OEE Formula: Availability, Performance and Quality

If you ask ten plants for their OEE you will get ten numbers, and if you ask the same ten plants for their definitions you will understand why the numbers cannot be compared. The formula itself is trivial and has been stable for decades. The inputs are where the judgement lives, and the inputs are almost never published alongside the figure. This guide walks through each factor, the input choices that move it, and a full worked calculation from raw shift data. Then it recomputes the same shift under a second, equally defensible set of definitions, because that contrast is the most useful thing anyone can show you about OEE.

The message up front: an OEE figure without its definitional record is not a measurement, it is a claim. The same shift of real production data can legitimately produce results tens of points apart depending on how you scope planned production time, which ideal cycle time you adopt, and how you treat rework. Publish the definitions with the number or the number means nothing, including to you.

1. The formula: three fractions and one multiplication

Overall equipment effectiveness is the product of three factors, each a ratio between zero and one:

Two structural properties are worth noticing before you calculate anything. First, the factors are sequential, not independent: performance is measured only across the run time that availability already established, and quality is measured only across the units that performance already produced. Time lost to a breakdown is not also counted as a performance loss. This is what stops the metric from double counting, and it is also why you cannot mix and match factors computed by different people on different scopes.

Second, because it is a product, OEE is dominated by its worst factor. Three factors at 90 percent give roughly 73 percent, which surprises people who expect something in the nineties. That compression is the point: it makes the cumulative cost of small losses visible rather than letting each one hide behind an individually respectable figure.

There is an equivalent single-line form worth knowing as a cross-check: OEE also equals (Ideal Cycle Time x Good Count) / Planned Production Time. If that disagrees with your three-factor result you have an inconsistency in the inputs, usually a run time or a count scoped differently in two places.

If you want the concept, the six big losses and the argument about what OEE is for rather than how to compute it, that belongs with the OEE explained pillar. This article stays on the calculation.

2. What ISO 22400-2:2014 settles, and what it does not

OEE is a defined key performance indicator in an international standard. ISO 22400-2:2014, "Automation systems and integration: Key performance indicators for manufacturing operations management, Part 2: Definitions and descriptions", defines OEE along with a large family of related manufacturing KPIs, and it is the document to reach for when you want a formal reference. It is published by ISO, it is paywalled, and it is currently under revision as ISO/DIS 22400-2, so always cite it with the year you are working from.

What the existence of that standard does not give you is a single universally agreed formula. Three things are true at once and you need all three:

• OEE is defined in ISO 22400-2:2014, with named time elements and named ratios. Anyone telling you OEE has no standard is wrong.
• The ISO definition diverges from Nakajima's original TPM formulation, which is where OEE came from and which most plant-floor practice still descends from. The two are not interchangeable line for line, particularly in how time buckets are named and nested.
• Peer-reviewed criticism has called the standard's treatment imprecise, and in the field practitioners compute OEE inconsistently regardless of which reference they claim to follow.

So the honest position is this: a standard exists, it is worth adopting deliberately, and adopting it does not make your number comparable to anyone else's unless they adopted the same interpretation and documented it. That is exactly why the two-ways calculation later in this article matters. For the wider TPM context OEE sits inside, see the TPM complete guide and the eight pillars of TPM.

You can confirm the designation, title and current revision status directly on the ISO catalogue entry for ISO 22400-2 , or browse the ISO site for the surrounding 22400 series. Check the catalogue rather than a secondary summary, because the revision status changes.

3. Availability: the denominator is the whole argument

Availability looks like the simplest factor and is often the most contested, because the numerator is usually not in dispute and the denominator regularly is.

Run time is the clock during which the equipment was actually producing. Almost nobody argues about that. Planned production time is the scheduled window you are holding the equipment accountable for, and every calendar minute you move in or out of it changes the answer. The decisions you have to make explicitly:

