The line is staffed for the worst piece you'll never run

The 18-trimmer line that needed 10

A Midwest meat processor runs 18 trimmers on a single shift. The reported baseline is 65,000 pounds per shift, which the math supports: 18 people, three minutes a piece, somewhere between 2,400 and 2,800 pieces. The plant manager will tell you the line is staffed correctly. He has been running it for twenty years.

We modeled it. Touch time per piece, including the muscle work and the trim variability, comes in around 100 seconds. At that pace the line needs 10 to 13 trimmers, not 18. Five to eight people of standing slack, every shift. That is roughly $250 to $300 thousand a year of labor on one shift, and closer to $575 thousand once the back shift is included.

Nobody put eight extra people on the line by accident. They put them there because the line had no pace-setter. The conveyor speed floated. So the trimmers self-paced. When humans self-pace a process with piece-by-piece variability, they pace to the worst piece they remember seeing this month. Not the average. Not the median. The worst.

Why this happens

In CPG manufacturing, formulation drives variability inside the same SKU. The "same" trim cut has different fat content, different muscle structure, different bone fragment risk. The "same" frozen meal has different ingredient drop times. The "same" baked good has different dough viscosity by batch.

The scheduler does not see any of this. The scheduler sees one OEE number per line, one average changeover, one staffing model. The system underneath looks like it is running on averages because that is what the spreadsheet reports.

But OEE rolls up three components, and the formulation interaction lives inside the performance component. At the same processor, Line 1 hit 39.8 percent OEE across 14 days of sensor data. Lines 2 and 3 came in at 51 to 52 percent. All three lines can run 20 parts per minute on the design spec. All three average 5 to 6. The gap is availability losses at startup, 30 to 60 minutes each shift, plus performance losses that show up as operator slowdowns on the harder pieces. Daily overtime ran 1.5 hours or more to hit volume.

The result is a plant manager who has run that line for two decades and genuinely believes he is near world-class. Until you put sensors on every line for two weeks, calculate proxy OEE against the best day plus 10 percent as ideal, and watch the actuals settle in the high 30s.

What to do this week

Three moves, in order.

First, instrument the line for two weeks. Banner sensors, wireless trackers, whatever you have on hand. You are not looking for an average. You are looking for the variance, how cycle time moves across the formulation mix, across the shift, across the changeovers. Two weeks captures enough of the actual mix to model.

Second, set the conveyor speed as the pace-setter. Pick a target throughput, dial the line, and stop letting the workers absorb variance through self-pacing. The pace-setter forces the system to expose its real constraints. If the conveyor is at the right rate and operators cannot keep up, you do not need more people, you need a buffer or an accumulator. If they are idle, you do not need more line speed, you need to staff down.

Third, run the simulation against the real formulation sequence. Not against averages. Against the actual order of SKUs that hit the line last month. At a national sliced-meat brand we modeled, the four-line, one-shift configuration ran roughly 5 to 6 million pounds annually. Simulating the same plant as two lines on two shifts, with one auto-loader per line and the conveyor set as the pace-setter, the model showed 14 percent more throughput and roughly $1.4 million in annual labor savings. The kicker: the same building could triple to 18 million pounds before the cook step became the binding constraint. Three times the latent capacity was sitting inside the existing plant. Nobody could see it because the scheduler was treating four lines as four parallel averages instead of one interacting system.

The lens-flip

The reflex when a plant misses rate is to add labor or buy a new line. Both are responses to a problem the scheduler invented by averaging the formulation mix. The capex case usually disappears once you answer the simpler question first: is the line already on the floor staffed for the formulation actually running, or for the worst one the operators remember from a bad shift six weeks ago.