The Variability Tax Your Capex Case Cannot See

The question that gives the game away

A 50-year meat-industry veteran asked me the obvious question last week. I had just walked him through how we model a processing plant, build a digital twin from CAD and production data, drop temporary wireless trackers on the lines, and simulate labor and changeover scheduling before anyone moves a machine. He nodded along, then asked the thing almost everyone asks first: "So you do your best work on long, dedicated runs, right?"

It is the natural instinct. Clean, predictable, one product all shift. That feels like where a model earns its keep. The truth is the opposite. The longer and more dedicated the run, the less there is to find. The value lives in the mess. The more a plant's schedule varies, the bigger the bill it is paying without seeing it, and the bigger the number a simulation pulls back out.

That bill has a name. It is the variability tax, and most plants pay it as a line of credit against capacity they think they are short.

Why the spreadsheet cannot see it

Open the capacity model on almost any food plant and you will find the same thing: one average speed per format, one average changeover, one average crew rate. The line gets modeled as if it runs the way it runs on its best day, all day. Reality does not average.

Every changeover needs a labeler reset, a coder reset, a film thread, a sanitation step. Each one looks like five to ten minutes, small enough to round away. Run the formats in a sensible order and they stay small. Run them in the order the order book happened to arrive and they cluster, and the clustering is where the shift disappears.

Watch it on a plant running more than two hundred SKUs. The top forty drive almost all the volume. The long tail drives almost all the schedule. Those minor runs each carry their own changeover, film, and reset, and they consume planning effort far out of proportion to revenue. Measure it and the labor minutes per thousand units on the tail run close to double the core lines. The plant is not slow. It is taxed.

Then there is the interaction, which is the part that fools good operators. A snack line had let its format mix drift from three pack sizes to seven. The wrappers started missing rate. The capex case said add a second cell. We modeled the line first. The extruder ran at design. The cooling tunnel kept up. The wrappers fell behind on every shift the schedule pushed past four format changes. They were not slow. They were drowning. Upstream was producing roughly thirty percent more work in process per shift than packaging could absorb at the new mix, so operators paused upstream to let packaging catch up, and every pause cost recovery on the cooker. No single machine was the constraint. The interaction between them was. System interaction governs throughput, and you cannot find a system effect by staring at one machine's spec sheet.

Price the variability before you pour concrete

The reason this matters is that the variability tax masquerades as a capacity shortage, and a capacity shortage gets answered with capital. A labeling deadline, a missed rate, a stretched schedule, and the request lands as a second line or a third shift. The math looks clean because the math is built on averages, and averages hide the tax.

So before approving the capex, price the variability. That is a modeling exercise, not a construction one, and it runs in three moves you can start this week.

First, pull labor minutes per thousand units by SKU, sort the list, and look at the tail. If the bottom half runs near double the top, you have found the tax and roughly sized it.

Second, map changeover minutes by transition type, not just total changeover. The number that matters is not how many changeovers you run; it is which sequences cluster the expensive ones. The order is a free variable most schedulers never treat as one.

Third, model the current schedule and a re-sequenced one side by side before you sign for capacity. We did exactly this on a plant the math said was fifteen percent short of a labeling deadline. The model said they were not short at all. Re-sequence the book, batch the like formats, and the shortage closed without a dollar of capital. The decision was not invest. It was defer, and fix the sequence first.

Most "we are short" cases are defer cases wearing an invest costume. Not all of them. Some plants really are out of runway and need the line. The point is that you cannot tell which is which from a model that runs every SKU at its best-day speed, and that is the only model most plants have.

The two numbers nobody puts in the same model

The capex case said the plant was fifteen percent short. The model said it was not short at all. Both were honest. They were just measuring different things, and nobody had put the variability and the line speed in the same model. The savings from a second line were real on paper. So was the tax the plant kept paying on every clustered changeover. It is almost always cheaper to re-sequence a schedule than to pour concrete against a shortage you do not actually have.