• Calendar time or scheduled time. Start from calendar time and you are measuring the utilisation of the asset as an investment. Start from scheduled shifts and you are measuring the equipment while you intended to run it. OEE conventionally uses the second, so unscheduled shifts, weekends and plant shutdown sit outside the denominator entirely.
• Breaks and meal periods. If the line genuinely stops for a contractual break, most implementations exclude that time. If the line is designed to keep running through breaks with relief operators and it stopped anyway, that is a loss and belongs inside. The mistake is applying the convention without checking which situation you are in.
• Planned maintenance. Excluding scheduled PM is defensible and common: you deliberately gave that time up. Including it is also defensible, because PM downtime is real capacity you cannot sell. Whichever you choose, PM that overruns its window should be split: the planned portion out, the overrun in.
• Changeovers and setup. This is the single biggest lever and the one most often quietly pulled. Excluding changeover time can lift availability by a large margin on a high-mix line, and it makes changeover reduction invisible in the metric. TPM practice treats setup and adjustment as a loss to be attacked, so the defensible default is to keep changeovers inside the denominator as downtime.
• Unplanned stoppages. Breakdowns always count as downtime inside the denominator. That part is not negotiable in any standard interpretation.
• Minor stops and the stop threshold. You need a documented minimum duration below which a stoppage becomes a performance loss rather than a downtime event, because a two-second jam is not realistically logged. Where you set that threshold shifts loss between availability and performance without changing OEE overall: fine for the total, misleading for the diagnosis, so record it.
• External starvation and blocking. Upstream material shortage or a blocked downstream buffer is not an equipment fault. Some organisations exclude it as an external loss, some include it because the capacity was still lost. Excluding it is the more generous reading and has to be declared.

Note also that the availability factor inside OEE is not the same quantity as the availability that reliability engineering computes from uptime and downtime over a long horizon. They share a name and a shape and answer different questions. The relationship is covered in the reliability metrics pillar, which has a section on where availability sits inside OEE. Use that rather than trying to reconcile the two here.

4. Performance: which cycle time is the ideal one

Total count is every unit the equipment produced during run time, good and bad together. Run time you have already established. So the entire content of this factor is the choice of ideal cycle time, and there are three candidates:

• Nameplate or design cycle time: what the manufacturer says the machine can do. This is the fastest of the three, and because it sits in the numerator, adopting it produces the lowest performance figure of the three. It is the conservative choice, not the flattering one. This trips people up constantly, so be clear about the direction: a shorter ideal cycle time gives a smaller performance number.
• Best demonstrated sustained rate: the fastest rate this specific machine has actually sustained over a meaningful period with this product and this material. This is the choice I would recommend in most plants. It is achievable by definition, so the resulting gap is a real gap rather than an argument with the vendor's datasheet, and it does not let the machine off the hook the way a negotiated standard does.
• Current standard or historical average rate: the rate in the routing or the planning system. This is usually the slowest of the three and therefore produces the highest performance figure. It is the flattering choice, and it has the additional problem that standards drift: once a standard is relaxed to match current output, performance returns to near 100 percent and the metric stops detecting speed loss at all.

Two further decisions belong here. Cycle time must be per product, not per machine, on any line running a mix: a single blended ideal cycle time will move your performance figure by whatever the mix happened to be, which makes shift-to-shift comparison meaningless. And if minor stops below your threshold are absorbed into performance rather than logged as downtime, say so, because that is the difference between performance meaning speed loss and performance meaning speed loss plus micro stoppages.

5. Quality: first-pass yield and the rework question

The intended meaning of this factor is first-pass yield: units that came off the equipment right the first time, with no further intervention. That is the definition I would recommend and it is the one that makes the metric useful, because the cost of a defect is mostly the cost of handling it again.

The decisions that move it:

• Does rework count as good? Under first-pass yield it does not. A reworked unit consumed additional labour, time and possibly material, and treating it as good hides all of that. This is the most common single inflation in a quality factor, and it is often not deliberate: the count simply comes from a warehouse receipt after rework rather than from the machine.
• Startup and changeover scrap. Units lost while the process settles after a start or a product change are a genuine loss, and TPM practice treats reduced yield at startup as a loss category in its own right. Excluding them because they are "expected" removes the incentive to shorten the ramp.
• Where is the count taken? A count at the machine and a count after a downstream inspection station are different numbers. If defects are detected two operations later and never attributed back, the quality factor will look better than it is. Decide where the boundary sits and hold it.
• Concessions and downgrades. A unit shipped on concession, or sold as a lower grade, is not a first-pass good unit even though it generated revenue.
• Partial and scaled units. On continuous or weight-based processes, define the unit before you define the yield. Mass, length or batch counts all work, provided total count and good count use the same unit.

6. The three factors side by side

This is the table I would put on the wall next to any OEE board, because it names the single input that moves each factor most:

7. A full worked example from raw shift data

All figures below are illustrative and invented for this article. They are not drawn from any client, plant or dataset. Use them to follow the arithmetic, not as any kind of reference point.

