Your Floor Reports 84 Percent And Runs Fifty

The 84.9 percent that was really 50

A meat plant we work with reported 84.9 percent efficiency. The facility manager stood behind the number and, for a while, withheld the raw line data that would have let anyone check it. When we finally got instrumentation onto the lines, actual performance sat closer to 50 percent. That is not a rounding error. That is a plant running at a little more than half of what its own reporting claimed, with every downstream decision, staffing, capital, and promised delivery dates, built on top of the higher figure.

This is the simulation gap: the distance between what a plant believes about itself and what its floor is actually doing. It is not usually fraud. It is a reporting system that rewards a clean number, an OEE calculation with a generous definition of "available," and a floor culture where nobody has an incentive to find the real figure. The number becomes a belief, and the belief gets defended.

Why the reported number and the real constraint drift apart

The deeper problem is that even an honest OEE number tells you less than you think. Throughput is not the speed of your fastest station or the rating on the slowest machine's spec sheet. It is a property of how the stations interact: how work arrives, how buffers fill and starve, how a changeover on line three strands labor on line four, how one micro-stoppage pattern cascades. System interaction governs throughput. No single station's rated speed does.

That is why the bottleneck is almost never where the floor says it is. We have walked into plants that were certain they were running near the top of their category and found the real constraint two stations upstream from where every manager was looking. The station that looks slow is often just the one that is visibly backed up; the actual constraint is the one quietly starving it. Add crew to the visible one and you have spent labor to make the pile in front of it move faster while total output does not budge.

You cannot reason your way to the real constraint by standing on the floor, because the interactions are faster and more tangled than intuition tracks. You have to measure the system and then model it.

Measure the system, then model it before you touch it

The sequence that closes the gap is boring and it works. Drop lightweight wireless trackers on the production lines; you can instrument a facility in a day with almost no disruption. Pair that with a 3D scan of the floor so the layout, distances, and buffer positions are captured, not remembered. Now you have a digital twin: a model of the plant that behaves like the plant.

Then you simulate. On one protein line we ran roughly 300 scenarios against that model and narrowed them to five viable operating configurations. One of them was a $1.4M labor path that also lifted throughput 14 percent, a combination no one on the floor had named because you cannot see it by watching. The model can hold every interaction at once; a person standing at one station cannot.

The discipline here is to model before you act, not after. This is the same logic that should gate any capital line decision: you validate the constraint and the expected gain in a model you trust before you reallocate a crew or sign a purchase order, not six months and a few hundred thousand dollars later. The trackers and the twin collapse what used to be a six-month factory study into a couple of hours of simulation, which is what makes modeling first affordable enough to be the default instead of the exception.

Run this and the labor decision changes shape. Instead of "the packout station looks slammed, add two people," you get "measured labor minutes per thousand units say the constraint is upstream, reallocate one operator there and pull one from the station that only looked busy." Same headcount, more output, and a number you can defend because it came from a measurement, not a belief.

What a well-run floor looks like

The reported OEE matches a hand-counted sample within five points, and when it does not, the floor treats the gap as the first thing to fix rather than a number to defend. The named constraint is the station that is measurably starved or blocked, confirmed against tracker data, not the one that looks busiest. Downtime minutes and scrap are counted, not estimated. Headcount is set from labor minutes per thousand units at the real constraint. And no crew gets added and no line gets bought until the change has been run through a model of the actual floor.

The savings were sitting in plain sight

The 84.9 percent was not a lie anyone told on purpose; it was a belief the reporting system protected. The savings and the throughput were there the whole time, two stations upstream from where everyone was looking. The plant did not need to work harder. It needed to stop trusting the number and measure the system that produced it.