You have an OEE number. It is probably somewhere in the sixties, it is reported weekly, and there is a target above it that somebody set before you arrived. Improvement projects get justified against it. It appears in the monthly review with a trend line. And if you are honest, you cannot point to a single decision in the last year that was made differently because of it.
That is the normal experience of overall equipment effectiveness in a mid-market manufacturer, and it is not because the people running it are unsophisticated. It is because OEE has a structural problem that gets worse the more seriously you take it. This piece explains the mechanism, sets out where OEE genuinely earns its place, and weighs the alternatives on the same criteria so you can decide rather than just react.
What OEE Actually Measures
OEE is the product of three ratios: availability, performance, and quality. Availability is the share of planned production time the equipment was actually running. Performance is the share of theoretical speed it ran at while running. Quality is the share of output that came out good. Multiply the three and you get a single percentage.
The intent behind it is reasonable and worth stating fairly, because a lot of criticism of OEE attacks a version nobody defends. The intent is to stop plants from optimizing one loss while creating another: running fast but scrapping, or achieving high uptime by running slow. Combining the three forces a conversation about total effectiveness rather than any single dimension.
That intent is sound. The execution creates a number that resists being acted on, and the reason is arithmetic rather than philosophy.
It is worth being precise about what the percentage represents, because this is where most of the confusion starts. An OEE of 70 percent does not mean the equipment ran 70 percent of the time, and it does not mean 30 percent of your capacity was wasted in any recoverable sense. It means that after three separate discounts were applied in sequence, 70 percent of the theoretical good output during planned production time was achieved. Every term in that sentence is doing work, and two of them are internal decisions rather than measurements.
Where the Metric Came From and Why That Matters
OEE was designed for a particular kind of plant: capital intensive, equipment constrained, running long stable campaigns of a small number of products, where the machine genuinely was the business and the dominant question was how much of the machine’s theoretical capability you were capturing.
That is a real kind of plant and OEE works well in it. The difficulty is that the metric travelled far beyond its original setting, into high mix operations with frequent changeovers, into assembly work where labor rather than equipment sets the pace, and into businesses where the binding constraint is demand, engineering capacity, or a single supplier rather than any machine on the floor.
When a metric moves outside the conditions it was designed for, it does not announce that it has stopped working. It keeps producing a number, the number keeps moving, and everyone keeps reviewing it. The failure is silent, which is why it can persist for years inside an otherwise well run operation.
So the first question about OEE is not whether it is a good metric. It is whether your plant resembles the plant it was built for. Long runs, few products, equipment constrained, stable mix: keep it. Short runs, high mix, labor paced, or constrained somewhere other than the floor: the fit is poor and no amount of discipline in the calculation will fix that.
Why Three Losses in One Number Is the Core Problem
The three components have nothing in common except that they reduce output. They have different causes, different owners, different fix timelines, and different costs per point.
Availability losses are usually maintenance, changeover, and material starvation. They are owned by maintenance, engineering, and planning respectively, and they are fixed over weeks to months with capital or process redesign. Performance losses are usually short stops, speed reduction, and operator variation, owned by production supervision, fixed over days with training and standard work. Quality losses are owned by quality engineering and fixed over months with process capability work.
Three owners, three timelines, three cost structures. Now multiply them together and report the product to a leadership team. What is anyone supposed to do with that? The number went down two points. Nobody in the room can tell which of three unrelated problems moved, or whether one improved while another deteriorated more.
Availability, performance, and quality have different owners, different fix timelines, and different costs per point. Multiplying them together produces a number nobody in the room can act on.
The standard answer is that you drill into the components. Correct, and that is the concession that gives away the argument: if every use of the metric requires immediately decomposing it back into its three parts, the composite added nothing except a step.
The Multiplication Trap
Multiplying three percentages creates a second problem that is less discussed and more damaging: it makes equal-sized movements in different components look identical when their value is not remotely equal.
Hypothetical: availability 90 percent, performance 85 percent, quality 98 percent gives an OEE of 75 percent. Now improve availability by five points, to 95, and OEE rises to 79. Alternatively improve performance by five points, to 90, and OEE rises to 79 as well. Identical movement in the reported number.
But those two improvements are not the same thing. The availability gain might have required a capital project and a maintenance contract. The performance gain might have been a standard work change that cost nothing. And if the line is not the constraint, neither one produced a single additional shippable unit. The composite reports them as equivalent, which is exactly the information a capital allocation decision needs and does not get.
Multiplication also means a component with little room to move can dominate the perception of the whole. Quality at 98 percent has two points available. Performance at 85 has fifteen. The number treats a point in each as worth the same, so effort gets spent where it is scarce rather than where it is available.
Why OEE Rewards the Wrong Machine
This is the failure with the largest financial consequence, and it follows from a basic fact about manufacturing: output is set by the constraint, and improvement anywhere else produces no additional output at all.