Raw data from one shift on a single packaging machine:

Definition Set A, the strict reading I would recommend as a default. Planned production time excludes contractual breaks only. Changeover, PM, breakdown, minor stops and starvation are all downtime inside the denominator. Ideal cycle time is the best demonstrated sustained rate. Quality is first-pass yield, so rework does not count as good.

Cross-check with the single-line form: (0.22 x 1,368) / 450 = 300.96 / 450 = 0.6688. The two agree, which tells you run time, counts and planned production time were scoped consistently. That check takes fifteen seconds and is worth doing every time.

Reading the result diagnostically, availability is the weakest factor and within it the largest single element is the 25 minute breakdown, followed by the 20 minute changeover and the 20 minutes of starvation. Two of those three are attacked by different teams: the breakdown by maintenance, the starvation by planning and materials. That split is the real value of computing the factors separately rather than reporting one headline. Tracking the downtime itself is a prerequisite, and the practical side of that is covered in the backlog and downtime tracking pillar.

8. The same shift, computed a second defensible way

Now take exactly the same raw data and apply Definition Set B. Nothing about the shift changes. Not one minute, not one unit. Only the definitions change, and every one of these choices is in common use and can be argued for in good faith:

• Planned production time excludes breaks and scheduled changeover and scheduled PM, on the grounds that all three were deliberately given up and are not equipment losses.
• Material starvation is excluded as an external loss, on the grounds that it is an upstream supply failure rather than a fault of this machine.
• Ideal cycle time is the current routing standard, on the grounds that it is the rate the business plans and commits to.
• Reworked units are counted as good, on the grounds that they were ultimately shipped as conforming product.

Notice which set is more useful operationally. Set A shows 90 minutes of loss and points at a breakdown, a changeover and a materials problem. Set B shows 40 minutes of loss and points at almost nothing, because the three losses a plant could most readily attack have been defined out of the measurement. A lower, stricter number that names its losses is a better management instrument than a higher number that hides them.

9. How the number gets gamed

Most OEE inflation is not fraud. It is a series of individually reasonable decisions, each made by someone who is measured on the result, accumulating in one direction. The patterns to watch for:

• Reclassifying unplanned stoppages as planned downtime. A breakdown that is logged after the fact as "planned maintenance" leaves the denominator entirely and lifts availability twice over: the lost minutes disappear and so does the downtime. Look for PM events created retrospectively, or PM durations that vary suspiciously with how the shift went.
• Quietly excluding changeovers. The single largest lever on a high-mix line. Often introduced with a genuine argument, then never revisited, and the effect is that setup reduction work produces no visible improvement in the metric because setup is not in the metric.
• Resetting the ideal cycle time. If performance is persistently low, relaxing the standard cycle time fixes the number without touching the machine. Standards that drift toward actual output turn performance into a near-constant 100 percent, which is the clearest sign the metric has stopped working. Freeze the ideal cycle time, version it, and require a documented engineering reason plus a dated change record for any revision.
• Raising the minor stop threshold. Lifting it from one minute to five moves loss from availability into performance. The OEE total is unchanged, so this is less about inflation and more about hiding which category the loss belongs to, which quietly protects whoever owns availability.
• Broadening "external" losses. Starvation, blocking, utility interruptions, labour shortages and quality holds all get classified as not the equipment's fault. Individually arguable, collectively they can remove most of the real loss from the measurement.
• Counting rework as good. Usually accidental, because the good count is sourced from a downstream receipt rather than the machine. It still removes the rework loss from view.
• Measuring only the good hours. Excluding shifts where the line barely ran, or the ramp after a shutdown, produces a number that describes the machine's best behaviour rather than its behaviour.

