OEE Guide: Formula, the Six Big Losses and a Worked Calculation

OEE (Overall Equipment Effectiveness) condenses into a single percentage the three ways a machine loses capacity: stoppages, reduced speed and defects. This guide derives its formulas from first principles, solves a complete real-world case and dismantles the most common measurement mistakes, with an interactive calculator to verify every step.

What OEE is and why it is the central KPI of industrial efficiency

OEE (Overall Equipment Effectiveness) measures what fraction of planned production time is converted into good output manufactured at the equipment’s ideal speed. It is a percentage, but not just any percentage: it is a composite metric that condenses into one number the three ways a machine loses capacity. It can stop when it should be producing, it can run more slowly than its design allows, and it can produce parts that are not fit for use. OEE multiplies the three factors that capture those leaks.

The formal definition, set out in ISO 22400-2 on key performance indicators for manufacturing operations management, is expressed as the product of three factors:

OEE = Availability × Performance × Quality

It is worth fixing from the outset what it measures and what it does not. OEE does not measure profitability: a line can run at 90% OEE making a product sold at a negative margin. Nor does it measure calendar utilisation: if a plant works a single shift Monday to Friday, OEE does not penalise nights or weekends, because its denominator is the time production decided to plan, not the 8,760 hours of the year. Other indicators exist for that (TEEP and OOE), which we cover later.

Throughout the article we will work with a single example: a beverage bottling line running one 8-hour shift (480 minutes). The line fills, caps and labels bottles with an ideal cycle time of 1.5 seconds per bottle, that is, a nominal capacity of 40 bottles per minute. Each section adds one piece of the calculation and at the end we solve the full case, with the option of verifying it in the interactive calculator.

Origins: TPM, Seiichi Nakajima and the Japan Institute of Plant Maintenance

OEE was not born in an IT department or at a consultancy. It was born on the shop floor. Seiichi Nakajima, of the Japan Institute of Plant Maintenance (JIPM), formalised it as the operational metric of TPM (Total Productive Maintenance), the methodology developed in the sixties and seventies at Nippondenso, a supplier to the Toyota group. The usual academic reference is his book Introduction to TPM (Productivity Press, 1988), the first work to present the concept in English along with the loss taxonomy that underpins it.

Within TPM, OEE is the measuring instrument of the focused improvement pillar (kobetsu kaizen): before attacking a machine’s losses you have to quantify them, and OEE is precisely that quantification. Nakajima conceived it as a diagnostic tool for maintenance and production teams, not as a board-level indicator, and that origin explains several of its properties: it is calculated per machine or per line, it relies on shop-floor records and its value lies in the breakdown, not in the aggregate number. It is the same logic behind any serious continuous improvement programme: measure before you improve.

Since then the indicator has spilled well beyond maintenance. MES platforms compute it continuously from PLC signals, the ISO 22400 standard codified it as a manufacturing KPI, and today it is the headline metric of any plant digitalisation project. The formula, however, is still the one from 1988.

The six big losses and how they map to the three factors

The OEE formula is the consequence of a prior classification: Nakajima’s six big losses. Each loss attacks one of the three factors, and each is captured by a specific shop-floor record.

Two clarifications that head off recurring arguments on the shop floor. First: changeovers count as an availability loss even though they are planned in the everyday sense. Nakajima’s criterion is clear: during a changeover the equipment should be producing and is not, so it is lost time that improvement (SMED techniques, for instance) can reduce. If they are classified as planned downtime, they vanish from OEE and with them the incentive to shorten them.

Second: minor stoppages (a bottle jam in the infeed star wheel, a sensor forcing a reset) usually last seconds and almost never appear on a manual report. They are not logged as stoppages: they show up as missing pieces at the end of the shift, which is why they live in the performance factor, not in availability.

The time cascade: from calendar time to valuable operating time

All the OEE formulas derive from a chained time model. Each level is obtained from the previous one by subtracting a category of loss:

1. Calendar time: 24 hours a day, 365 days a year. It is the physical base; nobody produces more time than exists.
2. Planned production time: calendar time minus unscheduled time (shifts not worked, weekends if not worked) and minus legitimate planned stops (staff breaks, scheduled preventive maintenance, shift meetings). This is the OEE denominator.
3. Operating time: planned time minus unplanned stoppages (breakdowns) and changeovers and adjustments. What remains is time the machine was actually running.
4. Net operating time: operating time minus speed losses (minor stoppages and running below nominal rate). It is the time that would have been enough to make the actual output at ideal speed.
5. Valuable operating time: net operating time minus the time spent making defective pieces. It is the time that produced value.

