Why Most OEE Benchmarks Are Wrong

Key takeaways

See the benchmarks done right.

• The "world-class OEE is 85%" benchmark is repeated everywhere and almost never useful. It came from a specific industry context (discrete assembly, mature programs, narrow loss definitions) and gets misapplied to plants that look nothing like it.
• Published OEE benchmarks can differ by double-digit percentage points for the same actual performance because plants define availability denominators differently, count quality losses differently, and apply ideal cycle times inconsistently.
• The right benchmark is your own plant's rolling 12-month baseline, segmented by line type. Comparing yourself to your own past is the only comparison where the numerator and denominator are defined the same way at both ends.
• If external benchmarks must be used, use them as direction-of-travel checks (is the gap closing) rather than as point targets. Plants that chase a specific external number usually find themselves optimising the calculation rather than the operation.

Where the 85% number came from

The "85% is world-class" figure traces back to Nakajima's foundational TPM work (Introduction to TPM, 1988), where the 85% is the product of 90% availability × 95% performance × 99% quality, and was based on a specific operational context: discrete repetitive assembly with mature OEE measurement, narrow availability definitions, tight cycle-time standards. The figure was reasonable for that context.

It is repeated today as if it were universal. It is not. A pharma plant running batch reactions does not have the same OEE math as a discrete assembly line. A continuous-process plant does not measure good units the same way as a packaging operation. A maintenance-heavy heavy-industry plant has a different planned-time denominator than a fast-changing CPG operation. Applying the same 85% target across them produces meaningless comparisons.

The article on manufacturing KPIs covers how the underlying components produce the OEE composite.

Why published numbers vary so much for the same performance

Three structural causes:

1. Availability denominator inconsistency

Some plants count "available time" as 24/7 minus scheduled shutdowns. Others count only scheduled production shifts. Others subtract planned PM windows from the denominator. The same physical operation can produce an availability number 15 percentage points different depending on which definition is used. Without standardising the denominator, comparisons are accidental.

2. Quality losses counted differently

Discrete operations count rejects in-line. Batch operations measure at release. Some plants count first-pass yield; others count final yield after rework. A reject rate of 1% in one accounting becomes 4% in another for the same actual quality outcome. The article on production loss analysis covers the related taxonomy problem.

3. Ideal cycle time set by different rules

Some plants set ideal cycle time as the OEM nameplate. Others set it as the historical best 5-minute window the line has ever achieved. Others set it as the design intent adjusted for product mix. The same line can show performance of 70% or 95% depending on which ideal is in the denominator.

Until two plants align on all three of these, comparing their OEE numbers is comparing different calculations. Most published benchmarks aggregate plants without standardising. The aggregate is the average of inconsistencies.

The benchmark that actually works

Your own plant's rolling 12-month baseline, segmented by line type. The math:

• For each line type (discrete, batch, continuous), compute the rolling 12-month average OEE.
• Report the current month against that line-type-specific baseline.
• Track the trend, not the absolute number.

This comparison is the only one where the availability denominator, the quality accounting and the ideal cycle time are guaranteed to be defined the same way at both ends. The number is honest. The trend means something.

The article on root cause analysis covers how trend analysis against your own baseline drives investigation more usefully than gap analysis against a published number.

If you must use external benchmarks

Three rules to make them less misleading:

1. Use them as direction-of-travel

"Industry plants in our segment are typically improving OEE by a point or two per year" is a useful statement, even when the absolute numbers are inconsistent. Year-over-year change is more comparable than levels.

2. Use ranges, not point estimates

"World-class is 85%" produces a target. "Plants in our segment typically run 65-80%" produces context. The range communicates the underlying uncertainty.

3. Disclose your own definitions

When reporting your OEE externally, disclose your availability denominator, quality accounting, and ideal cycle time methodology. The number becomes interpretable. Most plants do not do this, and most published benchmarks are missing the disclosure too, which is part of why they vary so much.

What about industry-vertical benchmarks?

"OEE in food and beverage averages 60%" is a more useful claim than "world-class is 85%" because the vertical narrows the operational variability. It is still subject to the three measurement inconsistencies above, but the denominator is at least conceptually closer.

The way to use a vertical benchmark: compare against the published range, not the published average. If the range for your vertical is 55-75% and you are at 62%, you are in the middle of the range and the next conversation is about which specific levers move you up.

If you are at 35%, the conversation is different. The number itself does not tell you what to do; the gap to the range tells you whether to look at the OEE program or at deeper operational issues. The piece on work order management systems covers some of the deeper-issue levers.

How to set your own internal target

Three inputs:

• Current rolling 12-month baseline. The starting point.
• Highest sustained month in the last 12. Roughly the floor of "what is operationally achievable today without structural change."
• Theoretical maximum if every identified loss class were 50% lower. The ceiling for the next year's work.

The next year's target sits between the highest sustained month and the theoretical maximum, biased toward the lower end. Targets that exceed the theoretical maximum produce gaming of the calculation rather than improvement in operation.

How Fabrico fits

The internal baseline approach works in any OEE system, but it requires the math to be applied consistently month over month. Where a unified platform helps is that the availability denominator, the quality accounting and the ideal cycle time are set once and applied uniformly, so the rolling-12-month comparison is honest.

Fabrico is built so the OEE calculation, the loss taxonomy and the historical trend stay consistent through asset and product changes. To see how your own internal baseline picture looks segmented by line type, book a demo .

Frequently asked questions

Is the 85% figure ever right?

For discrete repetitive assembly with mature TPM programs, narrow availability denominators and disciplined ideal cycle times, sometimes. For most other contexts, no.

How should we report OEE to senior leadership?

Three numbers per line type: current month, rolling 12-month average, and change vs prior 12-month. The third number is what makes the report decision-relevant; the first two are context.

What if our board insists on external benchmarking?

Provide it with explicit caveats. "Our 72% places us in the middle of the published industry range; the variance in published numbers makes precise placement unreliable, so we are also reporting our own year-over-year change of +2.3 percentage points." The combined framing is more honest than either number alone.

Should we report OEE by line or by plant?

Both, but in separate views. Plant-wide aggregates hide the actionable picture (which lines are improving, which are degrading); line-level numbers swamp leadership with detail. The plant-wide number with a per-line breakdown one click away is the working balance.

What is the most common implementation mistake?

Picking the highest single shift's performance as the ideal cycle time. The number becomes unachievable, the OEE looks artificially low, and the team stops believing the metric. The ideal cycle time should be the sustained best, not the once-ever best.