10. The minimum definitional record to publish with any OEE figure

This is the practical output of everything above. Any OEE number you publish, internally or in a contract report, should carry this record. It fits on one page and it is what makes the figure auditable and comparable over time:

• Scope and period. Which machine, line or cell; which shifts; which date range. State whether the figure is a single period or an average, and if an average, whether it is weighted by planned production time.
• The denominator rule. Exactly what is in and out of planned production time, category by category: unscheduled shifts, breaks, scheduled PM, changeover and setup, trials and validation runs, training, external starvation and blocking, utility interruptions.
• The minor stop threshold. The duration in seconds below which a stoppage becomes a performance loss rather than a downtime event.
• The ideal cycle time basis. Nameplate, best demonstrated sustained, or current standard; the value per product; the date it was set and by whom; and the change control rule.
• The good unit definition. Whether rework counts as good, whether startup and changeover scrap is included, whether concessions count, and where physically the count is taken.
• The unit of count. Pieces, cases, kilograms, metres or batches, used identically in total count and good count.
• Data source and capture method. Automatic from the controller or historian, manual from an operator log, or a mix; and for manual capture, who enters it and when.
• Version and change log. A dated record of every definitional change, because a step change in OEE that coincides with a definition change is not an improvement and everyone will forget that within a quarter.

On systems: this record is configuration, not a spreadsheet note, and it should live wherever the calculation runs. Most CMMS, MES and historian platforms can hold downtime reason codes, a planned-or-unplanned flag and a standard rate per product; whether they enforce the definitions consistently is a configuration question you should test with a known shift before you trust any dashboard. Where downtime reasons and PM records come from a maintenance system, the same coding discipline that makes PM reporting trustworthy applies here, and that is covered in the PM KPIs and schedule compliance pillar. If you are choosing the system that will hold this data, the CMMS buyer's introduction is the right starting point.

11. Why cross-company benchmarking does not work, and what to do instead

The two-ways calculation settles this. If definitions alone can move the same shift by more than seventeen points, any comparison between two organisations is comparing definitional choices at least as much as manufacturing capability. Add differences in product mix, asset age, process type and batch size, and an external OEE comparison carries essentially no usable signal.

I deliberately publish no target or benchmark figure for OEE, and I would treat any you are offered with suspicion. A figure presented as a universal standard of excellence cannot survive the definitional variation described above, and in practice it functions as a number to be reached by adjusting definitions rather than by improving operations.

What does work:

• Your own baseline, measured under frozen definitions. Establish the definitional record, compute several periods, and treat that as the starting point. The only OEE comparison that means anything is this machine against itself.
• Trend, not level. Direction of travel over months is robust to definitional choice as long as the definitions have not changed. The absolute level is not.
• Loss minutes and units, reported underneath the ratio. Ninety minutes of downtime split by reason, and thirty-six scrapped units, are directly actionable in a way that 66.9 percent is not. The percentage is the summary; the minutes are the work.
• Factor-level tracking. Report availability, performance and quality separately and permanently. A flat OEE hiding a falling availability offset by a relaxed cycle time is exactly the failure mode you are trying to avoid.
• Internal comparison across identical assets only. Two identical lines in the same plant, running the same products under the same definitions, can legitimately be compared. Nothing looser than that.

The same caution applies to the reliability metrics OEE sits beside. If you are building a metric set, the companion calculation guides are the MTBF formula guide and the MTTR formula guide, both of which have the same characteristic: the arithmetic is simple and the clock definitions decide the answer.

The idea to walk away with

OEE is not a hard calculation. Availability is run time over planned production time, performance is ideal cycle time times total count over run time, quality is good count over total count, and you multiply the three. Anyone can do that in a minute. The skill is in the seven or eight input decisions that sit underneath, and in being willing to make the stricter choice on each one even though it produces a lower number.

A plant reporting 66.9 percent with a published definitional record, ninety minutes of classified loss and three named owners is in a far better position than a plant reporting 84.2 percent on the same shift with nothing written down. The first number can be improved. The second can only be defended.

Final thoughts

The advice I would give anyone standing up an OEE measurement for the first time: write the definitions before you write the formula. Decide, in writing and with the people who will be measured by it, what counts as planned production time, which ideal cycle time you are adopting and why, where the stop threshold sits, and whether rework is good. Then compute one shift by hand, exactly as in section 7, and reconcile it against whatever your system reports. If they disagree, the definitions in the system are not the definitions you agreed, and you have just found that out cheaply.

After that, freeze it, version it, and resist every request to adjust a definition in a period where the number matters. OEE earns its place as a diagnostic that concentrates attention on real loss. It loses that place the moment the definitions become negotiable, and it is far easier to hold the line at the start than to recover credibility after a step change nobody can explain.

Related reading: OEE explained: the concept and the six big losses, Reliability metrics: MTBF, MTTR and availability, MTBF formula: how to calculate MTBF, MTTR formula: how to calculate MTTR, Total productive maintenance: complete guide, Maintenance backlog and downtime tracking.