One Percent Downtime Is a Configuration, Not a Fact

The data return that didn't survive contact with the floor

A mid-market meat processor sent back the data pull for one of its plants ahead of a leadership presentation. Two numbers stood out. The labor rate came in at 26.02 dollars an hour, four to six dollars under the 30-plus figure leadership had been budgeting against. And the downtime report read 1 percent. Not one percent on a good week; one percent as a standing state, with an issues column that repeated the same boilerplate text in every row.

The problem was that the same team had already spent 8 to 10 hours standing on that floor over two days. They had watched the trim line, the sealer, and the multivac stop for hours. Two data sets, same plant, same period, describing two different factories. One of them was wrong, and it was not the one with the stopwatch.

Where the hours go when the number says zero

Overall equipment effectiveness is availability times performance times quality. Downtime is the availability lever. When a plant reports near-zero unplanned downtime, one of two things is true, and neither of them is "the floor is perfect."

Either the target speed has been dialed down, so the machine hits attainment while actually running slow and the loss quietly leaves the availability bucket, or the stops are being coded as planned time (film change, break, changeover) or filtered out before the number reaches corporate. Both paths produce a clean report. Neither produces more product.

The tell in this case was a decision, not a data point. The plant had cancelled its improvement coaches on the downtime system, on the reasoning that there was "no opportunity." No floor has no opportunity. A downtime module that reports zero opportunity is not measuring a flawless plant; it is misconfigured. The 100 percent quality reading on the roll-stock lines was the same signal wearing a different hat. A number that cannot be true on a real floor is a configuration artifact, not a fact, and the moment you accept one of them you have started planning against fiction.

How to catch a phantom OEE before you decide on it

Four steps, in order, before that number reaches a slide.

First, reconcile against a stopwatch, not against another spreadsheet. Put one person at one machine for one shift with a clock. Every stop, its length, its reason. Then compare that to what the system logged for that machine on that day. The delta is your measurement error, and it is usually enormous.

Second, audit the taxonomy, not just the total. Pull the category breakdown. If unplanned downtime is zero or a rounding error, that category is either disabled or being dumped into planned time. Count how many distinct reason codes actually get used; a real floor uses many, a gamed one uses two.

Third, check the target speed against the nameplate. If performance attainment is high but throughput is low, the target was set to flatter the number. Reset it to the machine's rated speed and watch attainment collapse to the truth.

Fourth, get raw access, not the roll-up. In this engagement the unlock was direct read access to the downtime system, pulling raw logs instead of the filtered corporate summary. The size of the gap between the raw feed and the polished report is the size of the problem.

Do this before you sign off on a capacity or a capital decision, because that is where the fiction gets expensive. A plant reporting 85 percent OEE will tell you it needs a new line to grow. A plant actually running near 50 percent has that growth sitting inside its own four walls, hidden behind a downtime category pinned to zero. The whole point of validating the floor first is to prove the ghost capacity is real before anyone spends capital chasing capacity they already own.

The labor rate carried the same lesson in miniature. The staffing and cost models had been built on 30 to 32 dollars an hour; the real rate was 26.02. Every headcount and overtime calculation was off before anyone touched the downtime number. Bad inputs do not announce themselves. They just quietly move every decision one notch away from reality.

What a trustworthy downtime system reads like

Unplanned downtime is a live, non-zero category on every line, every shift. Target speeds equal rated speeds, so performance attainment measures the machine and not the setpoint. Reason codes are plural and get used. A floor walk with a stopwatch lands within about 10 percent of what the system logged for the same window. And the improvement coaches are still on the calendar, because a well-run plant assumes it always has opportunity and keeps looking for it.

Closing

The report said one percent. The floor said forty. The company was not lying to itself on purpose; it was planning against a number that measured its software's configuration and mistook it for a measure of its machines. The savings case and the loss on the floor were both real. Nobody had put the stopwatch and the spreadsheet in the same room.