OEE is the ratio between the last step and the second: valuable operating time divided by planned production time.

With the numbers from our bottling line: the shift lasts 480 minutes. There are 40 minutes of planned stops (a 30-minute break and a 10-minute shift meeting), leaving 440 planned minutes. During the shift, 55 minutes are lost to unplanned stoppages: a 22-minute labeller breakdown and a 33-minute bottle format changeover. That leaves 385 operating minutes. Speed losses eat up 38.5 minutes (we will prove it when calculating performance), leaving 346.5 minutes of net operating time. Defective pieces consume a further 16.5 minutes, so valuable operating time is 330 minutes. Of 440 planned minutes, 330 produced value: 75%.

OEE formulas with a full derivation

With the time model in place, the three factors are simply the ratios between consecutive steps of the cascade:

Availability = Operating time      / Planned production time
Performance  = Net operating time  / Operating time
Quality      = Valuable op. time   / Net operating time

When multiplied, the intermediate terms cancel out:

OEE = (Op. time / Planned time) × (Net op. time / Op. time) × (Valuable time / Net op. time)
    = Valuable operating time / Planned production time

That is the correct reading of OEE: the fraction of planned time that ended up as valuable operating time. The three factors are the decomposition of that fraction into its causes.

There is a second, equivalent form, more practical for checking calculations, that works with pieces instead of times. Since valuable operating time is, by definition, the time it would take to make the good pieces at ideal cycle:

Valuable operating time = Good pieces × Ideal cycle time

substituting into the previous expression:

OEE = (Good pieces × Ideal cycle time) / Planned production time

This compact form serves as a cross-check: if the OEE calculated factor by factor does not match the one calculated this way, there is an error in one of the records. Let us now look at each factor with its variables and data sources.

Availability = operating time / planned production time

Availability = Operating time / Planned production time
Operating time = Planned production time − Unplanned stoppages − Changeovers and adjustments

The delicate variable is planned production time. Only legitimate planned stops are subtracted from it: agreed breaks, scheduled preventive maintenance, absence of demand decided by planning. Everything else (breakdowns, changeovers, waiting for material, missing operator) reduces availability. The temptation to “plan” a stoppage after the fact so it does not count is the most common OEE fraud.

The natural source of the data is the stoppage log: in its automatic version, the machine status signal from the PLC or SCADA with reason coding; in its manual version, the shift report where the operator notes the start, end and cause of each stoppage.

In our example: 440 planned minutes, 55 of unplanned stoppages (22 of breakdown plus 33 of format changeover).

Availability = 385 / 440 = 0.875 → 87.5%

Performance = actual output / theoretical output

Performance = Actual output / Theoretical output
Theoretical output = Operating time / Ideal cycle time

And its dual form in times, algebraically identical:

Performance = (Ideal cycle time × Total pieces) / Operating time

Here the critical variable is the ideal cycle time. It must be the equipment’s nominal design capacity: the figure from the manufacturer’s data sheet or, failing that, the best sustained cycle the machine has demonstrated. Never the historical average. If the historical average is used, performance tends towards 100% by construction and the factor stops detecting losses: you are comparing the machine against itself in its degraded state.

There are two sources: the total piece counter at the machine outlet (photocell, PLC counter) and a master list of ideal cycles per product, because the cycle can vary by the format being produced.

In the example, the line produced 13,860 bottles in 385 operating minutes. Theoretical output at 40 bottles per minute would have been 385 × 40 = 15,400 bottles.

Performance = 13,860 / 15,400 = 0.90 → 90.0%

Check via the dual form: 13,860 bottles × 1.5 s = 20,790 s = 346.5 minutes of net operating time, over 385 operating minutes: 346.5 / 385 = 0.90. The 38.5-minute gap is the speed loss announced in the cascade: minor stoppages and stretches at reduced rate that no manual report ever recorded.

Quality = good pieces / total pieces

Quality = Good pieces / Total pieces
Good pieces = Total pieces − Rejects − Rework

A good piece means good first time (the first-pass yield criterion). A piece that comes out defective, is reworked and ends up being sold does not count as good in OEE: it consumed machine time twice and the indicator must reflect that loss. It is a demanding convention but consistent with the logic of the cascade: valuable operating time is time that produced value first time.

The source is the reject counter (automatic ejectors with a counter, the quality inspection record) plus the rework log if one exists.

