OEE in plain English, and the ways it gets flattered.

Overall Equipment Effectiveness is one number made of three, and it is popular for a reason: it is the only single figure that tells you, of the time you paid for, how much of it turned into good parts at the rate you expected. It is also the number most often quoted without anybody being able to say where it came from — which is why an OEE conversation so often turns into an argument about definitions rather than about machines.

The formula is Availability × Performance × Quality. All three are ratios, all three are between zero and one, and multiplying them is what makes OEE unforgiving. Three numbers that each look respectable on their own produce a total that does not.

Availability: did the machine run when it was supposed to?

Availability is run time divided by loading time. Run time is what it sounds like. Loading time is the argument.

Loading time is the period you intended to produce in — your real shift pattern, minus the time you have deliberately chosen not to count. Planned maintenance, breaks, a scheduled shutdown: most sites take those out, because holding a machine responsible for a shift nobody staffed is not useful. Everything left is availability loss: breakdowns, waiting for a setter, waiting for material, a changeover that ran over, and the fifteen minutes at the start of a shift that nobody has ever written down.

Two rules keep this honest. First, whatever you take out of loading time must be a category you defined in advance and can point at afterwards. Second, a window with no production in it should say no production in this window, not report zero per cent. A run of zeroes across a Saturday night drags a weekly average down and tells you nothing, and once it is in a board pack somebody will try to explain it.

Performance: did it run at the rate it should have?

Performance is what the machine actually produced against what it should have produced in the same run time, at its standard cycle time. It catches the losses availability cannot see: running slow, micro-stops too short to be worth an event, a mould that is not quite up to temperature, an operator who is covering two cells.

All of it rests on the standard cycle time, and that is the softest number in the whole calculation. If the standard came off a quotation five years ago, or was set by whoever had the stopwatch, performance is measuring the estimate rather than the machine. It is worth knowing where each of your standards came from before you defend a performance figure with one.

A machine beating its standard produces a performance figure over 100%, and most systems quietly cap it. That cap hides useful information: a press consistently running at 108% is telling you the standard is wrong, which is worth more than a tidy number. Reporting both — raw and capped — costs nothing and settles the argument.

Quality: how much of it could you sell?

Quality is good parts divided by total parts made. The complications are rework and concessions. A part that needed twenty minutes of hand-finishing before it passed is not free, and counting it as good makes quality look like 100% while the cost sits in somebody else’s budget. Decide which side of the line rework falls on, write it down, and keep it there — a definition that moves between months is worse than one you disagree with.

The arithmetic, on a ten-machine shop

Take ten presses running two shifts, five days a week: 80 machine hours each, so 800 scheduled hours in the week. Forty of those are planned maintenance and breaks, leaving 760 hours of loading time. Breakdowns, material and overrunning changeovers account for 76 hours, so run time is 684. The standard cycle is 36 seconds, which is 100 parts an hour. Actual output is 58,000 parts and 1,160 of them were scrapped.

Nothing in that shop is broken. Every component looks like a number you would be reasonably pleased with in isolation, and the product of them is 74.8%. That is the whole point of multiplying: OEE is deliberately hard to flatter by fixing one thing.

Which is exactly why people flatter the inputs instead.

Six ways an OEE figure gets improved without touching a machine

None of these is fraud. Every one of them is a defensible-sounding decision that somebody made for a sensible reason, and each one moves the number in the same direction.

• Reclassifying downtime as planned. Move 30 of those 76 unplanned hours into planned maintenance and loading time drops to 730 while run time stays at 684. Availability becomes 93.7% and OEE becomes 77.9%. Three points, from a reason code.
• Slackening the standard cycle. Change the standard from 36 seconds to 40 and the theoretical output falls to 61,560. Performance becomes 94.2% and OEE becomes 83.1% — over eight points from editing one field, and the machine ran at exactly the same speed all week.
• Counting rework as good. Quality goes to 100% and OEE to 76.3%, while the finishing bench works late.
• Dropping empty windows silently. If a system records 0% for a night nobody ran, the fix is usually to exclude quiet periods. That is right — but done by hand it becomes a judgement about which periods were quiet, made after the numbers are known.
• Absorbing unclassified downtime into “other”. If a third of your stopped time has never been given a reason, your Pareto is a chart of the third somebody bothered to classify. An unclassified-downtime figure, reported as a metric in its own right, tells you how much of the analysis is evidence.
• Averaging across machines. A factory OEE of 74.8% can be nine machines at 80% and one at 30%. The average is true and useless; the distribution is where the money is.

Why the honest number is worth more

A flattering OEE has one use: it looks good once. An honest one is something you can act on every week, and the difference shows up the first time somebody senior asks why. A figure that can be traced back to a machine-state ledger with timestamps survives that question. A figure assembled from a spreadsheet of classifications does not, and the moment it fails once, nobody believes the next one either.

That is the standard we hold SmartFactory to, and it is why the product will show you a caveat rather than a clean figure when a sensor dropped out mid-window, report raw and capped performance side by side rather than reconciling them, and put an unclassified-downtime figure next to your Pareto. It is also why an empty window says so. A system that quietly fills gaps with plausible numbers is worse than no system, because you act on it.

If you want the mechanics — how machine states are derived rather than declared, how a downtime event opens itself, how changing your working calendar re-states history — that is on the monitoring page. If you would rather put one of your own shifts through it, the OEE calculator does what this guide describes — the three components, raw and capped performance side by side, and an empty window that says so instead of reporting 0%. If you have got as far as needing a number with a pound sign on it, the ROI calculator takes your hours and rates and tells you what a single percentage point is worth, and the business case guide covers how to run a pilot that proves it.