Work Sampling in Logistics and Warehousing

Sampling can describe activity distribution in receiving, picking, packing and dispatch. It complements warehouse records when they do not capture the state being investigated. WMS coverage varies, and observation is not presumed always cheaper or less intrusive.

Define activities and population

Separate extraction, location search, packing, receiving, loaded and empty travel, resource waiting, administration and breaks/other. Searching may be necessary; verify method and cause before labelling it loss. Define simultaneous-activity rules.

State whether you estimate person-hours for the whole warehouse, a process or each shift. Proportions are neither units prepared per hour nor exact minutes per order; add volume, errors, distances and timestamps as required.

Plan size and coverage

Under independent Bernoulli observations:

n = ceiling[z²p₀(1−p₀)/e²]

For expected picking p₀=0.35, z=1.96 and e=0.03: n=ceiling(971.0711)=972. The margin is three absolute percentage points, not 3% relative.

Allocating by headcount represents person-hours only with equivalent exposure. For 40, 25 and 15 people on equal-duration shifts, one proportional allocation is 486, 304 and 182 readings. With different hours or days, use person-hours; separate shift estimates need their own precision.

Fifty rounds recording twenty people create up to a thousand participations, not a thousand independent visits. Check correlation between people, cycles and routes, recording actual time. No fixed fifteen-minute gap guarantees independence.

Field readings

Schedule instants with known probabilities, covering relevant peaks and troughs. Randomisation does not ensure a particular campaign captures every state. Avoid convenient-zone selection and do not replace missed readings without a predefined rule.

Use safe routes. Record the state at the selected instant and distinguish confirmed from inferred cause. Do not identify people by vest number unnecessarily; review privacy and access. A high “other” share may prompt review, but 5% is not a universal validity threshold.

Teaching example: 971 obtained records

Although 972 were planned, assume one reading was missed and 971 valid records remain. Analyse missingness and report achieved precision:

Extraction's approximate normal 95% interval is 22.5–28.0% under the simple model. This concerns the interval procedure, not posterior parameter probability or each person's productivity.

Equivalent time and decisions

If rounded 11.5% empty travel represents 80 people working eight hours, it corresponds to 73.6 aggregate person-hours daily. Across 22 days: 1,619.2 hours. At €18/h, that is €29,145.60 of valued time, not cash savings. Some empty returns may be necessary; validate feasible reduction and its use before budgeting benefits.

In another simulated scenario, 45 people ×8 h ×22 days ×12 months represent 95,040 annual hours. Search falling from 14% to 6% gives eight points: 7,603.2 hours or €136,857.60 at €18/h. Another non-overlapping eight-point reduction has the same equivalent. Adding them does not establish €273,715.20 savings: include implementation costs and realizable effects without double counting capacity.

Monitoring and technology

A p-chart may track a category using comparable subgroups and an appropriate model. With p̄=0.252 and n=50, normal three-sigma limits are approximately 0.068–0.436, bounded by zero and one. A signal prompts investigation; it does not identify the cause automatically.

WorkSamp supports observation planning and recording. Export is not verified WMS integration. Compare operational records without inventing AI, sensor or automatic-connection functions.

Sources

NIST: proportion intervals. NIST: p-charts. GDPR.