How to calculate OEE (and why most OEE numbers are wrong)

Overall equipment effectiveness answers one question with one percentage: of the time you set aside to make good product, how much actually produced good product at the rate the equipment was designed to run? It is a genuinely useful number, and one of the most commonly mis-stated figures on a shop floor — because almost every way of getting it wrong makes it look better rather than worse. This guide works the calculation through with numbers you can check, then sets out the four ways an OEE figure becomes wrong while staying entirely plausible.

What OEE actually measures

OEE is the product of three ratios, each describing a different loss. Availability is time you meant to be running but were not. Performance is running slower than the equipment is capable of. Quality is output you made but cannot sell.

That it is a product rather than an average matters, because losses compound. A line available 90% of the time, running at 90% of design rate, yielding 95% good product is not at 91.7%; it is at 76.95%, because the quality loss applies only to what the performance loss let through, which applies only to the time the availability loss left. Three respectable-looking factors routinely multiply out to a headline that surprises people, and that surprise is the point of the measure.

The three factors and their formulas

These are the calculations exactly as they are published on the tool page, in the order they are worked:

• Run time = planned production time − unplanned downtime
• Availability % = run time ÷ planned production time × 100
• Performance % = (ideal cycle time × total units) ÷ (run time × 60) × 100
• Quality % = good units ÷ total units × 100
• OEE % = Availability × Performance × Quality ÷ 10000

Three definitions carry most of the weight. Planned production time is the time you intended to make product, and the meaning of availability sits entirely in where you draw the line between time never scheduled and time scheduled then lost. Ideal cycle time is the time one unit takes at the rate the equipment was designed to achieve for that product — the theoretical fastest, not the practical usual; the multiplication by 60 is a unit conversion, since ideal cycle time is in seconds and run time in minutes. Good units are units that passed first time, because rework consumed exactly the time the measure exists to expose.

A worked example you can follow line by line

One shift, one line, one product.

• Shift length: 480 minutes
• Scheduled non-production time: 80 minutes (two 20-minute breaks and a 40-minute meal break, line stopped and unattended)
• Planned production time: 480 − 80 = 400 minutes
• Unplanned downtime: a 25-minute conveyor failure plus 15 minutes waiting for material = 40 minutes
• Ideal cycle time: 12 seconds per unit (the design rate — 5 units per minute)
• Total units produced: 1,620; good units: 1,539 (81 rejected at end-of-line inspection)

Run time = planned production time − unplanned downtime = 400 − 40 = 360 minutes. Availability % = run time ÷ planned production time × 100 = 360 ÷ 400 × 100 = 90.0%.

Performance % = (ideal cycle time × total units) ÷ (run time × 60) × 100. The numerator is 12 × 1,620 = 19,440 seconds of work at the design rate. The denominator is 360 × 60 = 21,600 seconds of actual run time. So 19,440 ÷ 21,600 × 100 = 90.0%. Read that as: in 360 minutes at 5 units per minute the line could have made 1,800 units, and it made 1,620.

Quality % = good units ÷ total units × 100 = 1,539 ÷ 1,620 × 100 = 95.0%.

OEE % = Availability × Performance × Quality ÷ 10000 = 90 × 90 × 95 ÷ 10000. Working it through: 90 × 90 = 8,100; 8,100 × 95 = 769,500; 769,500 ÷ 10,000 = 76.95%.

One cross-check catches most arithmetic slips. Over 400 minutes of planned production time at 5 units per minute, the line could have delivered 2,000 good units. It delivered 1,539, and 1,539 ÷ 2,000 = 76.95% — the same answer. OEE is, in the end, good units divided by the good units that perfect time, speed and quality would have produced.

The four ways OEE numbers go wrong

None involves bad arithmetic. Each is a defensible-sounding decision about what to count, and each pushes the number up.

1. Averaging row OEEs instead of summing the components

The most common error, and the largest. A month of production is a list of runs, each with its own OEE, and the instinct is to average the percentages. That treats every run as equally important, and runs are not equally long. Take two runs on the same line, with the same 12-second ideal cycle time:

• Run A — a 30-minute trial. Planned production time 30 minutes, no downtime, so availability is 100%. Design output at 5 units per minute is 150; it made 150, so performance is 100%. All 150 good, so quality is 100%. OEE = 100 × 100 × 100 ÷ 10000 = 100%.
• Run B — an 8-hour run. Planned production time 450 minutes, unplanned downtime 90, so run time 360 and availability 360 ÷ 450 × 100 = 80%. Design output over 360 minutes is 1,800; it made 1,440, so performance is 80%. Good units 1,368 of 1,440, so quality is 95%. OEE = 80 × 80 × 95 ÷ 10000 = 60.8%.

Average the two row percentages and you get (100 + 60.8) ÷ 2 = 80.4%. Now sum each component across both runs before dividing. Planned production time = 30 + 450 = 480 minutes; run time = 30 + 360 = 390, so availability = 390 ÷ 480 × 100 = 81.25%. Total units = 1,590, so performance = (12 × 1,590) ÷ (390 × 60) × 100 = 19,080 ÷ 23,400 × 100 = 81.54%. Good units = 1,518, so quality = 1,518 ÷ 1,590 × 100 = 95.47%. Multiply: 81.25 × 81.54 × 95.47 ÷ 10000 = 63.25%.

