The Average Is Not the Operating Condition

The award was one sentence on a Teams call

A refrigerated foods brand ran a corrugated and paperboard bid, picked winners, and told them so on a video call. The words were close to "good job, you got a majority of the brown box." No line level detail, no item list, no email record. Months later the awarded supplier came back with a number: it was seeing something like 15 percent of the volume it had expected, and it wanted to talk about price.

Nobody lied. The incumbents were still shipping some items during the transition, and at least one high volume item that had been awarded to the new supplier kept flowing through a different distributor entirely. The brand's sourcing lead was clear on his side of it too, and he was right: last year's spend is not next year's demand. Programs get cancelled. A category stops performing. Some of what showed up in the benchmark was safety stock, not consumption.

So there were two numbers in the room, both accurate, describing different things. The supplier priced one and is operating in the other. The gap between them is not a rounding error. It is the entire savings case.

What the supplier actually quoted

Work backward from the bid sheet and the mechanism is visible. The benchmark collection produced two columns at the item level: annual volume for the trailing twelve months, and the typical most frequent production quantity. That is a reasonable pair of numbers, and it is what suppliers quote against. Their cost model needs a run size and a yearly quantity to spread setup, plate, and freight across.

The problem is what happens next. Those two columns get handed forward as a forecast, because they are the only quantitative thing anyone has. A backward looking average, collected to describe last year, silently becomes the forward commitment a supplier prices against. Then the real order pattern arrives: lumpy, split across an award that was never written down at the line level, seeded with inventory nobody netted out.

The supplier discovers the variance first, because the supplier feels it in run size and idle press time. So the supplier prices it. That is the whole tax. The variability existed the entire time; the only question was who would measure it and who would end up paying for having ignored it.

The same mistake, priced in labor instead of resin

A protein processor asked for a labor validation at one plant, and the first draft of the scope proposed bringing pounds and throughput back on an aggregate basis. The plant manager pushed back inside two minutes, and his reason is the most useful sentence in this article: they do not feed the trim line with a mix. They run a steak day, or they run a barbacoa day.

An aggregate pounds-per-hour rate across those two states describes neither one. It is an average of two distinct operating conditions, and no shift ever runs it. Baseline on that number and every downstream conclusion inherits the blur: which crew size is right, which positions are genuinely removable, whether the conveyor investment pays.

Two more variability layers sat underneath it. The cut line is where the real labor sits, and cut content per primal had never been measured; some primals take zero cuts and some take close to ten. And the headcount visible during a visit is a sample of one. Fourteen people on the line the week you show up does not mean fourteen is normal, and it does not mean twenty is either. There were call-outs that week. There always are.

The correction was a data ask, not a cleverer observation: trailing twelve months of labor, segmented by run state, rather than the headcount standing there during the walk. Same fix as the packaging bid. Ask for the distribution, not the average, and ask for it at the level the floor actually operates.

Watch who claims the variance before it is measured

Here is the part that is about people rather than data. Before the study started, the plant manager named seven labor positions he was already working on, three on the front and four on the back, and offered to point them out on day one. That is a rational move. When the baseline is aggregate, savings credit is unassignable, so it goes to whoever stakes a claim first and loudest.

The supplier in the packaging story did the same thing from the other direction. Both are responses to the same vacuum. An aggregate basis does not just lose precision, it removes the ability to attribute anything, and attribution is what the negotiation is actually about.

What to do this week

Pull one recent award and check whether it exists at the line level in writing, with item numbers and quantities the supplier could recite back to you. If it lives in a meeting, it is not an award; it is a shared impression.

Then take the volume columns you send out to bid and label them honestly. If a number is trailing twelve month consumption, say so on the sheet, and net out safety stock and any inventory build. If you also have a forward forecast, send both, and mark which one the supplier is expected to price. Suppliers price risk they cannot see, and they price it high.

On the labor side, list the operating states your lines actually run: dedicated days, campaign runs, allergen segregated blocks, seasonal packs. Ask for twelve months of labor split those ways before anyone stands on the floor with a clipboard. If the split does not exist in the system, that absence is itself the finding, and it is usually a bigger one than the headcount question you came to answer.

What the well-run version reads

The bid file states annual volume and typical run quantity per item, dated, with the source period named and inventory netted out. The award exists as a written line item list both sides can open. Labor baselines are stated per run state, on trailing twelve months, not on the week the consultants visited. Cut or touch content is measured per product family rather than assumed uniform. And when a supplier's actual volume drifts from the quoted basis, somebody sees it inside a month, from an order report, not from a price increase letter nine months later.

Closing

Both stories are the same story. Someone summarized a variable process into one number so it could be quoted, awarded, or baselined, and the variance did not disappear when the number did. It just waited, and it charged interest.