The most common OEE mistakes all inflate the number: excluding real downtime as planned, padding the ideal cycle time, counting reworked parts as good, missing short stops, and averaging across dissimilar products. None is dishonesty, but together they can lift a reported OEE ten to twenty points.
OEE is only useful if it is comparable to itself over time and honest about the losses it names. The trouble is that every term has a soft spot where a reasonable-sounding choice quietly makes the number bigger. Left alone, those choices accumulate for years until a plant reports 75 percent while genuinely running in the high 50s, and then wonders why the "good" OEE never translates into the throughput or the delivery it promised. This post walks the specific slips that inflate OEE, why each one hides real loss, and how to correct it. If you want the clean method first, start with OEE calculation and use this as the debugging companion.
Mistake: excluding real downtime as "planned"
The single biggest inflator is quietly reclassifying operational losses as planned downtime so they never enter the OEE denominator. Changeovers, minor stoppages, unstaffed-but-runnable time, planned "meetings," and slow periods get carved out of planned production time, which shrinks the base OEE is measured against and lifts the score without changing a thing on the floor. The honest rule is narrow: planned downtime should cover only events genuinely outside the scope of operational improvement, like a scheduled major shutdown, not the routine losses you are supposed to be attacking.
The tell is a suspiciously small denominator. If planned production time is a lot less than the calendar and crewed time suggests, someone has been generous with the "planned" bucket. This is where the difference between OEE and TEEP matters: TEEP measures against all calendar time and refuses to let excluded hours disappear, so a wide OEE-versus-TEEP gap is often the first sign that hours are being hidden rather than genuinely unavailable. Changeover in particular belongs inside the loss, not outside it; it is setup and adjustment, one of the six big losses not a planned exclusion.
Mistake: padding the ideal cycle time
The second inflator hides in the performance term: using a soft ideal cycle time instead of the true theoretical best. Ideal cycle time is supposed to be the fastest the machine can produce a good part under perfect conditions. When teams instead use the nameplate rate the vendor was willing to commit to in writing, or worse, a "realistic" rate padded with the very slowdowns OEE should catch, the performance term flatters itself. The machine looks like it is running near ideal because the ideal was set low enough to reach.
A padded cycle time makes speed loss vanish by definition. If the ideal is set to the average observed rate, then reduced speed, the fourth big loss, can never appear, because you have defined it away. The fix is to anchor ideal cycle time to the genuine physical best, the fastest sustained good-part rate the equipment has actually demonstrated or is engineered to hit, and then let the performance term show the gap honestly. A painful performance number is the point; it is where reduced speed and minor stops live.
Mistake: counting rework as good
The quality term inflates when reworked parts are counted as good because they eventually ship. Commercially that feels fair, the customer gets a conforming part, but OEE quality is meant to measure first-pass good units, the ones that came off right the first time with no touch-up. Counting reworked parts as good erases the quality loss that the rework itself represents, along with the capacity the rework consumed. The first-pass yield is the honest input, not the final ship count.
The reason this matters is that rework is not free even when it succeeds. It uses labor, time, and often the very machine whose OEE you are measuring, so counting the reworked part as good double-flatters the number: the quality loss disappears and the capacity the rework ate never shows up as a loss anywhere. An honest quality term counts only first-pass good, which keeps the pressure on the process that produced the defect rather than on the rework cell that hid it.
Mistake: missing short stops and using paper logs
The fourth inflator is the loss you never record: short stops below the logging threshold and end-of-shift paper logging that rounds losses away. A five-minute minimum on the downtime log erases every chronic ninety-second jam, and a log filled in from memory at shift end reconstructs a tidier day than the one that happened. Both push OEE up by simply failing to capture loss. The chronic short ones are their own deep topic, covered in chronic minor stops but the point here is that uncaptured loss is indistinguishable from no loss in the OEE math.
Reason-coded, real-time machine downtime capture is the countermeasure, because it timestamps stops as they happen rather than reconstructing them later. When the capture is honest, OEE drops, and that drop is not the process getting worse; it is the measurement finally telling the truth. Teams that skip this step often celebrate an OEE that is really just an incomplete downtime log.
Mistake: averaging OEE across dissimilar products or machines
The fifth inflator is aggregation: rolling a single OEE across a mixed line or a whole plant, which produces a number that is not wrong so much as meaningless. You cannot improve "the plant"; you improve a specific machine running a specific product. Averaging blends fast products with slow ones and healthy machines with sick ones, hiding the actionable detail and often masking the one asset dragging everything down. A plant-level OEE is fine as a headline, useless as a work list.
The subtler version is averaging across dissimilar products on the same machine, where each product has a different ideal cycle time. Blend them and the performance term becomes a weighted fiction that describes no actual run. Calculate OEE per product-machine combination first, then roll up only for reporting, never for diagnosis. The rules for machine-versus-line and product-versus-product OEE are worth getting right before you trust any aggregate; the manufacturing KPIs scorecard keeps those levels separate on purpose.
How do you audit your OEE for inflation?
Run a deliberate audit rather than trusting the number. This sequence surfaces most of the inflation:
- Rebuild the time base from the calendar. Start from all crewed time and list every hour excluded as planned. Challenge each exclusion against the rule that planned means outside operational improvement, and put changeover and minor stops back in.
- Check the ideal cycle time against physics. Confirm the ideal is the fastest demonstrated good-part rate, not the nameplate or a padded average. If speed loss never appears, the ideal is too soft.
- Trace the good count to first pass. Verify that reworked, sorted, and concession parts are excluded from good count. Compare against first-pass yield; a gap is hidden quality loss.
- Test the downtime threshold. Find the minimum stop the log captures and estimate the loss below it. If short stops are uncounted, the availability and performance terms are both flattered.
- Disaggregate the number. Break the reported OEE down to product-machine level. If it was averaged across dissimilar products or machines, the aggregate hid the worst offender.
- Compare against TEEP. Measure the same asset against all calendar time. A wide OEE-to-TEEP gap points at hours excluded rather than genuinely unavailable.
- Recompute and expect a drop. An honest recomputation usually lands well below the reported figure. Treat that drop as recovered visibility, not as a step backward.
What does an honest OEE look like?
Lower than the inflated one, and far more useful. Industry write-ups estimate that plants reporting 70 to 80 percent are frequently running closer to 55 to 62 percent once these five slips are corrected, a ten-to-eighteen-point gap that is the sum of the mistakes above, not a single act of fudging. An honest OEE feels worse and works better, because now the losses it names are real and the improvements you make actually show up in throughput and delivery. For context on where the number should sit, see what is a good OEE score which puts the 85 percent world-class figure in its place.
The definitions that keep OEE honest are not a matter of opinion. The international manufacturing KPI standard, ISO 22400-2, specifies OEE and its supporting time model precisely so availability, performance, and quality mean the same thing across sites and cannot be quietly redefined (ISO 22400-2:2014). And the prize for honesty is concrete: with U.S. manufacturing capacity utilization in the mid-70s percent range, most recently about 75.7 percent (Federal Reserve, G.17), the losses an honest OEE exposes are recoverable capacity the plant already owns.
Honesty is easier when the data collects itself. Paper logs and manual roll-ups are where most of these slips creep in, because every hand-entry is a chance to round, exclude, or blend. Real-time, reason-coded capture at the station computes OEE from timestamps and first-pass counts as they happen, which removes the soft spots the five mistakes live in. That move from paper to live capture is what CLS built across its shops (see the CLS case study), and you can sanity-check your own numbers against the method with the OEE calculator. No rip-and-replace, just a number you can finally trust.