The cross-check confirms it: 480 minutes at 5 units per minute is 2,400 possible good units, and 1,518 ÷ 2,400 = 63.25%. The row average said 80.4%. A thirty-minute perfect trial carried the same weight as a run fifteen times longer, and moved the reported number by more than seventeen points.

2. A planned production time that quietly excludes the losses

Availability is a ratio and both halves are yours to define, so the temptation is to shrink the denominator. A breakdown becomes “unplanned maintenance, so the line was not scheduled”. Waiting for material becomes “a supply issue, not a line loss”. Each reclassification is arguable in isolation, and each deletes exactly the loss the factor exists to reveal.

Apply that to the worked example. Move the 25-minute breakdown and the 15-minute material wait out of planned production time, on the grounds that the line “could not have run anyway”. Planned production time falls from 400 to 360 minutes, run time is still 360, so availability becomes 100%. OEE = 100 × 90 × 95 ÷ 10000 = 85.5%. The shift produced the same 1,539 good units and the headline rose more than eight points.

The workable rule: time is outside planned production time only if the equipment was never scheduled to produce during it — no shift on, no demand, a shutdown agreed in advance. Everything else stays inside, including changeovers. Changeover is one of the six big losses precisely because it is improvable; excluding it as “planned” removes the largest lever most lines have.

3. An ideal cycle time set to the average achieved rather than the design rate

Ideal cycle time is a property of the equipment and the product, not of last month. When nobody can find the design rate, somebody eventually derives it from what the line usually does — and the arithmetic shows the circularity immediately. In the worked example the line ran 360 minutes and made 1,620 units, so the average achieved cycle time was 21,600 ÷ 1,620 = 13.33 seconds per unit. Enter 13.33 as the ideal cycle time and performance becomes (13.33 × 1,620) ÷ 21,600 × 100 = 100.0%, by construction. OEE = 90 × 100 × 95 ÷ 10000 = 85.5%. A performance factor derived from achieved output can only ever report 100%, so it measures nothing.

This is also the diagnosis for the most common OEE anomaly. If performance comes out above 100%, the equipment did not exceed its own design: the ideal cycle time is too slow, the unit count is wrong, or the run time is understated. That is an input error to correct, not good news to explain away. Note too that ideal cycle time is per product, not per machine — a single blended figure across a fast product and a slow one moves the headline with the mix.

4. Chasing the 85% figure instead of the losses

An 85% OEE is widely quoted as a world-class target, usually alongside roughly 90% availability, 95% performance and 99% quality. As a description of an excellent discrete line that is not unreasonable. As a target it does more damage than good, because it is easier to reach by moving a definition than by improving a process.

Notice what happened in the two previous sections. Redefining planned production time gave 85.5%; deriving ideal cycle time from achieved output gave 85.5%. Two adjustments, no extra good parts, both landing on the target. The shift still made 1,539 good units against a capability of 2,000, and that 461-unit gap — 40 minutes of downtime, 180 units of speed loss, 81 rejects — is what a real improvement plan is made of. The figure is also not comparable between sites, or usually between lines. Compare a line to itself under a definition that has not changed, and treat the absolute value as scenery.

The six big losses, mapped to the three factors

OEE is actionable because each factor decomposes into named, countable losses — commonly six, two per factor:

• Availability — breakdowns and equipment failure: the stop is long enough to be logged.
• Availability — setup, changeover and adjustment: last good unit of one product to first good unit of the next, including the fiddling at the end that nobody records.
• Performance — idling and minor stops: jams, misfeeds, sensor trips and short waits, usually under a few minutes and usually not logged. The largest invisible loss on most lines, because it shows up only as a performance factor that will not improve.
• Performance — reduced speed: the line runs below design rate because of wear, a cautious setting, material variation, or a habit nobody has revisited.
• Quality — process defects and rework during steady running.
• Quality — startup and yield losses: product scrapped while the process settles after a start or a changeover. Frequent short runs multiply this, which is why it travels with the changeover loss.

The mapping is why you calculate the factors separately rather than jumping to the headline. If availability is worst, your problem is stops and changeovers. If performance is worst, it is speed and micro-stops, and the fix is usually found standing at the line. If quality is worst, the fix is often upstream of the machine.

What to do first

1. Write the definitions on one page and date it: shift length, what counts as never scheduled, what counts as downtime, the design rate per product, what counts as a good unit. Almost every OEE dispute is a definition dispute in a numerical costume.
2. Fix the ideal cycle time first, from the design rate for the product being run. If performance exceeds 100%, correct the input rather than explaining the result.
3. Record losses as minutes against a named reason, at the time. Reconstructing them later produces fiction.
4. Aggregate by summing components, then run the factors once on the totals. Never average row percentages.
5. Rank losses in minutes rather than percentage points and work on the largest. Minutes are what people can go and look at.
6. Leave the headline alone for a quarter. If loss minutes are falling it will follow; manage it directly and the definitions will move instead.

OEE done honestly is uncomfortable at first: a line everyone believed was running well often lands in the fifties the first time the losses are counted properly. Resist the urge to adjust a definition, and the number stops being a score to defend and becomes a running total of the minutes and units you could get back.