The first automated OEE number almost always lands lower than the one on the whiteboard. Not a little lower. Sometimes twenty points lower. And the reaction is nearly universal: the system must be wrong, because the shop knows it runs better than that.
The shop doesn’t run worse than the whiteboard said. The whiteboard was never measuring the same thing.
The clipboard wasn’t lying, it just couldn’t see
Nobody hand tracking OEE is trying to inflate the number. A supervisor filling in a shift report is doing their best with what they can observe, and what a person can observe from the floor is a small fraction of what actually happens at a machine over eight hours. You notice the big stop, the one where you had to walk over and see what was wrong. You don’t notice the ninety second pause four times an hour that never rose to the level of a problem. You round a cycle time to “about what it usually runs,” not to the tenth of a second the control saw. You know the last part run was good, but you don’t know if six earlier parts drifted out of spec and got caught downstream instead of on the machine.
None of that is deception. It’s the limit of a human standing on a floor with forty other things to track. A clipboard number is built entirely out of what someone happened to notice, remember, and write down, and every one of those three steps loses information. The number that comes out the other end is optimistic because of how it’s built, not because anyone meant it to be.
What automated measurement catches that a person can’t
Turn on machine level data and the number drops because the losses were always there, they just weren’t visible before. A control reads run state every second, not once a shift. It catches the ninety second pause that never became a work order. It catches a cycle time that’s crept up two percent over three months, too slow to notice day to day, plenty fast to cost a full shift of capacity by the end of a quarter. It catches a changeover that ran twenty five minutes when the standard says fifteen, because nobody was standing there with a stopwatch, they were running the next job.
The quality side moves too. A clipboard quality number usually comes from final inspection or a customer return, which means the OEE input for quality was a downstream number all along, not what actually happened at the machine. Once scrap gets logged where it occurs, the quality factor starts reflecting reality instead of whatever made it through the rest of the process.
None of this is a scandal. It’s just what happens when a fifteen digit sensor replaces a person’s memory of an eight hour shift.
Why the lower number is the one worth trusting
A number that flatters you can’t be acted on. If the whiteboard says 78% and the plant believes it, there’s no felt urgency to fix anything, the number’s already good. The 51% automated reading isn’t worse news, it’s the first true description of where the capacity actually went, and every point of that gap is capacity you already own. Where a typical plant and a world class one actually land, and why the gap is worth chasing, is covered in the OEE benchmarks guide, the short version is that most shops run in the 40 to 60% range even with accurate measurement, so a first automated reading in that band isn’t a red flag, it’s a normal starting line.
The number to be suspicious of is a suspiciously good one. A shop that goes live on automated measurement and shows 85% in week one almost certainly has a standard set too loose to catch anything, an ideal cycle time padded to match reality instead of the spec, or a threshold reclassifying real downtime as planned. That number feels better and does nothing for you. A 51% you can build a plan against. A soft 85% just tells you the tool agreed to stop looking.
What the first accurate month is actually for
It isn’t a scorecard. Nobody should walk out of week one with a grade. It’s a map of where the losses are, and the only useful response to a map is to read it, not argue with it.
The mistake most plants make in month one is comparing the new number backward, against last year’s clipboard average, and treating the gap as a decline. That’s comparing two different instruments and calling the difference a trend. There is no trend yet. There’s a baseline. The comparison that matters starts in month two, against month one, on the same instrument.
Watch what actually happens: a machine reads 47% in week one. The Pareto shows the single biggest loss is a changeover that runs long on one specific job, twenty minutes over standard, four times a week. Someone fixes the fixture that’s causing the delay. By week three that machine’s at 54%, not because anyone worked harder, because the number finally showed where to look and someone looked. That seven point move is real, it shows up on the next invoice as parts made, and it happened because the number pointed at something specific instead of hiding it behind a rounded end of shift guess.
Part of what makes that seven points trustworthy is what the chart does with a gap. A machine it hasn’t measured yet shows a dash, not a hundred percent, and a shift with nothing to report leaves the chart instead of showing a zero. A number worth acting on has to be a number that wasn’t invented to fill space, and that’s what makes month two’s seven point gain mean something instead of noise from two different measurement standards.
What to actually do in month one
Don’t chase 85%. For most shops that’s the wrong target for years, maybe forever, especially with high mix, low volume work where setups eat into availability no matter how well the shift runs. Instead, pick the top one or two losses by dollars, not by how loud they are, and go after those specifically. A slow but steady changeover problem worth four hours a week is a better first target than a dramatic breakdown that happens twice a year, even though the breakdown feels more urgent.
Resist the urge to explain away the first number. “That reading’s wrong, we know this machine runs better than that” is usually the sound of a real loss being noticed for the first time, not a sensor malfunctioning. Give it two weeks before deciding the number’s broken. Most of the time it settles in exactly where the floor already suspected, once the floor stopped rounding in its own favor.
Quick recap
- Automated OEE reads lower than a clipboard number because it sees everything, not because anything got worse
- Micro stops, cycle drift, and downstream scrap were always there, they just weren’t visible before
- Trust the low number over a suspiciously high one, a soft 85% means nothing
- Month one is a baseline, not a score, compare month two against month one, not against last year’s guess
- Chase the top losses by dollars, not by drama, and expect the first accurate number to hold up under scrutiny