Two plants each report 82% OEE. One is genuinely world-class. The other is really running at about 60%, wearing makeup. The number on the slide is identical; the reality on the floor is not even close.
That is the quiet problem with Overall Equipment Effectiveness. The metric itself is excellent — it is the best single number manufacturing has for how well equipment converts available time into good product. The problem is that the same three letters mean very different things depending on choices made before the arithmetic ever begins. Change those choices and you can move the reported number ten or fifteen points without touching a single machine.
The number everyone quotes
OEE is the product of three factors: Availability × Performance × Quality. Run each at world-class and you land at the famous 85% — Nakajima's original TPM benchmark of 90% availability × 95% performance × 99.9% quality.
Those bands are useful — but only when the definitions underneath the three factors are held constant. In practice, they rarely are. And that is where a perfectly good metric quietly stops being comparable, plant to plant, line to line, and sometimes shift to shift.
Five choices that quietly inflate the number
None of these are dishonesty. Each is a defensible decision someone made for a reasonable local reason. Together, they can turn a 60% plant into an 82% slide.
| Excluding "planned" downtime | The biggest lever by far. Move changeovers, planned maintenance, and "no orders" out of the denominator and availability leaps. It can be legitimate — but it hides the time the asset was not making money. This is why TEEP (Total Effective Equipment Performance), which counts all calendar time, often tells a very different story. |
| Inflated ideal cycle time | Performance is measured against an "ideal" rate. Set that ideal to the comfortable rate the line actually runs today — rather than the equipment's nameplate capability — and performance loss simply disappears from the math. |
| Definition drift | Production's OEE, maintenance's OEE, and finance's OEE are often three different numbers built on three different assumptions. When no single definition is owned, the highest one tends to become the one that gets reported. |
| Rework counted as good | Quality should reflect first-pass yield. Count reworked or reprocessed units as "good," and the quality factor overstates how well the process actually ran the first time. |
| Bottleneck cherry-picking | Report OEE on a machine that happens to look good, rather than on the constraint that actually governs the plant's throughput. A high number on a non-constraint tells you almost nothing about the plant's real output. |
Change the choices, not the machines, and the reported OEE moves ten or fifteen points. That is the whole problem in one sentence.
Try it: how the three factors multiply
OEE is a product, not an average, so each factor punishes the others. Enter your own availability, performance, and quality and watch where you land.
Availability × Performance × Quality
Notice how a plant with a strong-looking 90% on each factor still lands at 73% — because the losses multiply. That compounding is exactly why an inflated single factor is so distorting.
What an honest OEE actually looks like
The goal was never a high number. It is a trustworthy one — a number you can act on. In practice that means a handful of disciplines:
| Nameplate ideal cycle time | Performance measured against what the equipment was designed to do, not the rate the line has drifted into being comfortable with. |
| One owned definition | A single OEE definition, written down and applied identically across every shift and line — so the numbers are actually comparable. |
| Measured at the constraint | OEE reported on the bottleneck that governs throughput, because that is the only place a point of OEE converts directly into more product. |
| Planned downtime made visible | Not excluded and forgotten. Shown alongside OEE — often as TEEP — so the organisation sees the capacity sitting inside changeovers and idle calendar time. |
| Traceable to raw data | The number can be walked back to actual run time and part counts in under a minute — not reconstructed from a summarised monthly report. |
Why this matters more in the AI era
An inflated OEE used to be a story you told leadership once a month. That was survivable. What is new is that the same number is now becoming an input to software.
Feed a flattering OEE into an AI scheduling engine or a predictive-maintenance model and it stops being a harmless story and becomes a premise the machine believes and optimises around. If the "ideal" cycle time is wrong, every capacity projection built on it is wrong. If planned downtime is invisible, the model never learns to attack it. The metric's integrity was always important; automation just raised the stakes on getting it right — the same order-of-operations problem that leaves most Industry 4.0 pilots stranded on data no one had cleaned first.
Is your OEE honest?
A short self-assessment. Nothing tracked, nothing saved.
What to watch next
Chasing 85% is the wrong goal for most plants. An honest 62% that everyone trusts beats a cosmetic 82% that no one can defend — because only one of those two numbers tells you where the next hour of capacity is actually hiding. Fix what the number means before you spend a dollar trying to move it.
Not sure your OEE is telling you the truth? Happy to compare notes.
Get in touch →Where these numbers come from
- Seiichi Nakajima, Introduction to TPM — original 90% × 95% × 99.9% = 85% world-class benchmark
- Fabrico, OEE Benchmarks: From Average 60% to World-Class 85% — 250-plant European study, 2025
- Evocon, World-Class OEE: Industry Benchmarks From 50+ Countries
- Lean Production, Understanding OEE in Lean Manufacturing — OEE / TEEP definitions and the Six Big Losses
- Hero image: site asset (blog-control.jpg)