When OEE is reported by asset and compared across assets, it creates pressure to raise the low numbers. The low numbers are frequently on non-constraint equipment, because non-constraint equipment is allowed to idle, and idling is precisely what availability punishes. So the plant invests attention, and often capital, in raising effectiveness on machines that were never limiting anything.
The result is not neutral. Running a non-constraint faster or longer does not increase shipments. It increases work in process, which has to be moved, stored, counted, and eventually revalued. The plant reports improved effectiveness and consumes cash. Every metric that improves when you overproduce will eventually do this, and OEE measured broadly across assets is one of them.
There is a subtler version of the same problem that shows up in capital requests. When effectiveness is reported per asset and compared across assets, the asset with the worst number becomes the natural candidate for investment, and it will have a credible improvement case attached because the gap to target is largest there. Meanwhile the constraint, sitting at a respectable number with very little headroom, attracts nothing. The plant systematically invests where the metric has room rather than where the business has need, and every individual decision in that sequence looks defensible on paper.
Running a non-constraint machine harder does not add shipments. It adds work in process. The metric improves and the cash position gets worse.
The defense is that OEE should only be applied to the constraint. That defense is correct, and it is also an argument for a different metric, since what you want on the constraint is the number of good hours it produced, not a compressed ratio of three loss types.
Does a Higher OEE Always Mean More Output?
No, and this is the single most useful thing to understand about the metric. OEE and output are only reliably linked at the constraint, and even there the link depends on the denominator.
Availability is measured against planned production time. That denominator is a decision, not a fact. Reduce the planned hours and availability rises without anything changing on the floor. A plant that cuts a shift can improve its OEE while shipping less, and the metric will report this as progress.
Performance is measured against theoretical or nameplate speed, which is also a choice, and one that is often decades old or set by an equipment vendor with an incentive. Where the standard is soft, performance is soft, and a plant can improve its performance ratio by revisiting the standard rather than the process.
Neither of those is fraud. Both are normal administrative decisions that move a headline metric without moving the business. A metric whose denominator is set internally and reviewed rarely will drift, and the drift will be in the comfortable direction.
The test I would apply is to ask when each denominator was last changed and who approved it. If planned production time has been quietly adjusted twice in three years and the nameplate speeds predate anyone currently employed, then your multi-year OEE trend line is not measuring the plant. It is measuring a mixture of the plant and a series of administrative decisions, drawn as a single continuous line, with no marker showing where one ends and the other begins.
Where Does the Target Number Come From?
In most plants, the target was inherited. It came from a training course, a consultant’s deck, a previous employer, or a widely repeated figure whose original derivation nobody in the building can trace.
That matters more than it sounds. A target with no traceable derivation cannot be argued with, which means it cannot be improved, and it also cannot be defended when someone asks whether it is appropriate for this equipment, this product mix, and this changeover profile. A high-mix plant running short runs has a structurally different achievable ceiling than a plant running long campaigns of one product. Applying the same target to both is not rigor, it is the appearance of rigor.
If you keep OEE, derive the target from your own demonstrated performance: the best sustained period you actually achieved on that asset with a comparable mix. That is defensible, it is specific to your equipment, and nobody can argue it is impossible because you already did it. It also has the advantage of changing when your mix changes, which an inherited industry figure never does.
How OEE Interacts With Mix and Changeover
Product mix moves OEE substantially without any change in how well the plant is run, and in a high mix operation this effect is large enough to swamp genuine performance movement entirely.
Two mechanisms drive it. Changeovers consume availability, so a month heavy in short runs reports lower availability than a month of long campaigns, regardless of how efficiently those changeovers were executed. And the performance ratio compares actual speed against a standard set per product, so a shift toward products with softer standards raises performance while a shift toward tightly set ones lowers it.
Hypothetical: a plant runs the same equipment, the same crew, and the same practices in two consecutive months, but the second month carries twice as many changeovers because the order book shifted toward smaller quantities. OEE falls several points. Nothing got worse. The order book changed, which is a commercial fact rather than an operational failure, and the metric has no way to say so.
This is why comparing OEE month to month in a high mix plant generates so many unproductive conversations. If you keep the metric under those conditions, report it alongside changeover count and a mix indicator, so the room can see which of the three moved before anyone is asked to explain the first one.
How OEE Distorts Behavior
Any metric that is reviewed will be managed, and the specific distortions OEE produces are predictable enough to list.
Longer runs. Changeovers hurt availability, so the rational response to being measured on OEE is to run larger batches, which increases inventory and lengthens customer lead times. A plant optimizing OEE will drift away from flexibility, which is usually the opposite of the strategic direction.
Reclassified downtime. The boundary between planned and unplanned is a judgment, and where a metric depends on that judgment, the judgment moves. Unplanned stops become scheduled maintenance. Planned production time shrinks to exclude periods nobody wanted to explain.
