The OEE that gets reported is not the OEE that exists
Overall Equipment Effectiveness — Availability × Performance × Quality — is the single most-quoted manufacturing KPI and one of the most-fiddled-with. We've audited a lot of factory OEE numbers. The real number is almost always 10-30 percentage points lower than what gets reported. The reasons are systematic, and they're worth understanding before relying on the metric.Availability: the easiest number to lie with
Availability = run time / planned production time.The way it breaks:
- "Planned downtime" gets generously defined. Cleaning, changeovers, breaks, "scheduled" PMs — all booked outside the denominator. The metric improves; the line doesn't.
- Short stops aren't counted. A 2-minute jam every 15 minutes doesn't show up if the data only captures stops > 5 minutes. But it adds up.
- Manual logging. Operators are busy; downtime gets categorised after the fact, often miscategorised toward "planned".
Fix: define planned downtime restrictively. Capture every stop, regardless of duration, automatically from the PLC.
Performance: where the rounding errors live
Performance = (actual rate × actual run time) / (ideal rate × actual run time) = actual rate / ideal rate.The way it breaks:
- Ideal rate is set conservatively. If the rated speed is 60 ppm but operators are pushed to keep the line at 50 ppm, performance looks 100 % even when the line is consistently slower than designed.
- Average rate masks micro-stops. If the line stops for 30 seconds every 5 minutes, the average over the hour looks fine; the actual experience is 6 micro-stops per hour.
Fix: ideal rate is the design rate, not a "negotiated" rate. Use the actual cycle time per part, not an averaged throughput.
Quality: usually the cleanest, sometimes hidden
Quality = good parts / total parts.The way it breaks:
- Rework counted as good. If a part fails inspection, gets reworked, and passes the second time, was that one part or two? Different factories answer differently.
- First-pass yield vs final yield. If you're reporting "quality", be clear which one. The first-pass-yield number is harder, and harder is better.
Fix: define first-pass yield strictly. Track rework separately. Don't conflate.
The honest OEE setup
- Automatic capture of every state change (PLC-tagged) — running, faulted, idle, blocked, starved
- Operator categorisation of unplanned stops (with a max of 8 reason codes — more and they're not used)
- Automatic capture of cycle times, not aggregate throughput
- Quality data from inspection systems, not operator self-report
- Rejection of any "OEE > 90 %" reading without manual investigation — it's almost always an artefact
The losses, ranked by frequency in our audits
- Setup / changeover — 5-15 % of theoretical capacity in mid-volume operations. Almost always the highest-leverage improvement target.
- Speed losses — running below design rate "to be safe". Often invisible.
- Minor stops — < 5 minute stops, almost never logged manually. Add up to 5-10 % alone.
- Quality losses — usually the most-tracked, often not the biggest.
- Major breakdowns — most visible, usually not the biggest either.
If your OEE programme isn't surfacing the first three, it's measuring the wrong thing.
The improvement loop that works
- Weekly review with operations + maintenance + engineering — 30 minutes, fixed agenda
- Pareto of losses for the week
- One specific countermeasure picked, owner assigned, due date set
- Review of last week's countermeasure — did it move the metric?
Boring. Effective. The factories where we've seen this run consistently outperform the ones with the fancy dashboards.
One thing that doesn't help
Reporting OEE up to senior management before fixing the data quality. Bad numbers shape bad decisions. Spend 4-8 weeks on the data pipeline before publishing the trend.One thing that does help
A second metric: TEEP (Total Effective Equipment Performance), which uses calendar time instead of planned production time. Catches the cases where "planned downtime" is generously defined. Useful for senior leadership, awkward for line management — exactly the right pair.What's your OEE methodology? And has anyone successfully driven micro-stop reduction with PLC-based root cause analysis?
