Pareto priority into an improvement forecast, specify the intervention, constraints and outcome measurement. A distribution of causes alone cannot predict savings or an OEE increase.
6. OEE and tools
OEE needs planned time, run time, ideal cycle time and total/good counts: Q=good units/total units. Delay shares alone cannot calculate A×P×Q, and no sensor-equivalent accuracy is established. See the OEE calculation.
WorkSamp supports random observations and reports; Cronometras is for time studies; Induly is for working-time records. No automatic integration or savings are promised. Additional resources include ASETEMYT, its blog, company listing and ProdCont.
Statistical reference
NIST: selecting sample sizes. An interval quantifies uncertainty under assumptions; it does not itself validate a study or investment.