In our shift there were 660 rejected bottles: 380 during the start-up after the format changeover (start-up loss, loss number 6) and 280 during the rest of the shift due to fill levels out of tolerance and misplaced labels (loss number 5).

Quality = (13,860 − 660) / 13,860 = 13,200 / 13,860 = 0.9524 → 95.2%

And with the three factors, the shift’s OEE:

OEE = 0.875 × 0.90 × 0.9524 = 0.75 → 75%

Interactive OEE calculator

Before continuing with the theory, you can verify the calculation yourself or enter the data from your own line. The calculator comes preloaded with the example data (440 planned minutes, 55 of stoppages, ideal cycle of 1.5 s per piece, 13,860 total pieces and 660 defective) and should return exactly the factors we have just derived. It is also available as a standalone page in the OEE calculator.

If your data returns a performance above 100%, the calculator will warn you: it is the classic symptom of a poorly defined ideal cycle time, something we address in the mistakes section.

OEE vs TEEP vs OOE: what each one measures

OEE answers a specific question: of what we planned to produce, how much came out right? There are two other legitimate questions that demand different indicators, and confusing them generates sterile conversations between operations and management.

TEEP (Total Effective Equipment Performance) is the most severe: its base is every hour of the calendar. An 85% OEE can coexist with a 60% TEEP in a plant running three shifts with weekend shutdowns, and with a much lower TEEP on a single shift. Our bottling line illustrates it well: with a 75% OEE and one shift a day, utilisation is 440 / 1,440 = 0.3056 (30.6%) and TEEP comes out at 0.75 × 0.3056 = 330 / 1,440 = 22.9%. The line uses its shift well, but the investment in that asset is yielding less than a quarter of its physical capacity.

The practical rule: OEE is the operations metric (improve what is planned), TEEP is the investment metric (decide whether you need to buy another line or simply open a second shift). Before investing in new capacity it pays to look at TEEP: the capacity often already exists, just in unscheduled hours.

Typical mistakes that inflate (or sink) OEE

The literature on OEE implementation (Nakajima himself and, with particular severity, Robert Hansen in Overall Equipment Effectiveness, Industrial Press, 2001) documents recurring errors. The five that follow appear in almost every plant that starts measuring.

Confusing calendar time with planned time. If the denominator includes hours in which no production was scheduled, OEE sinks artificially and stops being comparable between periods with different loads. TEEP already exists for measuring against the calendar; OEE uses planned time.

Classifying changeovers as planned downtime. The opposite effect: format changeovers disappear from the indicator and availability is inflated. In our example, removing the 33 changeover minutes from the calculation would lift availability from 87.5% to 385/407 = 94.6% and OEE from 75% to 81%, without the line having improved in any way. The changeover is an availability loss and one of the most profitable improvement levers there is.

Ideal cycle inflated or based on the historical average. If the “ideal” cycle is actually the average speed the machine has been running at, performance will always hover around 100% and speed losses will remain invisible. The ideal cycle must come from the data sheet or from the best sustained cycle demonstrated.

Double-counting rework. A reworked piece counted as good, or one that passes twice through the outlet counter and is counted as two total pieces, distorts performance and quality at the same time. The correct convention: it counts once as a total piece and zero times as a good piece.

Assuming an OEE is healthy because the factors look healthy. Losses multiply, they do not add. Three factors of 90%, which look acceptable in isolation, produce an OEE of 0.9 × 0.9 × 0.9 = 72.9%. More than a quarter of planned time lost, with three indicators glowing green on a dashboard. That is the whole point of OEE as a product: it forces you to look at the combined effect.

And one symptom that deserves its own mention: an OEE (or a performance figure) above 100% is not good news, it is a definition error. It almost always means the ideal cycle is set wrong (the machine can run faster than the master data says) or that planned stops are concealing real stoppages. The correct reaction is to review the ideal cycle, not to celebrate the record. Detecting this kind of inconsistency in the records is part of the operational diagnostic that kicks off any serious measurement project.

Benchmarks: the 85% world-class figure and its caveats

The most quoted figure in the entire OEE literature is 85% as the world-class threshold. It is worth knowing where it comes from: Nakajima proposed as a condition of excellence an availability above 90%, a performance above 95% and a quality above 99%. The product of those three values is 0.90 × 0.95 × 0.99 = 0.846, which tradition rounded to 85%. It is an indicative reference born of the JIPM’s experience with Japanese discrete manufacturing in the eighties, not a standard nor a threshold with universal statistical validity.

