Work Sampling: Production vs Logistics

Executive summary. Work sampling addresses a similar management question in two different environments: how working time is allocated and what contributes to the cost of producing or dispatching a unit. However, an observation plan designed for a production line cannot simply be copied into a warehouse, or vice versa. This article explains the differences—flow repetitiveness, work unit, measurement categories, and observation coverage—and how to avoid investment, pricing, or delivery decisions based on data that do not represent the operation.

---

1. The Business Problem: Data That Describe the Operation

When operations management asks what manufacturing or dispatching a unit costs, business systems can provide attendance hours, orders, and unit counts. They may not explain all the waiting, travel, or unplanned work on the shop floor or in the warehouse.

Work sampling helps address this gap through instantaneous observations. Instead of timing each cycle, observers record people's current activity at randomly selected times. The resulting proportions estimate the distribution of activity during the population and period sampled, with statistical uncertainty conditional on the design.

The technique is the same in a plant and a logistics center; the operating context is not. Who is observed, which categories are used, how many observations are taken, and which days are covered must fit that context. Activity proportions alone do not identify the full unit cost: compatible labor-time, output, and cost data are also needed.

2. What Changes Between a Plant and a Warehouse?

Aspect	Production plant	Logistics center
Flow	Often repetitive cycles involving a product	Order-driven flows with varying content
Work unit	Operation or relatively fixed workstation	Variable task: picking, packing, exceptions
Typical hidden loss	Waiting between stages, micro-stoppages, adjustments	Travel, searching, system waits
Worker assignment	Often linked to a machine or cell	Often linked to a route and order flow
Observation	Fixed workstations and relatively short routes	Longer routes across warehouse zones

In production, work often centers on a workstation and cycle, even when variants exist. In logistics, order content, item locations, and exceptions can change the work required. These are useful design distinctions, not absolute rules: some warehouses have highly repetitive flows, and some plants operate with high-mix, nonrepetitive work.

The classification system should therefore reflect the decisions and activities in each operation rather than being copied unchanged.

3. Designing the Study in a Production Plant

1. Define the scope. Select a line, cell, or specified group of workstations. Combining areas with very different conditions can conceal differences; stratify them if an overall result is also needed.
2. Define decision-relevant categories. One possible framework uses value-added (VA), non-value-added (NVA), and nonproductive/delay (IMP), with subcategories such as assembly, material waiting, adjustment, rework, and cleaning. This is a chosen taxonomy, not a mandatory statistical standard. Define boundaries so categories do not overlap, including how necessary support work and breaks are treated.
3. Calculate the sample size. For a two-sided 95% confidence level and an approximate ±5-percentage-point margin for the main category, n = z²·p·q/e² with p = 0.5 gives 385 valid observations after rounding up. The familiar approximation of 400 comes from rounding z. This assumes an appropriate random-sampling model, not an unconditional guarantee. WorkSamp includes a calculation of the number of observations for a 95% confidence setting; verify the precision and assumptions selected.
4. Distribute observations across shifts and days. Planned maintenance, inventory counts, or demand peaks need to be represented according to the study's scope. Oversampling them without suitable weighting can distort an overall estimate; if they matter separately, report them separately.
5. Minimize observer influence. Explain the study and define any familiarization period in advance. Do not automatically discard or “smooth” early rounds because the results look different; exclusions need a documented rule that does not select the preferred outcome.

4. Designing the Study in a Logistics Center

1. Define zones and processes. Picking, packing, receiving, and dispatch may require separate strata or studies.
2. Adapt categories to actual warehouse flow. Beyond broad VA/NVA/IMP groups, useful detail can distinguish active picking, packing, travel, location searching, waiting for a system or cart, exception handling, and recorded breaks. Without clear detail, a finding of “nonproductive time” may not indicate a specific improvement.
3. Design routes that cover distant zones. Cover the relevant locations without favoring the easiest corner to visit. A long round is a sequence of observations, not a simultaneous snapshot of the whole shift. Record times and consider whether route order could bias coverage.
4. Interpret system records carefully. A system's assigned task duration does not necessarily separate actual location searching from waiting between orders. Observations can add context; first check which events the particular system already records.
5. Represent workload variation. Peak and quiet days can have different distributions. Cover both as required, with weights or separate reporting appropriate to the target period. An average can be meaningful, but it should not hide important operational differences.

5. Common Mistakes When Transferring a Study

• Using production categories unchanged in a warehouse. Categories such as assembly or machine adjustment may not fit picking, leaving too much under “other.” Adapt the definitions instead of forcing the observations.
• Calculating the sample only once. Required observations depend on the desired precision and design. Under the same conservative normal approximation, ±3 percentage points requires about 1,068 observations, compared with 385 for ±5 points.
• Observing only the convenient shift. Order-release peaks or night-shift conditions may be missed. Determine coverage from the operation, not observer convenience.
• Treating a time distribution as the solution. Sampling indicates where activity or waiting is observed. Establishing causes and choosing changes requires further process investigation. If detailed timing of a specific operation is needed, time study can complement sampling; see Work Sampling vs. Time Study.

Simulated Practical Case

This example was constructed for teaching purposes. It does not describe an actual WorkSamp implementation or a particular customer.

A company with its own plant and an adjacent warehouse runs two studies, each with 385 valid observations. That sample count corresponds to an approximate 95% confidence level and a ±5-percentage-point planning margin under the stated simple-random-sampling assumptions.

In the plant, covering 14 workstations in an assembly cell, 78 observations fall into delay categories: 20.3% of the total. These are concentrated in material waits between stages.

In the warehouse, covering 9 picking workers, the distribution differs:

• 96 observations involve travel: 24.9%.
• 61 involve waiting for a system or cart: 15.8%.
• 14 are classified as purely nonproductive: 3.6%.

These findings suggest different investigations: material supply between plant stages, and warehouse routing or system response. Sampling alone does not prove which intervention will work.

Pooling the listed categories produces 249 out of 770 observations, or 32.3%, labeled broadly as “nonproductive time.” This mixes plant delays with warehouse travel and other states, losing the distinction needed for action. Because the studies have equal sample sizes but cover different workforces, the pooled result is not automatically a workforce-time-weighted estimate for the company.

Checklist Before Launching the Study

• Is the scope—a line, cell, or warehouse zone—clearly defined and sufficiently homogeneous, or appropriately stratified?
• Does each category support a management question, with mutually exclusive definitions?
• Is the sample size based on the required confidence level, precision, and sampling assumptions?
• Do observations cover the relevant shifts and peak and quiet periods?
• Is the observer's role clear, with predefined measures to address observer influence?

Conclusion

Work sampling uses the same basic technique in production and logistics, but study design must fit each environment. Business value comes from categories that represent the real flow, an appropriate sample calculation, and representative observation coverage.

With those foundations, activity proportions can inform improvement, staffing, and delivery discussions. They do not replace causal investigation, capacity analysis, or cost accounting.

To explore how a study could be organized for your plant or warehouse, request a demonstration through WorkSamp. If you are deciding between sampling and stopwatch timing, start with Work Sampling vs. Time Study.

Statistical reference

NIST: Experimental Statistics, handbook 91 describes normal-approximation planning for a proportion. The 385 and 1,068 counts are planning calculations, not guarantees against sampling bias, correlated observations, or misclassification.

Technical reference.