OEE, measured properly: definitions, the six big losses and the errors that distort the number

Why the same OEE figure can mean different things

OEE compresses three questions into one number: was the asset running when it was scheduled to run, did it run at the speed it was designed for, and did it make good product first time. That compression is what makes the metric useful in a daily review, and also what makes it easy to distort. Each of the three inputs depends on a local definition: what counts as planned time, which cycle time is treated as ideal, and what counts as a good unit. Two plants that report the same figure may be measuring materially different things.

Before comparing lines, setting targets or buying a monitoring system, the definitions need to be agreed and written down. The international reference for those definitions is ISO 22400.

The standard definition

ISO 22400-2, the international standard for key performance indicators in manufacturing operations management, defines the OEE index as the product of three ratios. In the standard’s terminology they are availability, effectiveness and quality ratio; most plants call the middle term performance.

Two points in these definitions matter more than they first appear. Availability is measured against planned busy time, not calendar time: time with no production scheduled is excluded by design, which is why OEE measures how well scheduled time is used rather than total capacity. Performance compares output with what the asset could have produced at its ideal cycle time (the planned run time per unit, in ISO 22400 terms) during the time it actually ran. Using an observed average instead would conceal speed losses.

Total effective equipment performance (TEEP) applies the same logic to all calendar time. It is the better measure when the question is capacity rather than operating discipline.

A worked example

Consider one eight-hour shift on a single machine.

A useful cross-check: OEE should equal the time needed to make only the good units at the ideal rate, divided by planned busy time. Here that is 330 minutes ÷ 450 minutes = 73.3%. If the two methods disagree, one of the inputs is defined inconsistently.

The example also shows why OEE is demanding. Three ratios in the high eighties or above still produce a result in the low seventies, because the losses compound. If all three factors were 90%, OEE would be 72.9%.

The six big losses

The loss categories most plants use come from Seiichi Nakajima, who introduced total productive maintenance, OEE and the six big losses at the Japan Institute of Plant Maintenance in the early 1970s. His book Introduction to TPM was published in Japanese in 1984 and in English in 1988.

The value of the six-loss structure is that it turns one number into a short list of owned problems. A plant that knows its OEE but not its loss split has a scorecard, not an improvement plan.

Seven errors that distort OEE

1. Moving time between categories. If changeovers or planned maintenance are excluded from planned busy time on some lines and not others, availability figures are not comparable. Decide once which stops are planned, write it down and apply it everywhere.
2. Using an average cycle time as the ideal. If the ideal cycle time is set to what the line usually achieves, performance sits close to 100% and speed losses disappear from view. Use the design rate or the demonstrated best rate, per product.
3. Counting reworked units as good. Quality in OEE is first-pass quality. A unit that needed rework consumed capacity twice and should not be counted as good first time.
4. Missing short stops. Manual logs rarely capture stops of a minute or two, yet these accumulate across a shift. Where short stops are suspected, automatic state capture from the machine is usually the only reliable source.
5. Averaging percentages across assets. Line or plant OEE should be calculated from summed times and counts, not as a simple average of machine figures, which gives a lightly loaded machine the same weight as the bottleneck.
6. Comparing dissimilar processes. A line making many products with frequent changeovers will carry more availability loss than an identical line making one product. Comparing their OEE says little about which team performs better.
7. Treating OEE as the goal. When the number itself becomes the target, the easiest gains come from reclassifying time. Review OEE alongside output against schedule and the loss split, and audit category definitions periodically.

What about world-class OEE?

A figure of 85% is widely described as world-class. It also traces back to Nakajima, who set out minimum levels of roughly 90% availability, 95% performance and 99% quality, based on his practical experience. Vorne, which publishes the oee.com reference site, notes that these figures have roots in a particular place, period and industry (Japanese automotive manufacturing in the 1970s) and apply to discrete rather than process manufacturing.

A more useful target is one set against the asset’s own baseline. Vorne recommends stretch targets that are achievable within three to four months, raised once they are reached. The baseline matters more than the benchmark, and the loss split matters more than either.

From measurement to improvement

• Agree and publish the definitions: planned busy time, ideal cycle time per product, and what counts as a good unit.
• Capture machine state and counts automatically where the asset allows it, and use structured operator input for reasons the machine cannot know. Our guide to connecting legacy machines covers the options.
• Use a short, fixed list of stop reasons mapped to the six losses. Long free-text lists are rarely analysed.
• Review the loss split daily at line level and weekly at plant level, with a named owner for the largest loss.
• Re-baseline after each significant change to product mix, equipment or shift pattern.

How OrbitX approaches this

In a Connected Floor Audit we reconcile the definitions before we look at the figures: how each line defines planned time, which cycle times are used, and how rework is counted. Only then do we recommend what to automate. Where the existing number is sound, we say so. Where it is not, the first deliverable is a definition that operations, quality and finance can all sign.