OEE: Calculation, Benchmarks and What the Number Misses

What OEE stands for and what it measures

OEE stands for Overall Equipment Effectiveness. It answers one question: of the time you planned to produce, how much produced good parts at full speed?

It is the product of three ratios:

• **Availability** — did the equipment run when it was supposed to?
• **Performance** — did it run at its designed speed while running?
• **Quality** — were the parts it made good first time?

OEE = Availability × Performance × Quality. Because the factors multiply, the number falls quickly: three respectable-looking rates of 90%, 90% and 95% give 77%, not 92%. That multiplication is deliberate — it reflects that losses compound rather than average.

The three factors, precisely

The definitions matter more than the formula, because most disagreements about OEE are really disagreements about what goes in the denominator.

**Availability** = run time ÷ planned production time. Planned production time is the scheduled time minus planned stops — breaks, planned maintenance, no-demand periods. Run time is what remains after unplanned stops: breakdowns, changeovers, material shortages, waiting for an operator.

**Performance** = (ideal cycle time × total count) ÷ run time. Ideal cycle time is the machine's designed rate, not the rate you usually achieve. This factor captures small stops and slow running — the losses that never appear in a downtime log because nobody records a 40-second jam.

**Quality** = good count ÷ total count. First-pass only. A part that was reworked into conformance still consumed the capacity of a defect.

A worked calculation

A shift runs 480 minutes. Planned breaks take 30 minutes and planned maintenance 20, so planned production time is 430 minutes. During the shift a breakdown costs 25 minutes and a changeover 45, leaving 360 minutes of run time. The line produced 600 parts, of which 570 were good first time. Ideal cycle time is 30 seconds, so 2 parts per minute.

There is a useful cross-check: good parts divided by the theoretical maximum in planned time. 570 ÷ (430 × 2) = 66.3%. If your two methods disagree, one of the inputs is defined inconsistently — usually the ideal cycle time, or what counts as planned.

Planned downtime is where most OEE arguments start

OEE deliberately excludes planned downtime from the denominator. Break time, planned maintenance and unscheduled shifts do not count against you.

This is reasonable — you cannot fault a machine for not running when nobody asked it to — but it also makes OEE trivially inflatable. Move a recurring stop from "unplanned" to "planned maintenance" and availability rises without anything improving on the floor. Schedule shorter production windows and OEE improves while output falls.

Two defences. First, write down the planned-stop rules and keep them fixed; the number only means anything over time if the denominator does not move. Second, track TEEP alongside OEE (below) — it puts the planned time back in and shows what the plant is actually doing with its capacity.

One related trap: a changeover is normally unplanned downtime, and it belongs there. Reclassifying changeovers as planned is the single most common way an OEE figure becomes flattering and useless at the same time. If changeover is hurting your number, that is the number working correctly — see the SMED guide.

OEE, OOE and TEEP

Three metrics, three different denominators, and confusing them is why plants compare figures that were never comparable.

Using the earlier example: if the plant runs one shift, five days a week, planned time is 2,400 minutes out of 10,080 calendar minutes, so utilisation is 23.8%. TEEP = 66.3% × 23.8% = 15.8%.

That number is not a failure — it is the answer to a different question. A plant with 66% OEE and 16% TEEP does not have an equipment problem; it has spare capacity. Knowing which of the two you are looking at determines whether the right investment is improvement or a second shift.

Where the 85% benchmark actually comes from

The 85% "world class" figure is quoted constantly and understood rarely. It originates in Seiichi Nakajima's TPM work as a target composed of three component targets: 90% availability × 95% performance × 99% quality = 84.6%.

Two things follow from that, and both matter more than the headline number.

First, it is a **target**, not an observed industry average. It was proposed as what a well-run discrete manufacturing asset should aim at, not as a description of what plants typically achieve.

Second, the component targets tell you where the tolerance lies. The quality target is 99% — meaning the model assumes defects are essentially solved before availability and performance are attacked. A plant at 92% quality cannot reach 85% OEE no matter how well its maintenance performs: 0.90 × 0.95 × 0.92 = 78.7% is the ceiling.

So an OEE target inherited without its components is close to meaningless. Decompose it or do not use it.

What a realistic target looks like

Published "typical" OEE figures by industry circulate widely and are worth very little: they rarely state their planned-time rules, their sample, or whether changeovers were counted. Two plants reporting 65% may be measuring different things entirely.

