What OEE measures, and what it deliberately ignores
OEE answers one question: of the time you planned to produce, what fraction produced good parts at full speed? Everything else is subtracted. It was formalised by Seiichi Nakajima as the central metric of Total Productive Maintenance, and it survives because it collapses three unrelated failure modes into one comparable number without hiding which of them caused the loss.
The three rates are deliberately independent. Availability is a time question — was the equipment running when it was supposed to be? Performance is a speed question — while it ran, did it run at its designed rate? Quality is a yield question — of what it made, how much passed first time? Multiply them and each loss compounds, which is why an operation with three respectable 90% rates lands at 0.9³ = 72.9%.
What OEE ignores is just as important. Planned stops are excluded from the denominator entirely, so OEE says nothing about whether you scheduled enough shifts. That is TEEP's job. OEE also says nothing about whether the parts you made were needed; a line can run at 95% OEE building inventory nobody ordered. Pair it with takt time to answer the demand question, because OEE alone will happily reward overproduction.
How each rate is built, and the definitions that trip people up
Availability = run time ÷ planned production time. Planned production time is the shift minus stops production was never expected to make: breaks, scheduled maintenance, a shift you did not staff. Run time subtracts the stops it was expected to avoid: breakdowns, changeovers, material starvation, waiting for an operator. The line between the two categories is a management decision, and it is the single most common way OEE numbers become incomparable between plants. Write it down and stop moving it.
Performance = (ideal cycle time × total count) ÷ run time. Ideal cycle time is the fastest cycle the equipment can sustain, not the rate on the routing and not the historical average. Using a soft standard is how a plant reports 95% performance while running two thirds as fast as the machine can go — the loss does not disappear, it just stops being measured. If performance ever exceeds 100%, the standard is wrong by definition, because the machine cannot beat its own theoretical maximum.
Quality = good count ÷ total count. Good means passed first time. Reworked units count as defects, because the equipment time that produced them was spent twice. Counting rework as good is the second most common way OEE gets inflated.
Multiply the three and the run time terms cancel algebraically, leaving OEE = ideal cycle time × good count ÷ planned production time. That identity is worth carrying in your head: it lets you check any reported OEE from two numbers on the production board and the standard cycle, without touching the downtime log.
TEEP extends the denominator to all calendar time. It answers a capital question rather than an operations one: how much of the theoretical 24/7 capacity of this asset is being converted into good parts? A plant running one shift a day cannot exceed a TEEP of about a third no matter how well the shift goes, which is exactly the point — it makes the cost of unscheduled capacity visible before anyone buys a second machine.
Worked example: a 480 minute shift with 47 minutes of downtime
A packaging line runs an eight-hour shift. There are 60 minutes of scheduled breaks, and the downtime log records 47 minutes of unplanned stops. The machine's ideal cycle is 1.0 second per unit. It produced 19,271 units, of which 423 were rejected.
- Planned production time. 480 − 60 = 420 minutes.
- Run time. 420 − 47 = 373 minutes.
- Availability. 373 ÷ 420 = 0.8881, or 88.81%.
- Performance. Run time in seconds is 373 × 60 = 22,380 s. Ideal time for the units made is 1.0 × 19,271 = 19,271 s. So 19,271 ÷ 22,380 = 0.8611, or 86.11%.
- Quality. Good count is 19,271 − 423 = 18,848. So 18,848 ÷ 19,271 = 0.9780, or 97.80%.
- OEE. 0.8881 × 0.8611 × 0.9780 = 0.7479, or 74.79%.
- Check by the identity. Ideal cycle × good count = 1.0 × 18,848 = 18,848 s = 314.13 min. Divided by 420 min of planned production time, that is 0.7479. The two routes agree.
- TEEP over a calendar day. 0.7479 × (420 ÷ 1,440) = 0.2181, or 21.81%.
Now convert the rates into minutes, which is where the improvement conversation actually happens. Downtime loss is the logged 47.00 min. Speed loss is run time minus the ideal time for what was made: 373 − 321.18 = 51.82 min. Quality loss is the ideal time spent on rejects: 1.0 × 423 = 423 s = 7.05 min. Fully productive time is 314.13 min. Those four add to 47.00 + 51.82 + 7.05 + 314.13 = 420.00 minutes, the whole of planned production time.
Read that split rather than the three percentages. Quality is the worst-sounding number to a plant manager because 97.8% feels close to perfect and yet defects always feel expensive — but here quality costs 7 minutes while speed costs 52. The largest single opportunity on this shift is the seven and a half minutes per hour the line spends running below its designed rate, not the scrap.
