Work Sampling in Manufacturing: Find Hidden Delays

Home » Resources » Work Sampling in Manufacturing: Find Hidden Delays

Estimated reading time: 12 minutes

A walk through the factory can make the same shift look busy at one moment and underused at another. Neither visit tells you how much time goes into making parts, preparing work or waiting for something to arrive.

Work sampling replaces those impressions with a planned set of random observations. It helps you find the conditions that interrupt production and choose an improvement worth testing. The useful result is not a score for how busy people look. It is a clearer answer to what prevents work from moving.

Once you have verified what an improvement changes, SkyPlanner APS turns better operation times and resource availability into a coordinated production schedule. You spend less time rebuilding separate machine queues and can see what the change means for the orders you need to deliver.

What is work sampling?

Work sampling is a statistical method that estimates the proportion of time spent on defined activities by observing people or machines at randomly selected moments. Each observation records the activity happening at that instant. The share of observations in a category estimates its share of the time covered by the study, provided the sampling design represents that time.

For example, an assembly-area study might distinguish assembly, setup, inspection, waiting for material and other interruptions. Recording those categories over representative shifts gives you more useful evidence than calling everything either productive or idle.

The method is particularly useful when activities vary, interruptions are scattered and continuous observation would take substantial effort. You do not have to follow every operation from beginning to end to understand the broad distribution of activity.

Choose a different measurement when you need a different answer:

• A manufacturing time study examines defined work elements and can establish a justified standard time. Random activity snapshots do not tell you how long each short element takes.
• Manufacturing time tracking records events and actual time against jobs or operations. Those records answer which work used the booked hours, rather than providing a random sample of all activity.

A formal work-measurement study can use sampling alongside other evidence to establish standards. An activity percentage alone, however, is neither a standard time nor a measure of someone’s working pace.

Start with a decision, not a busy-versus-idle tally

Choose a question whose answer could change how the factory works. “How much of the assembly team’s time is spent waiting for components?” is a useful starting point if you can act on material preparation. “How efficient is everyone?” is too vague to design or interpret well.

Agree the purpose with the people doing the work. Explain what will be observed, how categories are defined, who will see the results and what decision they support. Ask operators which interruptions the existing records miss. A study is more useful when people can explain a missing kit than when they feel encouraged to conceal the wait.

Define the population: the people or machines, shifts, days and production conditions the result should describe. A day-shift study of one assembly area does not automatically describe night work or the whole plant.

Then choose the observation unit. An operator at one instant is one unit; a machine at one instant is another. Keep their states separate. A machine can be cutting automatically while its operator inspects another part. The machine is running, and the operator is inspecting: neither observation needs to be labelled idle.

Make the time boundary explicit, too. If the study covers the whole paid shift, include the applicable break category. If it covers only defined operating periods, report that denominator. Do not compare the two as though they measure the same thing.

Define activities that observers can recognise

Use a small set of categories that are mutually exclusive for the chosen observation unit and cover the situations you expect. Write a short definition for each, with examples from the actual area.

“Waiting for material” should mean that the selected work cannot proceed because the required material is unavailable, not that the operator happens to be standing near the material rack. “Setup” should have an agreed boundary from preparation to readiness for production. Keep necessary inspection visible rather than burying it in a general non-production category.

Include an unknown or not-observable code. If you cannot establish what is happening, record the uncertainty rather than guessing a cause. Review those cases separately; they should not quietly disappear from the denominator.

Test the definitions in a short pilot. Have observers classify the same situations and discuss disagreements. If one person calls a tool search setup and another calls it waiting, the study will partly measure your definitions rather than the work. Fix that before collecting the main sample.

The pilot also reveals practical problems: inaccessible observation points, a route that takes too long, or categories so detailed that a momentary glance cannot distinguish them.

Collect random observations across representative production

Generate observation times in advance using a random-number tool, and spread the sampling plan across the shifts and days the decision concerns. Select the person or machine according to the plan rather than choosing whichever looks most interesting when you arrive.

A regular visit every half hour can repeatedly coincide with replenishment, a break or a machine cycle. Visiting only when the supervisor has spare time creates a different bias. Random timing helps avoid those patterns; covering the relevant operating conditions makes the result useful beyond a convenient morning.

At each selected moment, record the current activity without waiting for the next one to begin. Retain the timestamp, observation unit, category and relevant shift or product context. Keep a brief factual note when it will help investigate a delay. Do not interrupt the activity to obtain an explanation and then classify the conversation you just created.

Record missed observations and why they were missed. Replacing every inaccessible or awkward observation with an easy nearby one can change what the sample represents.

If you deliberately sample one shift more heavily, report the shift results separately or combine them using weights that reflect the time or population being estimated. Simply pooling unequal sampling rates can make an unusual shift dominate the answer.

Also distinguish observation count from independent information. Looking at several operators during one tour produces several records, but a shared material shortage can make their states related. Repeated snapshots during the same stoppage have a similar limitation. Keep tours and time groups identifiable so the analysis can account for that clustering.

How many work-sampling observations do you need?

The required sample depends on the precision needed for the decision and the expected share of the activity. There is no observation count that makes every study representative.

For planning a simple random sample of independent observations, a common approximation for one activity proportion is:

n ≈ z² × p × (1 − p) ÷ e²

Here, n is the observation count, p is the expected proportion written as a decimal, e is the desired absolute margin of error, and z is the normal-distribution value for the chosen confidence level. For an approximate 95% confidence level, z is 1.96. Round the resulting count up.