Constraint neglect. Effort flows toward whichever asset has the worst number rather than whichever asset limits the business, because the worst number is the one that gets asked about in the review.
Deferred improvement on the constraint. Because effort follows the reported gap, the constraint is often the last asset to receive attention, and it is the only one where attention converts into shipments.
None of this requires bad faith from anyone. These are rational responses to what is being measured, which is the point. The distortion is a property of the metric, not of the people, and the correct response is to change the measurement rather than to ask people to behave irrationally against it.
What About the Six Big Losses?
The loss taxonomy that sits underneath OEE is better than the composite it feeds. Breakdowns and setup time under availability, small stops and reduced speed under performance, startup rejects and production defects under quality. That categorization is genuinely useful and it is the part of the framework I would keep in almost every plant.
What the taxonomy gives you is a common language for where time goes, and a way to prevent one team from claiming an improvement that another team paid for. Those are real benefits and they survive entirely without the multiplication step at the end.
So treat the taxonomy and the composite as separable, because they are. Categorize losses six ways, record them in hours, and total them. The result is a loss profile rather than a percentage, and a loss profile is directly comparable across weeks, additive across assets, and convertible into capacity and then into money.
The one refinement worth making is to keep the categories in the language your floor actually uses. Six imported category names that nobody says out loud will get recorded inconsistently, and inconsistent categorization is worse than fewer categories. Three categories everyone applies the same way beat six that get interpreted differently on each shift.
What OEE Is Genuinely Good For
There are three situations where I would keep it, and it is worth being specific about them, because the honest case against OEE is not that it is useless.
The first is a single dedicated asset running a stable product where the equipment genuinely is the business, such as a continuous line or a bottleneck press that everything flows through. Under those conditions the three components are stable, the denominators are meaningful, and the composite tracks output closely.
The second is capital justification for equipment replacement, where you need a defensible before and after on a specific asset and the comparison is to itself rather than to another machine.
The third is a plant so early in its improvement journey that no structured measurement exists at all. OEE imposes a loss taxonomy, and having any taxonomy is better than having none. The value there is the discipline of categorizing losses, and that value fades once the categories are established.
Notice what those three cases share. In each one, the comparison is to the same asset at a different point in time, under stable conditions, with the components visible. None of them involves ranking assets against one another, and none of them uses a target borrowed from outside the plant. Those two uses are where the metric stops informing and starts misdirecting, and they are unfortunately the two most common.
When Should You Use OEE?
Use it when the asset is the constraint, the product mix is stable, the denominators are documented and reviewed, and you report the three components alongside the composite rather than instead of them.
Do not use it as a cross-asset comparison, as a plant-level headline, as a driver of capital allocation across different equipment, or as a target imported from outside your own history. Those four uses account for most of the OEE reporting I encounter and all of the damage.
The practical test is the one I would apply to any metric: name the decision it changes. If the OEE number moved three points next month, what would you do differently, and could you do it without first decomposing the number? If the answer requires decomposition, the components are your metric and the composite is a summary you are reviewing out of habit.
The Alternatives, Weighed on the Same Criteria
Three candidates, judged on what each one answers, how directly it can be acted on, how easily it distorts, and what it misses.
| Criterion | OEE | Constraint good hours | Attainment and first-pass yield as a pair |
| Question it answers | How effective was this asset across three loss types | How many good hours did the limiting resource produce | Did we build what we promised, and did it come out right the first time |
| Directly actionable | No, requires decomposition first | Yes, each lost hour has a named cause | Yes, each has a single owner and a short fix loop |
| Distortion risk | High. Denominators are internal decisions | Low. Hours are observable and hard to reclassify | Moderate. Sequence substitution can flatter attainment |
| Links to output | Only at the constraint | Directly, by construction | Directly, through the promise and the rework loop |
| Rewards overproduction | Yes, when applied off the constraint | No | No |
| What it misses | Which of three problems moved | Non-constraint deterioration that has not yet bound | Equipment condition trends |
| Best used for | Single dedicated asset, before and after capital | Daily and weekly operating decisions | Weekly operating review and supervisor accountability |
Read across the actionability row. That is the practical case: the alternatives can be acted on without an intermediate analytical step, and the intermediate step is where operating attention gets lost. Notice also that the two alternatives fail differently rather than failing less. Constraint good hours will miss a non-constraint asset degrading quietly until the day it becomes the constraint. The attainment and yield pair can be flattered by substituting easy jobs for hard ones within the same week. Both of those failure modes are visible and correctable once you know to look for them, which is the most you can ask of any metric.
What to Measure Instead, Specifically
Start with constraint good hours. Identify the limiting resource, then count the hours it spent producing good output against the hours it was available to. Report the lost hours by cause in a short list rather than as a ratio: hours lost to changeover, to breakdown, to material starvation, to quality stoppage, to no demand.
