The Clean Number Is the Red Flag

A downtime report that reads one percent

A second plant of a Midwest protein processor sent over its downtime report. It read one percent. Ninety-nine percent of available line hours, productive. On paper the plant was already excellent, nothing left to optimize, no reason to let an outside team in.

The number was boilerplate. When we finally pulled the raw labor data, a month of asking, the plant's labor rate came back at $26.02 an hour. Leadership had been quoting $30-plus into every opportunity conversation for years. The gap was not rounding. It was that nobody had sourced the number; they had inherited it. And the same plant that reported one percent downtime had 8 to 12 heads of removable labor sitting on the trim line and in manual loading, no capital required, easy to validate. The report said the floor was clean. The floor was not clean. The report was.

Data fails in a flattering direction

Bad plant data is not random noise. It fails in one direction: toward looking good. A downtime tracker configured wrong, or runs clocked in late, or events simply not logged, all push the same way. They under-report loss. So downtime trends toward a clean single digit, uptime trends toward ninety-nine, and OEE, which should physically top out in the low nineties, starts printing numbers above 100 percent.

That last one is the tell. On another engagement we built a digital twin of a further-processing operation and calibrated it against a full year of real production data. The model landed at 98 percent accuracy to the actual floor. In the process it surfaced seven weeks where the plant's own OEE had been recorded above 100 percent, a physical impossibility, a structural data error nobody had caught because nothing in the building was checking the number against reality. The model was not valuable because it predicted the future. It was valuable because it was the first instrument on site calibrated to the present.

This is the trap. Leadership does not sit on a $300K opportunity for four years out of laziness. They sit on it because the instrument on the wall told them the floor was already running at ninety-nine percent, and no rational operator hunts for savings in a system that reports itself full. The flattering error does not just hide the opportunity. It manufactures the confidence that there is no opportunity to hide.

Validate three numbers before you trust anything

Before you model a line, build a business case, or approve a headcount, validate three numbers against physical reality. This is the work you can do this week.

First, the downtime distribution. Do not accept a summary percentage. Pull the top five downtime causes over the last three months. If the system cannot produce that list, or the list comes back suspiciously flat, you are not measuring the floor. Real plants carry a long tail of ugly, specific losses: a dicer that needs replacing, a changeover that runs long, a line that starts twenty minutes after the crew clocks in.

Second, the OEE ceiling. Any reading above 95 percent deserves suspicion; any reading above 100 is proof the inputs are wrong. Find out whether the denominator is scheduled time or a number someone picked.

Third, the labor rate provenance. Ask where the rate in the business case came from. If the answer is "that is what we have always used," pull the actual from last quarter's payroll and watch the case move. A four-dollar error on an hourly rate reprices every labor scenario you will ever run.

Only after those three survive contact with reality should you build the model or make the call. Calibrate to the present before you simulate the future. A model tuned to a flattering number will confidently recommend the wrong line, the wrong shift pattern, the wrong capital, and it will be exactly as accurate as the data you fed it and no more.

What the trustworthy version reads

The reference plant in that same company had run its floor-tracking system since 2017. Supervisors check real-time line-down status on a screen on the floor. Anyone can pull the top five downtime causes on demand when intuition says it is time to replace a machine. Quality checks are paperless and forced: you cannot log an X-ray check on a product that does not exist, so the record cannot drift away from the floor. Downtime there reads in the double digits, because the double digits are true. And the operator who runs it grades his own engagement a C-minus.

That is the standard, and notice how modest it is. Not a perfect plant. A plant where the numbers are merely honest, and the person closest to them still sees room. Honest data does not flatter you. It shows you the losses, names them, and refuses to round them away. A plant that grades itself a C-minus on a trustworthy instrument is in far better shape than a plant reporting one percent downtime on an instrument nobody calibrated.

Nobody sits on $300K of savings for four years on purpose. They sit on it because the report told them the floor was already clean, and a clean report is the one number on the floor you should trust the least.