OEE explained: three factors, six losses, and the 85% rule of thumb

OEE — overall equipment effectiveness — answers one question: of the good units this asset could have made in the time you gave it, how many did you actually get? It is a single percentage, and that is both why it spread through manufacturing and why it gets misused. A single number cannot tell you which loss to attack, and the three factors underneath it can, which is why nobody should ever report OEE without them.

Overall equipment effectiveness (OEE)

The proportion of planned production time that was genuinely productive — running at full speed, producing units that passed first time. Calculated as availability × performance × quality, each defined against planned production time rather than calendar time.

The three factors, defined precisely

• Planned production time — the time the asset was scheduled to make product. Take out breaks, planned maintenance, unstaffed shifts and periods with no demand before you start. Charging a scheduled PM against availability penalises the maintenance you want people to do.
• Run time — planned production time minus every unplanned stop, including the two-minute jams nobody logged.
• Ideal cycle time — the fastest cycle the asset has ever sustained for this part. Not the routing standard, which carries allowances. This one input causes more bad OEE numbers than everything else combined.

A worked example: one shift, one asset

An 8-hour shift on a packaging line. Two 15-minute breaks and a 20-minute scheduled changeover are planned downtime, so 50 minutes come out before the clock starts.

Availability = 378 ÷ 430 = 87.9%. Performance = (0.20 × 1,674) ÷ 378 = 334.8 ÷ 378 = 88.6%. Quality = 1,617 ÷ 1,674 = 96.6%. Multiply the three: 0.879 × 0.886 × 0.966 = 75.2%.

Turn the percentage into units, then into money

Nobody argues with a unit count. In 430 minutes of planned production at 0.20 minutes a unit, the line could have made 430 ÷ 0.20 = 2,150 good units. It made 1,617. The 533-unit gap splits cleanly across the three factors:

At $9.40 of contribution per unit, 533 units is $5,010 of margin left on the floor in one shift. Across 480 shifts a year — two shifts a day, 240 production days — that is 255,840 units and about $2,404,896, assuming the loss rate holds. That assumption is doing real work, so state it, but a number like that is what gets a capital request read. Price your own losses per hour with the downtime cost calculator.

Why performance above 100% means your ideal cycle time is wrong

This is the single most common OEE error, and it always has the same cause. Performance compares actual output against the theoretical maximum at the ideal cycle time, so exceeding 100% means the asset produced faster than the number you called ideal — which means the number is not ideal. Almost always someone used the routing standard, padded with allowances for setup, fatigue and expected minor stops. Take the same shift and enter that standard of 15 seconds instead of the true 12-second ideal. Performance becomes (0.25 × 1,674) ÷ 378 = 418.5 ÷ 378 = 110.7%, and OEE reads (0.25 × 1,617) ÷ 430 = 404.25 ÷ 430 = 94.0%. A line genuinely running at 75.2% reports 94.0%, the 216-unit speed loss vanishes from the chart, and the improvement team goes to work on a machine that is already excellent.

The six big losses, and which factor each one lands in

The three factors are how you calculate OEE. The six big losses are how you fix it. Every OEE point you lose belongs to exactly one of these, and each one has a different owner.

On the example shift, halving unplanned downtime from 52 to 26 minutes raises run time to 404 minutes and availability to 404 ÷ 430 = 93.9%. Hold performance and quality steady and OEE rises to 93.9% × 88.6% × 96.6% = 80.4% — 5.2 points, and 111 more good units per shift. That is the useful form of an OEE improvement plan: a specific loss, a specific minutes reduction, and a unit count attached.

The 85% figure, and what it is actually worth

You will be told that 85% OEE is world class and 60% is typical. Treat that as a rule of thumb, not a statistic. It circulates widely in lean and TPM material without a dataset behind it, and its origin is visible in the arithmetic: 90% availability × 95% performance × 99.9% quality = 85.4%. It is the product of three aspirational component targets, not a measured distribution of plants.

It also assumes high-volume discrete production of a single part. A job shop running 40 changeovers a week is spending availability on the flexibility its customers pay for: judged by the 85% bar it looks broken, judged against its own trend and its quoted lead times it may be excellent. Use the figure to open a conversation, never as a target handed down to a line.

TEEP: the calendar-time variant

OEE judges the asset only during the hours you scheduled it, which makes it silent on the question managers most often want answered: do we need another machine? TEEP — total effective equipment performance — closes that gap by measuring against all calendar time.

TEEP = OEE × Utilization
Utilization = Planned production time ÷ Calendar time

The example asset ran one 8-hour shift, so 430 minutes of planned production out of 1,440 calendar minutes gives utilization of 29.9%, and TEEP = 75.2% × 29.9% = 22.5%. Add a second shift and planned production time becomes 860 minutes, utilization 59.7%, and TEEP 44.9%. Before signing a capital request, that comparison is the honest one: the second shift is available capacity you already own, and it costs labour rather than capital.

Four ways an OEE programme goes wrong

• Averaging OEE across machines. The mean of five machines' OEE describes nothing physical. Only the constraint's OEE converts into shippable units, so raising OEE on a non-constraint machine improves a chart and not the output.
• Charging planned downtime to availability. It makes the number look bad for the wrong reason and teaches supervisors to skip PMs to protect the metric. Planned time comes out of the denominator; it reappears in TEEP where it belongs.
• Counting rework as good output. Quality is first-time-through. Reclassifying rework as good is the single easiest way to make an OEE number rise without anything improving.
• Using OEE as an operator scorecard. Availability is mostly maintenance and material supply, and performance is mostly tooling and machine condition. Tie OEE to individual performance reviews and the downtime log becomes fiction within a month — which destroys the data before it destroys the trust.

Getting the inputs, honestly

Be straight about where OEE data lives. Total count and good count are usually in the ERP as work-order completions and scrap transactions. Ideal cycle time is on the routing, often wrong. Downtime minutes by reason are the hard part: unless the line is instrumented, they sit on a paper log, and short stops are simply absent from it. That absence is why performance so often shows up as the largest loss — the minutes were real, they just never got recorded as downtime.

You can still start. Take one week on the constraint, log stops by reason with a stopwatch and a clipboard, compare the counts against the ERP completions, and you have a defensible baseline. From there the reporting becomes routine: ask ERPray for "good and scrap quantity by work centre and shift for the last 12 weeks" and it computes the answer live from your own account with the query shown underneath, so your CI engineer argues about the definition rather than whose export is current. The scrap and yield calculator works the quality leg from the same data; purchase price variance explained covers the same trap on the buying side; and inventory turnover: how to read it covers what lost constraint output does to stock.