Ghost Capacity Hides in the Seams Between Systems

When a packaging line books 140 percent OEE on a system the operations team trusts, two things are true. The line is not running at 140 percent. And whatever you do next based on that number is going to make the plant worse.

I sat in a working session three weeks ago looking at a Red Zone dashboard at a Tier-1 protein processor. The reported OEE was 140. The team in the plant believed actuals were 50 to 60. The supervisor running that line had been recoding events for months. Stops became lunch. Multi-step changeovers got fragmented because the system retriggers a new downtime code every time a machine stops and starts. By the time the data hit the schedule meeting, the line looked exemplary.

Ghost capacity. Throughput that was already there, hidden inside a metric the schedule never modeled correctly.

Most plants think of capacity as a property of equipment. Line N runs 25 pounds per minute. Lines O, P, Q run closer to 19. Two shifts, one shift, weekend cover. People reach for capex when those numbers come up short. New line. New tooling. New automation cell at the end of pack-off. The capex case gets built around the equipment number that lost.

The line is rarely the problem. The equipment number is rarely the problem.

Throughput emerges from the interaction of equipment, data, scheduling, and pacing. The ceiling on that interaction is almost always lower than the ceiling on any single piece of it. And every layer obscures the loss in a different way.

Three places ghost capacity hides

The data layer obscures it. A sausage facility I assessed last month runs four packaging lines: N, O, P, Q. Reported Red Zone changeover averages came in at 6.1 minutes on Q, 7.9 on O, 10.7 on P, 13.4 on the lunch meat line. The operations team described those numbers as physically implausible. They were. Operators were splitting a single 30 minute changeover into three Red Zone events because a knife adjustment showed up as a separate downtime code from the wash. Two of the events also got recoded as lunch coverage because that is what the supervisor knew how to log fast. The schedule averaged the fragments. The plan baked in changeover assumptions that were 40 to 50 percent of actual. By the time the supervisor sequenced the next week, ghost capacity had already been spent on optimistic dwell.

The interaction layer obscures it. A protein plant I worked with this spring is preparing for a major QSR customer's pre-marinated protein transition. The first market trial lands in April. The prior format transition the same plant ran a few years earlier produced a 12 to 15 percent line run-rate reduction. That number is going to land again, on a fixed footprint, with seasonal demand holding constant. Equipment did not change. The packaging profile did. Tumbling and marinade got added. The interaction between forming, packaging, and pack-off shifted by inches and the line lost ground by yards. The capex committee had already approved spend. The line layout had not been finalized. That is not a capacity problem. That is a system-interaction problem masquerading as one.

The pacing layer obscures it. At a deli operation last quarter we watched trimmers accumulate three or four pieces in front of them and stop the conveyor every minute or so. Worker utilization policy said 85 percent. Observed utilization on the floor was lower because the conveyor itself was the queue. The model said 26 trimmers were needed for 78,000 pounds a day. The model was wrong. Sixteen trimmers paced correctly, with the conveyor metering one piece every 42 seconds, would hit the same throughput. The 10 people the line did not need were pacing-induced ghost headcount, not capacity headcount. Adding them creates their own micro-stops.

Find ghost capacity before you sign the capex

Three specific moves you can run this month. Each one costs less than $5,000 to execute. None of them require a new line.

Audit the changeover codes against video for one week. Pick the line with the most variable downtime numbers. Pull every changeover your event-tracking system logged that week. Pull the matching DVR. Match start to end without believing the system. The first audit at the sausage facility I described earlier found that a 6.1 minute changeover averaged 18 to 22 minutes in reality, with 30 to 40 minute washdowns counted as three separate stops. That is a 12 minute per event swing on 31 events a month. Six and a half hours of monthly capacity nobody knew was missing because the report said it had not been spent.

Run a discrete-event simulation against 30 days of cleaned data before you approve capex. The protein facility I worked with built an OptQuest model on 30 days. It projected $1.3 to $1.4 million in annual labor savings and a 14 percent capacity increase by consolidating four lines to two and running two shifts. That recommendation reverses the direction every prior planning conversation had been moving. Twelve months of data would stress-test the answer further. Thirty days were enough to flip the bias of the conversation. Do not approve a new line until the model says the existing footprint is genuinely full.

Pace the conveyor explicitly, then count. Time the center belt to deliver one piece every 42 seconds, or whatever your cycle math demands. Do not let operators dictate pace by piling. Then count the headcount needed to hold rate. Most lines find they were carrying buffer headcount, not throughput headcount. Reallocation, not layoff. The number drops on its own once the pacing logic owns the metering, not the trimmers.

The savings are real on paper. So is the ghost.

I have watched plants approve a multi-million-dollar packaging line on the strength of an OEE number the floor knew was overstated by 80 points. I have watched another plant defer a six-figure conveyor pacing investment that would have unlocked seven figures in annual labor.

The line was not underperforming. The schedule was misreading the line. Until somebody puts the data layer, the interaction layer, and the pacing layer in the same room and lets them argue, the ghost stays where it is, and the next capex case gets pointed at the wrong thing.