OEE (Overall Equipment Effectiveness) is Availability times Performance times Quality. The math takes thirty seconds. Whether the answer means anything depends on five raw inputs: planned production time, run time, ideal cycle time, total count, and good count. To calculate OEE without fooling yourself, use the nameplate cycle time, count every unplanned stop as downtime, subtract rework as well as scrap from good count, and write the rules down so nobody changes them when the number looks bad. A packaging lead I worked with kept two numbers. The one on the whiteboard, 82 percent, went to the morning meeting. The one in his notebook was a list of every time the line stopped and what he did about it. When I asked why he kept two, he said the board number had never once matched a shift he had actually worked. He was not being difficult. He was being accurate. What each part of the OEE calculation measures Availability. Of the time you planned to make product, how much did the line actually run? Run time divided by planned production time. Performance. While it was running, did it run at the rate it is designed to run? Ideal cycle time times total count, divided by run time. Quality. Of everything it made, how much came off good the first time? Good count divided by total count. Multiply the three. That is it. Nobody gets the OEE formula wrong. What people get wrong is what counts as planned, what counts as running, what counts as ideal, and what counts as good. Where plants make the OEE number look better than it is Three places make the number look better than the shift was, and I have found all three in the same plant on the same day. Planned production time shrinks. Someone decides the changeover was scheduled, so it is not downtime. Then the material wait was the supplier’s fault, so it is not downtime. Then the training hour was a management decision, so it is not downtime. Every exclusion is defensible on its own. Together they take the hard hours out of the denominator, and availability rises without the line shipping one more case. Ideal cycle time gets set to the best shift instead of the nameplate. The nameplate says 1.8 seconds per unit. The line has hit 2.0 on its best shifts, so 2.0 goes into the spreadsheet, because otherwise performance looks bad. Now the gap between the rate the machine was designed to run and the rate it actually runs is gone from the report. That gap was the whole point of measuring performance. Good count includes rework. Scrap gets subtracted, because scrap is visible and someone has to sign for it. Rework does not, because the unit eventually shipped. But a unit that took two passes cost you a second pass. Rework is often the largest loss on a line and the one least likely to appear in the OEE calculation. There is a fourth problem, and it is harder to see. Downtime reasons get typed in at the end of the shift from memory, in whatever categories the software offered five years ago. When four different stoppages all land in “Other” or “Minor stop”, the number can be perfectly accurate and still tell you nothing you can act on. A worked example: one shift, counted two ways An eight-hour shift. 480 minutes. Two 15-minute breaks. One 20-minute changeover. A 45-minute wait on material. 25 minutes of breakdown. 9,000 units made, 150 scrapped, 400 reworked. Nameplate cycle time 1.8 seconds. The version that makes the shift look good. Breaks, changeover and the material wait all get excluded from planned time, leaving 385 minutes. Only the breakdown counts as downtime, so run time is 360. Availability is 93.5 percent. Ideal cycle time is set at the best observed 2.0 seconds, which makes performance 83.3 percent. Good count is 8,850, because rework shipped, so quality is 98.3 percent. OEE 76.6 percent. A number that looks fine in the morning meeting. The honest version. Only the breaks come out of planned time, leaving 450 minutes. Changeover, material wait and breakdown are all downtime, 90 minutes, so run time is still 360. Availability is 80 percent. Ideal cycle time is the nameplate 1.8 seconds, so performance is 75 percent. Good count subtracts scrap and rework, 8,450, so quality is 93.9 percent. OEE 56.3 percent. Same shift. Same product. Twenty points apart. The second number is worth more, because it still shows what the shift cost: 90 minutes of downtime, a line running well under its design rate, and 400 units that had to be handled twice. Audit your own OEE inputs this week Take one shift, any line, and answer five questions in writing. It takes an hour. What time was excluded from planned production time, and who decided each exclusion? Where did the ideal cycle time come from: the nameplate, or the best shift anyone remembers? Does the good count subtract rework, or only scrap? Who enters downtime reasons, and how long after the stop? What is the shortest stoppage the system records, and what happens to everything below it? If any answer is “I would have to ask”, that is the finding. You do not have an OEE problem yet. You have a bookkeeping problem, and it is upstream of every improvement decision you make from that report. What an honest OEE number is worth An inflated OEE score is worse than no score, because people believe it and spend money based on it. Plants buy capacity they already have. They chase a performance loss that was really a changeover loss. They put a line on overtime while the real constraint sits two stations upstream, uncounted, because it never generated a record anyone could see. The packaging lead with two numbers was not hiding anything. He had simply learned that one of them described his shift and the other described a meeting. Whichever one your operators trust is the one they act on, whatever the report says.