What number is good, and what to do with it
The figure most often quoted as world class in the TPM literature is 85% OEE, built from roughly 90% availability, 95% performance and 99% quality. Treat it as a target shape rather than a universal benchmark: a discrete assembly cell and a continuous process line have completely different achievable ceilings, and comparing OEE across dissimilar equipment is the fastest way to make the metric meaningless.
Compare against your own trend instead. OEE is at its most useful as a same-machine, same-definition time series, where a step change points at something specific — a new tooling set, a supplier change, a shift pattern. Once definitions move, the series is broken.
Work the minutes, not the percentages. The three rates are proportional and can mislead about magnitude, as the worked example shows. Convert every loss to minutes on the same denominator, rank them, and attack the largest. That is what the loss table on this page does.
Beware the two ways OEE gets quietly inflated. A soft ideal cycle time makes performance look good while hiding speed loss, and classifying a stop as planned rather than unplanned moves it out of the denominator entirely. Both raise the number without changing a single thing on the floor. If your OEE improves and your output does not, one of these is usually why.
Finally, put OEE beside a demand measure. A cell at 90% OEE that is producing faster than takt is building inventory, and a cell at 60% OEE that comfortably meets takt may not be worth improving at all. OEE ranks equipment losses; it does not tell you whether removing them is worth anything.
The six big losses and which OEE factor each one lands in
| Loss | OEE factor | Typical examples | What it looks like in the data |
|---|---|---|---|
| Equipment breakdown | Availability | Tooling failure, motor burnout, unplanned maintenance | Long stops, few in number |
| Setup and adjustment | Availability | Changeovers, warm-up, first-article approval | Stops clustered at product changes |
| Idling and minor stops | Performance | Jams, misfeeds, blocked chute, sensor trips | Many stops too short to be logged |
| Reduced speed | Performance | Running below rated rate for wear, noise or quality | Steady shortfall against the cycle standard |
| Process defects | Quality | Scrap and rework in steady-state running | Reject rate roughly constant over the shift |
| Reduced yield at startup | Quality | Parts scrapped while the process stabilises | Reject spike after every start |
Idling and minor stops land in performance rather than availability because they are usually too short to appear in a downtime log — they show up as units not made rather than as time not run.
Assumptions, limits and common errors
- The planned/unplanned boundary is a policy, not a fact. Two plants can compute honest OEE figures that differ by ten points purely from where they draw it. Document it and freeze it.
- Ideal cycle time must be the machine's fastest sustainable cycle. Using the routing standard or a historical average converts real speed loss into invisible loss.
- Rework is a defect. Counting reworked units as good double-counts the equipment time that produced them.
- OEE is a single-machine metric. Applying it to a whole line means choosing a constraint machine, or defining a line-level cycle time, and both choices change the answer.
- It says nothing about demand. High OEE on the wrong part number is overproduction, which lean explicitly treats as waste.
- Short shifts exaggerate variance. One long breakdown in a four-hour run swamps the number. Aggregate over a week before drawing conclusions.
- Do not average OEE across machines by taking the mean of percentages. Sum fully productive time and sum planned production time, then divide.
OEE, TEEP and the standards that define them
OEE as three multiplied rates comes from Nakajima's work on Total Productive Maintenance and has no single ISO standard behind it, which is why definitions vary between plants. The semiconductor industry did formalise it: SEMI E79, the standard for definition and measurement of equipment productivity, sets out an equipment-state model and a family of efficiency metrics including OEE, and it is worth reading if you need definitions rigorous enough to write into a supply agreement. TEEP extends the same arithmetic to calendar time and is normally attributed to the same TPM tradition rather than to a standards body.
Where OEE sits among the other operations metrics
OEE tells you how much of your scheduled time became good parts. It does not tell you how many parts you need, how capable the process is, or how reliable the equipment is between failures — and each of those has its own tool.
Demand comes first. Takt time converts customer demand into the rate the line must sustain, and comparing takt with your actual cycle time tells you whether an OEE improvement is even needed. A line comfortably inside takt has slack; improving its OEE produces inventory, not throughput.
Capability is a different question again. Quality rate in OEE counts defects after the fact, while process capability predicts them from the spread of the process against its tolerance. A high quality rate with a poor Cpk means you are catching defects rather than preventing them, and that is a warning about the future.
Reliability sits underneath availability. MTBF and MTTR decompose downtime into how often failures happen and how long they take to fix, which is the difference between a maintenance strategy and a spares strategy. Availability is their product; OEE only sees the total.
And on the commercial side, the reason any of this matters is the fixed cost sitting behind the machine. Every minute of lost production time carries the same overhead as a productive one, which is the logic of the break-even point in units — and it is why TEEP, which exposes unscheduled capacity, is usually the more uncomfortable number of the two.