Walk onto almost any plant floor and you will find a dashboard glowing with OEE numbers — 82% here, 88% there, a line manager quietly relieved the figure crept up two points this quarter. Ask a harder question and the room goes quiet: did throughput actually increase? Did more sellable product leave the building? Often, nobody knows. The number went up and the trucks did not get fuller.
That gap is not a measurement error. It is the metric working exactly as designed — and the design is wrong for the job most plants are using it for.
Throughput is the goal. OEE is a proxy that forgets it.
Eliyahu Goldratt settled this argument forty years ago in The Goal: the purpose of a manufacturing system is to make money, and on the floor that means throughput — the rate at which the system generates sellable output. Everything else is a local measure that only matters if it moves that rate. An hour saved at the bottleneck is an hour of throughput added to the whole plant. An hour saved anywhere else is a mirage. It produces work-in-process that piles up in front of the constraint and ties up cash.
OEE does not know where your bottleneck is. It reports the same 88% whether the machine is the constraint or sitting idle-rich behind it. Average it across a facility and it gets worse: an 85% plant figure can hide a constraint running at 70% and an over-built station running at 95%, and the average tells you to celebrate. Used as a target, it gets gamed — run the non-bottleneck harder, push the number up, and watch throughput stay flat while inventory climbs.
The point is not that OEE is useless. Calculated at the constraint, availability-times-performance-times-quality is a sharp diagnostic. The point is that as a headline, plant-wide score, it obscures the one thing that pays — and it weights every machine as if they all mattered equally. They do not.
The real lever is variability, not utilization
If throughput is the goal, the question becomes: what suppresses it? Hopp and Spearman answered that in Factory Physics with a law most operators feel but rarely name — the variability buffering law. Variability in a production system is always paid for, in one of three currencies: inventory, capacity, or time. Reduce variability and throughput rises for the same buffers. Increase it and the system degrades no matter how high any single machine's utilization reads.
This is why the plants that win are not the ones chasing the last efficiency point on every asset. They are the ones eliminating the constraint and then attacking the variation that feeds it — unplanned downtime, changeover scatter, quality escapes, the setup that takes nineteen minutes one shift and forty the next.
And this is where standard work stops being a binder on a shelf and becomes the actual mechanism of improvement. A standard is not bureaucracy. It is the lowest-cost variability reducer available. Every cycle made identical is variance removed from the system — and by the buffering law, variance removed is throughput returned. In an era where everyone wants to bolt sensors and models onto the line, disciplined standard practice matters more, not less.
Why this is the order of operations
Here is the trap. A line with an unstable process and a misplaced constraint does not have a data problem. It has a process problem wearing a data costume. Automating it — adding the dashboard, the sensors, the model — does not fix the variation. It instruments it. You spend the capital and end up with a high-resolution picture of the same chaos.
Solve the problem first. Find the constraint. Reduce the variation feeding it. Standardize the cycle. Then automate — because automation laid over a solved process compounds, and automation laid over a broken one only scales the breakage.
That sequence is not a slogan. It is the reason ForgeShift starts every engagement with the ForgeShift Compass: before any technology recommendation, you find where the real constraint is and how stable the process around it actually is. The plant floor is where the next industrial revolution gets forged — from the floor up, in that order. Get the order wrong and the most advanced tooling in the world just helps you make the wrong amount, faster.