That list is the whole point. It is additive, so you can total it. It is in hours, so it converts directly into a capacity conversation and from there into dollars. Each line has an owner. And nothing in it can be improved by running a machine that was not limiting anything.
Then add the pair: schedule attainment and first-pass yield. Attainment asks whether you built what you said you would build in the sequence you said, which protects the customer promise. First-pass yield asks whether it came out right without rework, which measures the process rather than your inspection. Together they cover the ground that performance and quality were trying to cover inside OEE, with one owner each and no multiplication.
Lost hours by cause is additive, it is owned line by line, and it converts straight into capacity and then into dollars. A ratio does none of those things.
How Do You Replace OEE Without Losing the Data?
You do not turn anything off. The underlying capture, which is usually the expensive part, stays exactly as it is. What changes is what gets reported to leadership and what carries a target.
Keep collecting availability, performance, and quality data at the asset. Stop multiplying them for the executive review. Report constraint good hours and the lost-hour causes to leadership, keep the component detail available to engineering and maintenance, and retire the composite target rather than the composite calculation. If someone wants the OEE number for a capital comparison later, it is still there.
Run both for one quarter. At the end of it, look at which number actually changed a decision. This is a better argument than any case anyone can make in a meeting, because it uses your own plant as the evidence and it lets the metric defend itself if it can.
Expect resistance from two directions. Engineering may have built improvement programs around the composite and will reasonably ask what happens to the roadmap, and the answer is that the roadmap is built on the components, which are unchanged. Corporate may require OEE reporting on a template, in which case you supply it and simply do not manage to it, which is a cost worth paying rather than a battle worth fighting.
Does Any of This Change in a Small Plant?
The arithmetic is identical and the practical answer is simpler, because a small plant usually knows exactly where its constraint is without needing an analysis to find it.
In a business with one bottleneck machine and a couple of dozen people, everyone on the floor can already tell you what limits the place. The measurement job is therefore small: count the good hours that resource produced, list where the rest went, and check the promise and the yield. That can be maintained on a whiteboard and reviewed in ten minutes, and it will outperform an OEE reporting system that costs software, data capture, and somebody’s Friday afternoon.
I spent four years owning a small industrial tank liner manufacturer before selling it to its largest vendor, and the lesson that transfers most directly from that scale to the roughly one billion dollar division I run now is about what a measurement has to earn. At small scale the cost of a metric is visible, because it is somebody’s time and there is no slack to hide it in. At large scale that same cost is real and invisible, spread across a reporting function nobody questions.
The discipline worth importing from the small end is the question itself: what did this number cost to produce, and what decision did it change? Asked honestly of an OEE program in a plant where it does not fit, the answers are usually a great deal and nothing.
What Good Looks Like Without an OEE Number
A weekly operating review that opens with constraint good hours, a five-line loss list in hours with a name against each line, schedule attainment against the original sequence, and first-pass yield. Four items, all additive or binary, none requiring decomposition before anyone can respond.
The conversation that follows is different in a way that is noticeable within a month. Instead of discussing why a ratio moved, the room discusses forty lost hours, who owns them, and what it would take to recover half of them next week. That is an operating conversation. The ratio version was an analytical one, and analytical conversations in an operating review are how a week gets spent without anything changing.
Where those four numbers sit in the week, and which meeting owns each decision, is a structural question rather than a measurement one. The operating cadence is where that gets built, the twelve manufacturing KPIs are the wider scorecard these compete for a place on, and the classification logic behind picking the upstream versions is covered in leading versus lagging indicators.
One limitation of everything above, stated plainly: constraint good hours is only correct while you know where the constraint is. In a high-mix plant the constraint moves with the mix, sometimes weekly, and a business that installs constraint-based measurement without a routine for re-identifying the constraint will end up precisely measuring a resource that stopped limiting anything two months ago. That routine is a real cost and it is not optional.
About the Stagnation Assassin
Todd Hagopian is a Fortune 500 transformation executive who has generated $3B in shareholder value across Berkshire Hathaway, Illinois Tool Works, Whirlpool, and JBT Marel, where he serves as Global VP of Product Strategy and Marketing. Known as The Stagnation Assassin, he is the author of two published books: The Unfair Advantage: Weaponizing the Hypomanic Toolbox and Stagnation Assassin: The Anti-Consultant Manifesto. His blog is published in 15+ languages and read by operators worldwide. Bring him to your stage via his speaking page or connect with him on LinkedIn.
Decision Velocity Audit. Send me the metrics your weekly operations review opens with and what happened the last three times each one moved. I will tell you which ones are changing decisions and which ones are being reviewed out of habit. Free, and there is no pitch at the end of it. Request the audit here.
See also: 12 Manufacturing KPIs That Matter