The dashboard says OEE is 85 percent. The plant manager smiles. The finance team asks why the line still doesn’t ship enough product. They’re both right, because the OEE number is only as good as the assumptions behind it.

What OEE actually measures

OEE is availability times performance times quality. A line that runs 80 percent of scheduled time, at 90 percent of rated speed, making 98 percent good parts gets 0.8 × 0.9 × 0.98 = 70.6 percent. World-class is around 85. Most plants are in the 50 to 60 range.

The problem isn’t the formula. It’s how you feed it.

Availability: the scheduled-time trap

Availability is run time divided by planned production time. If you define planned time as eight hours and the line only runs for six because two hours were lost to a missing operator or a late material delivery, availability is 75 percent. But many plants quietly reclassify those two hours as “not planned,” because material shortages aren’t the machine’s fault. Suddenly availability is 100 percent, and the line still shipped 25 percent less.

This is the most common way OEE lies. Every plant has its own definition of planned time. Compare two lines on OEE and you’re comparing accounting choices, not performance. Pick one definition and stick to it, even when it makes the number look worse.

Performance: the speed nobody checks

Performance is actual output divided by the output you should have made at rated speed. The rated speed comes from the machine datasheet. Real machines run slower than the nameplate, especially as they age. If you divide by a theoretical rate that no one has ever measured, performance always looks low, and teams start adjusting the denominator until the number looks reasonable.

The fix is to measure the realistic best speed: run the line at full output for a sustained hour, with a good operator, known-good material, and no stops. Use that number as the baseline, not the catalog. Then performance reflects real degradation, not fantasy.

Quality: the hidden losses

Quality is good parts over total parts. The trap here is rework. A part that comes off bad, gets reworked, and ships as good gets counted as good, even though it consumed extra time and material. True first-pass yield is what you want, not shipped-part quality. If your OEE doesn’t track rework separately, you’re overstating quality.

Start-up losses are the other blind spot. The first 20 minutes of every shift produces scrap as the machine warms up and the operator resets parameters. That scrap often doesn’t hit the quality metric because it’s recorded as downtime instead. Either way, it’s lost output. Count it once, in one place.

Why a high OEE can still lose money

OEE measures efficiency on the line, not profitability. A line running at 85 percent OEE making a product that sells at a thin margin, on a shift that wasn’t needed, loses money. OEE doesn’t know about demand. It tells you how well you’re running, not whether you should be running at all.

Plants that chase OEE as the only metric will optimize the wrong things. They’ll run long campaigns to reduce changeover downtime, building inventory that the warehouse can’t sell. They’ll avoid stopping for maintenance, because downtime hurts the number, until the machine breaks for eight hours instead of four.

What to track instead

Use OEE to find where time is going, not to grade the shift. Break it into the six big losses: changeover, minor stops, speed loss, start-up scrap, production scrap, and downtime. The biggest loss usually isn’t the one the dashboard highlights. A line with 70 percent OEE might be losing 15 points to minor stops, which nobody has even investigated because the dashboard only shows the total.

Pick the largest loss category and attack it for a month. Then pick the next. Don’t try to raise OEE by 10 points all at once; attack one loss at a time, and the number moves on its own.

Bottom line

OEE is a diagnostic, not a score. If the number looks good but shipments don’t follow, audit the definitions: planned time, rated speed, rework, and start-up scrap. Most plants find they’ve been measuring something that looks like OEE but doesn’t reflect what the line actually lost.