Imagine a warehouse timing the same packing task five times. This is a fictional teaching example. The cycles come back at 40, 42, 41, 80 and 43 seconds.

Four of those readings sit within three seconds of each other. One does not. What happens to that eighty determines the estimate, and through it the headcount, the budget and the roster.

The average has already made the decision for you

Take the mean and you get 49.2 seconds. Take the median and you get 42. The gap between them is more than seven seconds per cycle, and nobody has yet asked a single question about the work.

That is the uncomfortable part. Choosing a formula before investigating the observation lets a statistic settle an operational question.

Measure Result What it tells us
Mean of all five 49.2s Average observed time, including the slow cycle
Median 42s The middle reading; less sensitive to the slow cycle
Mean excluding the 80 41.5s Average of the remaining four; exclusion needs evidence

Three correct calculations, but not three automatically valid labour standards. The median does not establish that the slow cycle is irrelevant, and excluding it does not establish that it was a different task. Validate the task definition, sample and explanation before choosing a planning estimate.

Go and ask about cycle four

Only the floor can settle it. Was that order genuinely bigger? Did the strapping machine jam? Was it a different product that happens to run through the same station? Was the packer interrupted?

Each answer leads somewhere different. A bigger order may call for a separate task category or a volume driver that reflects item count. A jam should be recorded separately from manual work, with its frequency investigated. An interruption may be another task or a recurring delay. If it belongs in the labour requirement, account for it once, in the appropriate place.

Record the reason next to the reading. An observation without its explanation cannot be defended six months later, when someone asks why the standard is what it is.

Ten seconds is not a small number

It is tempting to treat a few seconds as noise. Run the arithmetic before deciding that.

Ten extra seconds per order across a thousand orders a day is 10,000 seconds, near enough to 2.8 hours of work every day. At an assumed 250 operating days, that is about 694 hours per year. Against 2,000 paid hours per full-time equivalent (40 hours × 50 weeks), it is roughly 0.35 FTE before allowances or absence cover. That is a workload equivalent, not an automatic change to headcount or payroll.

This is also why allowances deserve their own conversation. An allowance covers defined needs such as recovery and unavoidable delay. If a delay was already recorded during the study, adding it again inside the allowance pays for it twice.

Test it yourself

Measure a task, question the estimate starts with a separate, twelve-cycle dataset broken into task elements and a recorded delay. It lets you inspect the slow cycle, change the method, adjust the allowance and the volume, and follow the result all the way into an annual budget. Try moving the allowance by a single percentage point and watch where it lands.

A question for your next study: when someone hands you a task time, can you find out how many cycles it came from — and what happened during the slowest one?