If you do not yet have a credible estimate of p, using 0.50 gives the largest count for a given absolute margin. With a margin of five percentage points, e = 0.05, the calculation gives 385 observations. That is a planning example for these assumptions, not a universal minimum or a guarantee for every subgroup.

Be precise about “five percent.” A waiting share of 20% with a margin of five percentage points means roughly 15% to 25%. A relative margin of 5% of that share means only one percentage point either way, which requires a much larger sample.

Plan precision for the category and comparison that matter. A total adequate for an overall waiting share may be inadequate for separate shifts, rare delays or individual workstations. Clustered tours also need design-aware analysis rather than treating every tick as independent. If the decision involves a small change or substantial investment, get a sampling specialist to check the design before collecting the main dataset.

Example: turn activity counts into an improvement question

Imagine a manual assembly area where the team suspects that incomplete material kits interrupt work. A study collects 400 equal-probability operator-moment observations across the production periods in scope. For this simplified example, assume the observations are independent and every observation can be classified.

Observed activity	Observations	Share of observations
Assembly work	200	50%
Setup and preparation	60	15%
Inspection	40	10%
Waiting for material	80	20%
Other identified delays	12	3%
Planned breaks	8	2%
Total	400	100%

The estimated material-waiting share is 80 ÷ 400 = 20%. This describes the pooled operator time represented by the study. It does not mean every operator waits for exactly one fifth of every shift, or that each interruption lasts the same length of time.

For this single proportion, a simple normal approximation gives:

Approximate 95% margin = 1.96 × √[0.20 × 0.80 ÷ 400] = 0.0392

That is about 3.9 percentage points, giving an approximate interval from 16.1% to 23.9%. It expresses sampling uncertainty under the stated assumptions. It does not cover a biased observation schedule, misclassified activities or a different product mix next month.

For small samples or activity shares close to zero or one, use a more suitable interval, such as a Wilson or exact binomial interval, rather than relying on this symmetric approximation. Use a method that also reflects the sampling design when observations are grouped or weighted.

The operational conclusion is clearer than the decimal places: material waiting warrants investigation. Start by checking where the affected kits became incomplete, which components were missing and whether the next assembly could have started instead. The category identifies where to look; the team and the supporting records explain why it happened.

Improve the condition, then verify what changed

Review the findings with operators, material handlers and the person responsible for the affected process. Separate necessary work from avoidable interruption. Inspection may protect quality, setup makes the next batch possible, and breaks support the working day. None becomes removable waste just because it is not assembly.

Choose a focused change and an owner. In the assembly example, that might mean confirming kit completeness before releasing work to the area. Check whether the missing-component problem decreases and whether the work can actually progress sooner, rather than merely moving the wait to another department.

Repeat observations under comparable conditions, keeping category definitions and time boundaries consistent. Compare product mix, staffing and unusual events as well as the headline percentages. A small percentage change within the study’s uncertainty is not convincing evidence on its own; a changed mix can also make the result look better without improving the method.

Most importantly, do not turn the original 20% waiting share directly into 20% additional capacity. The released time may be fragmented, require another person or coincide with a machine already running. It may also be time during which no alternative job could use the workstation.

Verify the effect on the resource you intend to schedule. If an improved preparation method genuinely shortens setup, measure and approve the new setup duration. If it makes a workstation available for a longer operating window, confirm that the people and equipment can support that availability. Keep labour effort, elapsed operation time and machine occupancy distinct.

Put verified improvements into the production schedule

A validated improvement changes more than the workstation you studied. A shorter setup can let the next order start earlier, but that order still needs material and space at its following operation. Other orders may be competing for the same machines. This is where a local improvement needs a shared production plan.

SkyPlanner lets you define process-step durations as time per piece, an output rate or a fixed time, with setup and teardown kept identifiable. Workstation shifts define the available time windows. Arcturus, SkyPlanner’s scheduling AI, calculates the schedule against configured capacity, dependencies and priorities, and the Gantt timeline makes the result visible.

Imagine that follow-up measurement validates a setup change from 40 to 30 minutes for a defined family of jobs. Maintain the affected planning inputs so the relevant jobs use that approved setup time, then recalculate their place alongside the rest of the workload. Review the downstream operations and delivery dates, not just the gap left at the first station.

SkyPlanner handles that connected scheduling calculation, saving the manual work of fitting each revised job into separate machine queues. You decide which improvement is valid and which orders take priority; Arcturus applies those instructions to the plan. Better evidence becomes useful production capacity where the work can actually use it.

Fredman Operations’ SkyPlanner experience illustrates the scheduling benefit. The food-packaging manufacturer previously lacked clear shift-by-shift capacity visibility, and urgent rescheduling could take an hour or more. Its implementation brought machine schedules and shift capacity into a shared Gantt view. The company describes faster responses to changes and a clearer basis for delivery commitments. Its story shows the value of coordinating the workload after circumstances change.

Choose SkyPlanner when verified shop-floor improvements need to become a workable plan for the whole order book. It integrates with any ERP or MES through its open REST API, so you can keep the systems that manage your business while improving how production is scheduled.

Book a SkyPlanner demo with one improvement, its approved effect on duration or availability, and other orders using the same resources. That gives you a concrete way to see what the change means for the work you have promised. If you prefer to explore first, try SkyPlanner with demo data.