OEE Is a Thermometer, Not a Report Card

There's a moment in the life of many OEE programs that decides everything, and it usually looks harmless. A monthly review. A slide comparing lines, or shifts, or crews. A manager — reasonably, by their lights — asks: "Why is B shift ten points behind A shift?"

From the front of the room, it's accountability. From the floor, something else just happened: the number stopped being a measurement and became a grade. And graded numbers, everywhere and always, obey a law worth pinning above every dashboard — when a measure becomes a target that people are judged on, it stops being a good measure. Goodhart's law isn't a management aphorism; on a shop floor it's a schedule of exactly what happens next.

What happens next

None of it is dramatic, and none of it is dishonest exactly. Stops get logged a little shorter. Ambiguous downtime drifts into categories that don't count — "waiting on material" has broad shoulders. The threshold for what's worth writing down creeps up. A crew under scrutiny discovers, without anyone deciding anything, that the rated speed on the hard product was probably set too optimistically anyway.

Each adjustment is small and defensible. Compounded across shifts and months, the program ends up managing a fictional plant. The reports improve; the plant doesn't; and — this is the expensive part — leadership *believes* the reports, so the real losses are no longer just unmeasured but actively invisible. A weaponized metric doesn't merely stop helping. It converts your measurement system into a machine for hiding problems, staffed by people you trained to hide them.

The cruelest part: the blamed number was unfair to begin with. OEE differences between shifts and lines mostly encode things nobody on the crew chose — product mix, machine age, staffing, which shift inherits the Monday startup. Grading people on a number dominated by factors outside their control isn't just corrosive. It's inaccurate.

The thermometer discipline

The alternative isn't abandoning accountability. It's aiming it correctly. A thermometer reading of 39°C starts a *diagnosis* — nobody scolds the patient for the reading, and nobody negotiates with the thermometer. The discipline, concretely:

Interrogate losses, not people. The working question is "what stopped the line?" — never "who was running it?" Same curiosity, different target. The first question produces information; the second produces defense. A manager who asks the first question consistently, especially on the bad weeks, is doing more for data quality than any audit could.

Compare a line to itself. Line 2 this month against line 2 last month is a fair fight — same machine, same products, mostly same conditions. Line 2 against line 5, or B shift against A shift, is mostly a comparison of circumstances wearing the costume of a comparison of effort. If you must look across lines, look for *transferable practice* ("their changeover is 15 minutes faster — what do they do?"), not ranking.

Hold people accountable for actions, not readings. There's a version of accountability that works: the feeder rebuild was assigned three weeks ago — done or not? The changeover checklist exists — is it being followed? Actions are chosen; numbers are inherited. Grade the choosing.

Let bad numbers be safe. The tell of a healthy program is what happens after a terrible week. If the meeting is a calm dig into the three losses that caused it, the program is a thermometer. If the meeting is an inquest, the floor will make sure there's never a terrible week again — on paper.

The payoff for restraint

Here's what plants that hold this discipline get in exchange: numbers they can believe. When logging is safe, operators record the embarrassing stops, flag the recurring jam even when they've been clearing it silently for a year. The data gets *worse-looking* and more true — OEE often drops several points in the months after a plant de-weaponizes its metrics, which is the sound of reality re-entering the records.

And real numbers compound. Every fix aimed with honest data actually hits something. The plants with the best OEE five years in are, almost without exception, the ones that were most relaxed about how bad the number looked in year one.

The thermometer never made anyone healthy. But nobody was ever cured by shooting it, either.