A more useful approach is internal and arithmetic:

1. Measure your current OEE with fixed, documented rules for at least four weeks. That is your baseline — not a benchmark you found.
2. Decompose it. If you are at 66%, find out whether it is 84/83/95 as in the example above or 95/75/93, because those two need completely different projects.
3. Set the target on the **weakest factor**, not on the OEE number. Moving performance from 83% to 90% in the worked example lifts OEE from 66.3% to 71.6% — and it is a specific, assignable piece of work, unlike "raise OEE by 5 points".
4. Re-baseline whenever the product mix or the planned-time rules change, and say so, rather than letting a step change look like an improvement.

The comparison worth making is against your own line last quarter. Comparisons against a number from an article are, at best, motivational.

The six big losses

OEE's three factors map onto six classical loss categories, which is what makes the metric actionable rather than merely descriptive:

Minor stops are the ones that hide. They are individually too short to log, collectively large, and they appear only in the performance factor — which is why a plant that tracks only downtime tickets is often mystified by its own OEE.

Collecting the data

OEE is only as trustworthy as its inputs, and the inputs are unusually easy to get wrong.

**Manual collection** — operator log sheets — is cheap and immediate but systematically under-reports minor stops, because a person cannot write down a 30-second jam while clearing it. It also tends to record what happened, not how long it lasted.

**Automated collection** from PLC or machine signals catches the small stops that manual logging misses, but it records that the machine stopped, not why. Without a reason code entered by someone who was there, you get an accurate number and no cause.

Most workable systems are hybrid: automatic timing, manual reason codes, and a fixed rule set for planned time. Whatever you choose, freeze the definitions before you start collecting. Changing what counts as planned halfway through a baseline destroys the baseline.

Where OEE stops seeing the loss

OEE is an equipment metric. That is not a criticism — it is the scope it was designed for — but it has consequences that get overlooked when OEE is used as a general productivity KPI.

On a manual assembly line, OEE says very little. There is no machine cycle to compare against, so availability and performance become artefacts of how you define an "ideal rate" for a person. Plants that impose OEE on manual areas usually end up measuring their own assumptions.

Even on automated lines, OEE is blind to the manual work around the machine. If the operator walks 9 seconds to a rack and searches 5 seconds for a gauge every cycle, the machine may be running perfectly and OEE will report a healthy number — while a quarter of the operator's cycle is waste. That loss shows up in labour cost and in the line's ability to meet takt, not in OEE.

This is why OEE and work measurement are complements rather than alternatives: OEE tells you how well the equipment ran, and a Yamazumi or standard work analysis tells you what the people around it were doing. Plants that improve only what OEE measures tend to arrive at excellent machine utilisation and an unchanged labour cost per unit.

Common mistakes

• **Comparing OEE between lines or plants** with different planned-time rules. Almost every cross-site OEE comparison is invalid for this reason alone.
• **Reclassifying stops as planned** to lift the number. It works, and it destroys the metric's usefulness permanently.
• **Using an achievable rate as the ideal cycle time.** This makes performance look near-perfect and buries speed loss.
• **Tracking OEE without decomposing it.** The composite number cannot be assigned to anyone; the factors can.
• **Chasing OEE on a non-constraint machine.** Improving a machine that is not the bottleneck raises its OEE and produces nothing extra — and often produces excess inventory.
• **Ignoring minor stops** because they are not in the downtime log. They are usually the largest single performance loss.
• **Applying OEE to manual work** and treating the result as a productivity measure.

Where to start

1. Pick one line — the constraint, if you know it. OEE across a plant is an average that hides everything.
2. Write the rules: what is planned time, what is a changeover, what is the ideal cycle time for each product, what counts as good first time.
3. Measure four weeks without trying to improve anything. Improvement during baselining makes the baseline useless.
4. Decompose and rank the three factors. The weakest one is the project.
5. Attack the specific loss, not the composite metric. Changeover loss goes to SMED; minor stops go to root-cause work at the machine; quality loss goes upstream of the machine more often than to the machine itself.
6. Re-measure and compare like with like — same rules, same products, same definitions.

The honest summary

OEE is a good metric doing a specific job: quantifying how much of your planned equipment time turned into good product at speed. It is precise, decomposable and hard to argue with when the rules are fixed.

It is not a productivity metric, it is not comparable across sites without identical definitions, and its famous benchmark is a target with components rather than an industry average. Most importantly, it measures the machine — and in most plants a substantial share of the recoverable time is standing next to it.