World-Class OEE by Industry: The 85% Myth and Real Bands

World-class OEE is commonly pegged at 85%, but that number is the arithmetic product of Seiichi Nakajima’s 1980s TPM targets, 90% availability × 95% performance × 99% quality, set for high-volume discrete manufacturing. Realistic OEE varies widely by industry, and most plants live somewhere between 45% and 75% by the design of the work they do.

The 85% figure gets quoted like a law of physics. It is closer to folklore with a footnote. It is a useful reference point, but applying it to a food plant that runs clean-in-place cycles every shift, or a job shop that changes over three times a day, tells you almost nothing except that your work is not high-volume discrete assembly. This post traces where 85% actually came from, lays out realistic OEE bands by industry with the reasons behind them, and gives you a way to set a benchmark that fits your line instead of somebody else’s. If you need the mechanics first, start with the OEE calculation method.

Where does the 85% world-class OEE number come from?

It comes from one book. Seiichi Nakajima, the engineer who formalized Total Productive Maintenance, published Introduction to TPM in 1988 with three component targets: 90% availability, 95% performance, and 99% quality. Multiply them, 0.90 × 0.95 × 0.99, and you get 84.6%, which the industry rounded to 85% and started calling “world-class.”

Two things about that origin matter. First, it is a target product, not a measured average. Nakajima was not reporting what the best plants achieved; he was proposing what a well-run discrete line should aim for. Second, no standards body certifies it. ISO 22400-2 the international standard that defines OEE and 33 other manufacturing KPIs, gives you the formula and the time-state definitions, it sets no benchmark and blesses no threshold. So when someone says your line “should” hit 85%, they are quoting a 1980s engineering target for a specific kind of plant, not an audited industry standard.

Is 85% OEE realistic for every industry?

No. The 85% target assumes a plant whose losses are mostly avoidable, breakdowns you can prevent, speed you can recover, defects you can eliminate. High-volume discrete assembly fits that picture: long runs, few changeovers, stable products. Push those levers hard enough and 85% is reachable.

Most plants are not that plant. A dairy line spends real hours every day on sanitation and clean-in-place that no maintenance program removes. A contract job shop changes over between unrelated parts several times a shift because that is the business. A pharmaceutical line runs validated cleaning between products because a regulator requires it. Those hours are not breakdowns; they are the cost of doing that kind of work. Holding those plants to a discrete-assembly target measures how different their work is, not how badly they run it. That is also why OEE versus TEEP matters, how you draw the time boundary changes the number as much as anything on the floor.

What are realistic OEE bands by industry?

Realistic OEE bands cluster by the structure of the work, not by how good the plant is. The ranges below are commonly cited across the field, treat them as orientation, not audited law, because measurement method moves any of them by ten points. What is not debatable is the pattern: the more changeovers, cleaning, and small lots the work requires, the lower the achievable ceiling.

Why do process and job-shop plants score lower by design?

Because their unavoidable losses are large and built into the product. Three structural forces set the ceiling before anyone touches a wrench:

• Cleaning and sanitation time. Food, beverage, dairy, and pharma lines lose hours per day to clean-in-place, allergen changeovers, and validated cleaning. That time is planned production time the line cannot produce during, it lands squarely in the availability factor, and no reliability program removes it.
• Changeover-to-run ratio. A job shop running lots of 50 parts spends a bigger share of the shift setting up than running. Even world-class setup reduction only shrinks that share; on small lots, changeover can genuinely exceed run time.
• Batch and startup dynamics. Process lines ramp, stabilize, and purge. First-off and last-off product may be scrap by specification. Those are first-pass yield losses the recipe requires, not defects a better operator prevents.

The honest move is to separate the two questions OEE quietly blends. One: how much of my loss is structural (the cost of this kind of work)? Two: how much is avoidable (breakdowns, unforced slow-downs, scrap I could stop)? A 55% OEE food line with 20 points of unavoidable cleaning and a well-run avoidable loss is in better shape than a 70% discrete line hiding soft cycle times. The number alone will not tell you which is which, the decomposition will.

How should you set your own OEE benchmark?

Set it off your own line’s demonstrated best, then chase your own trend. Here is the seven-step version that beats quoting 85% at a plant it was never meant for:

1. Fix the measurement method first. Lock the time-state definitions, the ideal cycle times, and what counts as good, before you record a single benchmark. A benchmark on shifting definitions is noise. If manual logs undercount minor stops, wire counts and stops to the source, the way machine monitoring does.
2. Baseline four to eight weeks. One good shift is luck; a month of shifts is a baseline. Capture the normal range, including the bad days, so you know the real starting point.
3. Segment by product and line. A plant-wide average of unlike lines is a number nobody can act on. Benchmark each line, and within a line, each major product family, since changeover and speed differ by SKU.
4. Split structural loss from avoidable loss. Tag cleaning, validated changeover, and specification scrap as structural. What is left is your real improvement runway. Set the target against the runway, not the total.
5. Set the target off your best demonstrated shift. Your best real shift on that product proves what the line can do. A target 5–10 points above the recent median, anchored to that proof, motivates. A borrowed 85% demoralizes.
6. Decompose which factor to attack. Availability, performance, or quality, one usually dominates the gap. Aim improvement there, and use the six big losses to name the specific loss.
7. Re-baseline when the mix changes. New product, new line speed, new constraint, the benchmark moves with them. A stale benchmark quietly punishes or flatters the crew for changes they did not make.

What does the data actually say about where plants run?

Two reference points worth keeping on the table, stated with their provenance:

• The 85% “world-class” figure traces to Seiichi Nakajima’s Introduction to TPM (Productivity Press, 1988): 90% availability × 95% performance × 99% quality. It is a discrete-manufacturing target, widely cited, never audited or certified by any standards body.
• OEE has a real standard behind its definition ISO 22400-2:2014 but that standard defines the formula and time states, not a benchmark. There is no ISO-blessed “good” score.
• For macro context, the U.S. Federal Reserve’s G.17 Industrial Production and Capacity Utilization release has shown manufacturing capacity utilization running in the mid-70s percent, a different metric than OEE, but a reminder that real plants operate well below theoretical maximums by any measure.

What OEE score should you actually chase?

Chase a trustworthy number that is trending up on the factor you chose to attack. A well-measured 58% climbing two points a quarter, with cleaning and changeover honestly separated out, beats a flattering 80% built on soft cycle times and changeovers quietly excluded. The value of OEE is direction and decomposition, not the badge. For the full picture of what a defensible score looks like, read what counts as a good OEE score and put your own inputs through the calculation method.

All of that depends on measuring the same way every shift, which is exactly where end-of-shift spreadsheets fall down and source-connected measurement wins. Harmony computes OEE from machine signals and operator entries in real time rather than from morning-after estimates, so the benchmark you set is the benchmark you actually track. You can see how that played out on a real floor in the CLS case study or walk the module map on the features section of our homepage.