Overall equipment effectiveness: availability × performance × quality
| OEE | Reading | Typical factor mix |
|---|---|---|
| 100% | Perfect production | Theoretical: only reachable on paper |
| 85%+ | World class (OEE.com convention) | ≈ 90% availability × 95% performance × 99.9% quality |
| ~60% | Typical first honest measurement | Realistic baseline for most plants |
| ~40% | Low / data problems | Unmeasured changeovers or minor stops |
| Availability | Performance | Quality | OEE |
|---|---|---|---|
| 90% | 90% | 90% | 72.9% |
| 95% | 85% | 95% | 76.7% |
| 85% | 95% | 90% | 72.7% |
| 90% | 95% | 99.9% | 85.4% |
The second table is the multiplication lesson: three 90s average to 90 but multiply to 72.9%. The benchmark bands are the widely used conventions popularized by OEE.com (Nakajima's framework), not regulatory standards.
OEE compresses the three ways a machine loses output into one number: it wasn't running (availability), it ran slow (performance), or it made bad parts (quality). Seiichi Nakajima developed the metric at JIPM in the 1970s, and it remains the standard yardstick for equipment effectiveness in discrete manufacturing.
Availability = run time ÷ planned production time, where run time = planned time − downtime. Performance = (ideal cycle time × total count) ÷ run time. Quality = good count ÷ total count. OEE = A × P × Q. The count shortcut gives the same answer: OEE = (good count × ideal cycle time) ÷ planned time.
Take one shift. Planned production time is what the schedule says the line should run, not the shift length; breaks and meetings you planned out of the schedule don't count. Downtime is everything that stopped the line inside that planned window. The ideal cycle time is the machine's nameplate or best-demonstrated rate, not today's average, and the counts are total and good parts including startup scrap.
One 8-hour shift schedules 480 minutes of production and loses 47 minutes to a jam and a changeover, leaving 433 minutes of run time: availability is 433 ÷ 480 = 90.2%. The line produces 260 parts at a 1.5-minute ideal cycle, needing 390 minutes of perfect running, so performance is 390 ÷ 433 = 90.1%. Of the 260 parts, 248 pass first-pass inspection: quality is 248 ÷ 260 = 95.4%. Multiply: 0.902 × 0.901 × 0.954 = 77.5% OEE.
The shortcut confirms it: 248 good parts × 1.5 minutes = 372 minutes of effective production ÷ 480 planned minutes = 77.5%. At 100% OEE the same shift would have made 320 good parts (480 ÷ 1.5), so the three loss buckets cost 72 parts. Getting from 77.5% to 85% is worth roughly 9.7% more good output (85 ÷ 77.5), often the cheapest capacity a plant can buy.
The long-standing benchmark set popularized by OEE.com calls 85% world class, built from roughly 90% availability, 95% performance, and 99.9% quality. Most plants that start measuring honestly land near 60%, and sites with unreliable data or unmeasured changeovers often show 40%. Chasing 85% everywhere is the wrong move; a bottleneck line justifies aggressive targets, while a low-utilization backup line may not justify the effort.
Because losses compound. Three factors at 90% each multiply to 72.9% overall, not 90%. A 10% loss in availability, another in performance, and another in quality leaves you with 27% of perfect production gone. That multiplicative structure is exactly why OEE is useful: it makes small compounding losses visible in a single number.
Both measure the same three factors, but against different denominators. OEE divides by planned production time, so scheduled downtime, holidays, and shifts you never planned to run don't count against it. TEEP (total effective equipment performance) divides by all calendar time, exposing the capacity you're choosing not to schedule. A line can run 85% OEE on two shifts and still show a modest TEEP because 16 hours a day are unplanned.
If the shift was planned as production time, yes: a changeover during a planned run reduces availability, which is the point of tracking SMED improvements. Only time formally removed from the planned schedule, like a plant-wide shutdown week, belongs outside the denominator. Fudging changeovers out of the calculation is one of the fastest ways to make OEE lie.
Yes, and it's the cleanest method: OEE = (good parts × ideal cycle time) ÷ planned production time. The three-factor breakdown is diagnostic, but the shortcut collapses it correctly, because performance and quality losses both show up as missing good parts. The shortcut only needs an honest ideal cycle time, real counts, and real planned time.
Because the first weeks of data replace assumptions with measurements. Unlogged minor stops, slow cycles the schedule absorbed, and startup scrap that never hit the yield report all show up at once. The drop isn't the plant getting worse; it's the plant becoming visible. Baseline after about a month of clean data, then trend from there.