With that caution in mind, the indicative scale the literature uses for discrete manufacturing is this:

Two essential caveats. First: comparing OEE across different plants, lines or sectors is almost always a misleading exercise, because the result depends on definition choices (which stops count as planned, which ideal cycle is used, how rework is treated) that rarely match between organisations. A 70% calculated with strict criteria can reflect a better-run plant than an 85% calculated with a generous hand. Second: the real value of OEE lies in your own time series. Benchmarking against yourself, with stable definitions, shift after shift, is what turns the indicator into an improvement tool rather than a contest.

How to measure OEE: automatic vs manual data

All the theory above rests on three pieces of data: how long the machine was stopped and why, how many pieces came out and how many were rejected. How you capture them determines how reliable the result is.

Manual capture. Paper or spreadsheet shift reports in which the operator logs stoppages and output. It is cheap and fine for getting started, but it has two documented biases. It misses minor stoppages: nobody writes down twenty fifteen-second jams, and those losses vanish from the record (though not from production). And it tends to round upwards: downtime is logged short, awkward causes are softened. The result is a systematically optimistic manual OEE.

Automatic capture. Sensors, PLC signals and MES or SCADA systems logging machine status and counters continuously. It captures the losses invisible to paper, removes the recorder’s bias and lets you compute OEE per shift, per product or per hour with no administrative effort. The experience gathered in the MES implementation literature suggests, as an indicative figure, that when moving from manual to automatic logging the measured OEE typically drops by 10 to 20 points. The plant has not got worse; it has simply started seeing losses that were always there. That initial drop is proof the measurement is working.

The minimum requirements for automating are modest: three signals per machine (run/stop status, total piece counter, reject counter) and a master data record with the ideal cycle for each product. You do not need a full Industry 4.0 architecture to get started; you need those three signals to be reliable. It is the kind of tightly scoped project we tackle through industrial automation of data capture. Plants already operating under management systems with continuous measurement (ISO 50001 in energy is the obvious parallel) usually have part of the data-capture infrastructure journey already behind them.

Complete worked example, step by step

We close the bottling line case with the full calculation, in the style of an exam problem.

Shift data:

Step 1. Time cascade.

Planned production time = 480 − 40         = 440 min
Operating time          = 440 − (22 + 33)  = 385 min

Step 2. Availability.

A = 385 / 440 = 0.875   (87.5%)

Step 3. Performance.

Theoretical output = 385 min × 40 bottles/min = 15,400 bottles
P = 13,860 / 15,400 = 0.900   (90.0%)

Net operating time = 13,860 × 1.5 s = 20,790 s = 346.5 min
Speed loss         = 385 − 346.5 = 38.5 min

Step 4. Quality.

Good pieces = 13,860 − 660 = 13,200
Q = 13,200 / 13,860 = 0.952   (95.2%)

Valuable operating time = 13,200 × 1.5 s = 19,800 s = 330 min
Quality loss            = 346.5 − 330 = 16.5 min

Step 5. OEE by both routes.

Via factors: OEE = 0.875 × 0.900 × 0.952 = 0.75   (75%)
Via pieces:  OEE = (13,200 × 1.5 s) / (440 × 60 s) = 19,800 / 26,400 = 0.75

Both routes match, a sign that the records add up. You can verify it by entering these same figures in the calculator above.

Interpretation. Of the 110 minutes lost in the shift (440 planned minus 330 valuable), 55 are availability, 38.5 speed and 16.5 quality. Within availability, the format changeover (33 min) weighs more than the breakdown (22 min), and it drags a second loss with it: 380 of the 660 rejects occurred during the start-up after the changeover. A SMED project on that changeover would attack the biggest availability loss and more than half the quality loss in one move. That is the right way to use OEE: not as an exam grade, but as a map of where the time is buried, the starting point of any process improvement initiative.

One final warning about using the indicator as a target. OEE carries no energy cost dimension inside it: lifting the performance factor by speeding the line up can worsen consumption per unit produced, and the indicator will not record it. When that second dimension matters, the trade-off between rate and energy is analysed with crossed production and consumption data, which is exactly what the production and consumption optimisation module of our platform does.

Frequently asked questions about OEE

If you want to move from a paper calculation to an OEE measured continuously with real shop-floor data, at Captia Technology we help manufacturers capture their machines’ signals and turn them into reliable indicators. Get in touch and we will look at your case.