The useful question is not the number. It is whether you can defend it.
Four tests that work regardless of what your figure is — starting with whether a high score is an achievement or a measurement artefact.
There is no universal answer, and anyone offering one without asking about your part mix, shift pattern and time base is quoting a number rather than answering a question. What can be said usefully is how to judge whether your own figure is good, which is a different and more answerable problem.
This article gives four tests that work regardless of what your number is.
A high OEE arrived at quickly is more often a measurement artefact than an achievement. The common ways a figure is inflated without anything improving: setup reclassified as planned time, short stops below the detection threshold, produced parts counted as good parts, and an ideal cycle time that flatters.
The test is simple. Ask what the figure would be if the time base were the full scheduled shift with nothing excluded, and how short a stop the system can detect. If the answer to the second is 'fifteen minutes', your availability figure is missing the category that usually matters most.
| Split | Reading |
|---|---|
| Availability well below performance | A stopping problem. Changeovers, breakdowns, waiting, manning. Usually the largest recoverable block on an unmonitored floor. |
| Performance well below availability | A running-slowly problem, or micro-stops being absorbed into performance. Check whether short stops are being detected at all. |
| Quality at exactly 100% | Almost certainly not being measured. A control counts cycles, not acceptable parts. |
| All three above 95% | Verify before celebrating. This pattern is rare outside genuinely continuous production and usually indicates a generous time base. |
A plant at 55% improving by two points a quarter is in better shape than a plant at 70% that has not moved in two years, because the first has a working improvement mechanism and the second does not.
The trend is also the only comparison that is reliably valid, because your own plant measured the same way is the one benchmark where the definitions are guaranteed to match. This is worth more than any published figure, and it costs nothing beyond consistency.
OEE on a non-constraint machine is a weak signal. Raising it produces work-in-progress rather than output, and plants that set uniform OEE targets across every machine reliably discover this the expensive way.
The useful question is what the constraint machine's OEE is, and whether the floor around it is organised to keep it running. A plant with 45% average OEE and 85% on its constraint may be performing better than one with 65% everywhere and 65% on its constraint.
If you want a figure to work with today, the OEE calculator computes both definitions from one shift's numbers, and the world-class OEE article explains why the benchmark you have probably been quoted is less informative than it appears.