Blog / Performance
Performance · Guide

Cycle Time Variation: Causes and How to Measure

OL By OEE Lab Editorial |Updated 2 November 2026 |6 min read
Short answer

Cycle time varies because of the machine, the material, the method, the person loading it and the stations around it. Measure it by timing at least 30 consecutive cycles, then working out the mean, median and standard deviation. A big gap between mean and median means a few long cycles are doing the damage, which often points to hidden short stops.

Two machines can have the same average cycle and behave nothing alike. One gives you 42 seconds all day, and the other swings between 38 and 60.

The second machine makes planning harder, starves the next station and hides problems inside an average. That is why the spread deserves its own number.

What to time

Decide first what one cycle means, and mark the start and end at the same physical event each time. A common choice is one part leaving the machine to the next part leaving it.

Timing from output to output catches everything, including loading, waiting and short halts. Timing only the machine's working motion misses most of the variation, because that part is often very steady.

For definitions of the terms around it, see takt time vs cycle time vs lead time. The cycle time calculator handles the simple average.

How to measure it

1
Fix one product
Mixing products mixes their ideal rates and looks like variation.
2
Time 30 or more cycles in a row
Consecutive cycles only. Do not skip the odd ones, because they are the point.
3
Note what happened
Write a few words beside any cycle that looks long.
4
Calculate three numbers
Mean, median and standard deviation.
5
Repeat
Another shift, another lot, another day.
The method, with a stopwatch or a machine signal.

A worked example

Example: an operator-loaded press with an ideal cycle time of 40 seconds, timed for 20 cycles in a row. That is short enough to follow here, but use at least 30 for real work.

58 s 62 s Ideal 40 s (dashed), cycles 1 to 20
The data: 41, 42, 40, 43, 41, 44, 42, 41, 58, 42, 43, 41, 42, 40, 45, 42, 41, 62, 43, 42 seconds.

The sum is 875 seconds, so the mean is 875 / 20 = 43.75 seconds. The median, the middle value, is 42 seconds.

The standard deviation is the usual measure of spread. Any spreadsheet gives it: use =STDEV.S(range).

CV = standard deviation / mean
All 20 cycles: 5.73 / 43.75 = 13.1%
Without the two spikes: 1.31 / 41.94 = 3.1%
Coefficient of variation (CV) puts the spread in percent, so you can compare machines with different cycles.

Look at what two cycles did. They raised the mean by almost 2 seconds and made the spread more than four times bigger.

The gap between the mean (43.75) and the median (42) was the first warning. When the mean sits well above the median, a few long cycles are pulling it up.

Read the shape, not just the number

A simple count of cycles by time often says more than any formula. Here are the same 20 cycles, grouped.

40-41 s7 cycles
42-43 s9 cycles
44-45 s2 cycles
46-57 s0 cycles
58-62 s2 cycles
Share of the 20 cycles in each time band. The empty band in the middle is the clue: two different things are happening.

One hump with a tail means normal scatter. Two separated groups mean two processes, and here the second one is a stop in disguise.

The notes beside those cycles might say "part stuck in the chute". Whether you call that a long cycle or a short stop depends on your stop threshold.

Split the loss into its two parts

Against the ideal 40 seconds, the 20 cycles lost 875 - 800 = 75 seconds. They split into two very different problems.

  • Spikes: 36 seconds. That is (58 - 42) + (62 - 42). They point to jams, feeding and micro-stops.
  • Steady gap: 39 seconds. That is all 20 cycles running about 2 seconds over the ideal. It points to speed settings, loading method or a wrong ideal rate, which causes of reduced speed helps you test.

The average alone would have told you only that performance is 40 / 43.75 = 91.4%, as the performance formula shows. The split tells you where to walk next.

Why cycle time varies

SourceExamplesCompare
MachineWarm-up, hydraulic oil temperature, worn tools, slow actuatorsFirst hour against later hours, before and after service
MaterialTolerances, hardness, curl, feed problemsLot against lot
MethodDifferent loading sequence, parts not in reach, hand tools movedVideo of a fast and a slow cycle
PersonExperience, fatigue, one person minding two machinesCrew against crew, hour against hour
NeighborsWaiting for parts, or for room at the next stationCycle time against buffer level
A mostly automatic machine tends to be steady, so large spread there deserves a look. Manual steps naturally vary more.

Split each cycle into machine time and handling time if you can. If machine time barely moves, stop looking at the machine and watch the hands.

Why spread costs output, even with the same average

Variation hurts most where stations hand parts to each other with no buffer. The slower of the pair decides every cycle.

Example: two linked stations each take either 40 or 50 seconds, at random, with equal odds. Each one averages 45 seconds, so you might plan for 45.

But a pair finishes in 40 seconds only when both are fast, which is one time in four. The real average is (40 + 50 + 50 + 50) / 4 = 47.5 seconds.

That is 3,600 / 47.5 = about 76 parts an hour instead of the 80 you planned for, from variation alone. A buffer between the stations or a steadier process wins that back.

What is a good number?

We found no neutral benchmark for cycle time CV across machines, and it would mean little if we had. A manual bench and a servo press are not comparable.

Use your own baseline instead. Measure the best stable stretch of a machine, make that your reference, and track the trend after each change.

How the measurement goes wrong

  • Removing outliers before you look at them. Those are the cycles you came for.
  • Mixing products. Two products with different cycles look like one very unstable process. See OEE by product.
  • Timing while people watch. Crews tend to work differently with a stopwatch nearby, so a machine signal is better for long runs.
  • Using the mean as the ideal. The mean includes your losses. See ideal cycle time.

Where to start

Take the spikes first, because a few causes usually sit behind them and removing them is quick. Then work on the steady gap with a speed trial.

For a lasting view, a cycle signal from the machine beats a stopwatch. What is machine monitoring covers the options.

See it live in Fabrico

A stopwatch gives you 30 cycles, while a machine signal gives you all of them. Fabrico takes stops, speed and counts live from PLCs or IoT sensors, including external sensors on older machines, with OEE per line and shift. See how Fabrico connects to machines.

What people ask next

How many cycles do I need to time?

At least 30 consecutive cycles is a common starting point, and more is better on long-cycle machines with rare problems. Then repeat on another shift. A signal from the machine, as in PLC signals for OEE, removes the limit.

What is the difference between cycle time variation and speed loss?

Variation describes how much cycles differ from each other. Speed loss is how far the average falls short of the ideal. A line can be steady and slow, as speed losses shows.

How do I calculate standard deviation in Excel?

Put your times in a column and use =STDEV.S(A2:A31), then =MEDIAN(A2:A31) and =AVERAGE(A2:A31) for the other two. Divide the standard deviation by the average to get CV.

When does a long cycle become a stop?

When it passes your stop threshold, which you should write down and apply the same way for all crews. Below it, the time stays in performance. Stop threshold in OEE explains how to choose.

Should I set takt time using the average cycle?

Takt comes from demand, not from the machine. Compare it with your cycle time, but plan with room for the spread. The takt time calculator does the first half of that.

See OEE per line and shift on your own machines

Fabrico is a light MES for OEE. It takes stops, speed and counts from PLCs or IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and operators tag the cause at the line.

Book a Fabrico demo

Related guides: Assembly Line Efficiency: Formula and Example · Cycle Time vs Process Time, With Takt Time

← More OEE Lab guides

OEE essentials: OEE calculator · What is OEE · Six big losses · Availability · TEEP · Cost of downtime · OEE benchmark · Free OEE tracker · Free OEE software · How to track downtime · How to implement OEE · Light MES vs full MES