The Number Said One Percent, The Floor Said Sixty

The number said one percent. The floor said sixty.

A multi-plant cooked-protein manufacturer wanted help sizing a labor opportunity at one of its plants. The reported numbers looked tight. Downtime on the lines was running around one percent. On paper, that plant was nearly flawless, and a one-percent-downtime line leaves almost nothing to optimize.

Then the team went on-site for two days and watched the lines actually run. The trim line and the sealer were down 60 to 79 percent of the time. Not one percent. Sixty to seventy-nine. The gap between the reported number and the observed number was not a rounding error or a bad week. It was the difference between a plant that has no opportunity and a plant sitting on an 8 to 12 head labor reduction with no capital required, low-hanging fruit that the reporting system had made invisible.

This is the failure mode that quietly wastes more optimization effort than any other. The instrument that measures the floor is wrong, and every decision downstream of it inherits the error. You are not optimizing the plant. You are optimizing a story about the plant.

A miscalibrated instrument does not just miss, it lies in a specific direction

The reflex when reported and observed numbers diverge this far is to suspect manipulation. That is almost always the wrong read. The account lead's assessment was blunt and correct: the plant is not sophisticated enough to be doctoring or filtering the data. The far more likely causes are mundane. Runs start late and the clock never catches the gap. A machine is not being tracked at all. The thresholds and settings in the system are wrong, so genuine stops never register as downtime.

None of that is fraud. It is an uncalibrated instrument, and an uncalibrated instrument does not fail randomly. It fails in a direction. Unlogged downtime always reads as uptime, so the error only ever runs one way: the floor looks better than it is, never worse. That is what makes it dangerous. A number that is randomly noisy invites suspicion. A number that is consistently flattering earns trust it has not paid for, and the plant manager who reports one percent downtime is not lying. He is repeating what his screen tells him.

Once you accept that the instrument lies in a flattering direction, the second-order damage is obvious. A constraint analysis built on this data will point at the wrong machine, because the real bottleneck is the one whose stops are not being recorded. A capacity decision built on it will under-size the opportunity, because the slack the system hides is exactly the slack you would have gone after. And a savings case built on it cannot survive contact with the floor, because the first person who walks the line for a shift will see the sixty percent the spreadsheet swore was one.

Calibrate the instrument before you touch the crews

The order of operations is the whole lesson. You do not reallocate labor against a number you have not validated. You validate the number first, and only then do you turn the labor lever.

Start with a reconciliation walk. Put someone on the line for a full shift with a clipboard and time the stops by hand, then lay that against what the system logged for the same window. You are not auditing to assign blame. You are calibrating the instrument. If the hand count and the system agree within a few points, trust the data and move to optimization. If they diverge the way this plant did, the data is your first project and the labor study waits.

Then fix the capture, not the report. The temptation is to correct the downtime number on a slide and move on. That patches the symptom and leaves the instrument broken for the next analysis. The durable fix is upstream: get runs clocked at their real start, get every machine tracked, get the downtime thresholds set so a real stop registers as a real stop. When the capture is right, the report is right for free, and it stays right after the consultants leave.

And use the sister plant as the proof that the bar is reachable in-house. This same company runs the same monitoring system at another plant, and that plant has run it well since 2017. A supervisor there checks an iPad for live line-down status and per-line progress against the day's order. The team pulls the top five downtime causes over three months when intuition says it is time to replace a piece of equipment. Quality checks, hourly date codes, and supervisor standard work are paperless and auditable. That is not a vendor demo. It is the same firm, two plants, one running the instrument honestly and one not. The bar is not theoretical, and it is not a capital project. It is a configuration and a discipline the company already owns down the hall.

The labor opportunity was always real. The 8 to 12 heads did not appear because consultants walked the floor; they were there the whole time, hidden by a downtime log that read one percent. The work was never to find capacity. It was to fix the instrument that was hiding it.

What a well-instrumented line looks like

Reported OEE reconciles to a walk-the-floor observation within a few points, every time someone checks. Downtime is captured at the event, logged when the line stops, not reconstructed from memory at the end of a shift. Run start times in the system match the scheduled start times within minutes. The top five loss causes for any line can be pulled in minutes, not assembled over a week. A supervisor can show real-time line-down status on a screen without caveats about which machines are not wired in. When those conditions hold, the number on the screen and the reality on the floor are the same number, and a labor decision built on it will still be standing after the first person walks the line.

The plant that reported one percent was not the best-run plant in the company. It was the worst-measured one. The two are easy to confuse, and confusing them is how a real opportunity stays buried under a flattering number.