Price The Machine After You Prove The Baseline

The plant said there was no opportunity, for a year

A Midwest protein further-processing plant runs a sous vide operation feeding two product families off one trim line. Plant leadership spent roughly a year insisting the line was already as lean as it could be. Wrong plant. Nobody to take out. You are wasting your time. That position was escalated more than once, and it was well documented, which turned out to matter later.

Two facts sat in plain view the entire time. The line monitoring system was reporting overall equipment effectiveness around 180 percent, which is not a performance result, it is a broken denominator. And the two aging packaging lines the trim operation feeds run one wide at about eight packages a minute, while the same equipment at sister facilities runs two wide. Upstream of those two machines stood 23 trimmers, plus a set of loaders working a conveyor that runs faster than anything downstream of it can consume.

The question that opened the work was blunt: you are putting out eight a minute here, so why do you have a hundred people cutting to feed it? That is not a labor question and it is not an equipment question. It is a question about which of those two you are allowed to answer first.

Capital confidence is a sequencing problem, not a modeling problem

Every capital request is a subtraction. Current state minus future state, divided by the check. Operators spend their review time interrogating the future state: the vendor's rate claim, the uptime assumption, the ramp. Almost nobody interrogates the current state, because it arrives as a headcount, and a headcount reads like a fact.

It is not a fact. It is the number of people currently standing there, which is a different thing entirely. And the moment you touch the line, that number moves, which means the baseline you quoted against expires the week you start working.

The discipline that fixes this is a sort, done before any integrator sees a layout. Every claimed reduction goes into one of three buckets. Bucket one needs nothing built: the handler stationed after the X-ray comes off the line the day someone in authority says so. Bucket two is already bought: a lift table was purchased and approved, and when it lands, trim loading goes from four people to two. That is not a test, it is an installation date. Bucket three is the only real capital question: the orientator position that needs a mechanical solution, the manual check-weigh position that needs a check-weigher.

Skip the sort, and here is what happens. An equipment integrator gets walked through the facility, gets the layout files, and builds a rough order of magnitude estimate for backroom automation, case erection, bag-in-box loading, labeling, single-operator palletizing, plus the trim lines and a check-weigher. All of that is legitimate scope. But if the integrator prices against the pre-sort baseline, the resulting business case earns its return partly on bucket-one and bucket-two labor. The number is not fabricated. It is misattributed. And misattributed savings are precisely the ones that never appear in the P&L, because the free reductions happen anyway, the machine gets installed, and the delta the model promised is already spent.

The mirror-image failure is just as expensive: freezing every removal until the measurement is perfect, which is how a plant goes a year with nobody taken out while a study matures.

Run the removals before you run the quotes

The sequence that produces a defensible capital number looks like this on a real schedule.

Start with the stated baseline, not your own count. The plant says 23 trimmers, so 23 is the number of record. Day one goes to 20, which is three heads out. The following week, 19. The week after, 18. Then stop and read all three weeks together, rather than declaring victory or defeat on week one, because a line absorbs a three-head cut differently than it absorbs the fifth. Pace the conveyor to the machines it feeds instead of to the people loading it, and test the alternating break pattern in the same window.

Measure in pounds per employee hour against the actual product mix, not on a single observation day. A steak day and a barbacoa day produce different labor arithmetic on the same equipment, and a baseline built from whichever day you happened to visit will be argued away in the first review meeting. Segment it, or you will relitigate it.

Then document what is already captured before it disappears. This is the rule worth stealing: write down every reduction that has already been agreed, with the date and the signature, before the heads actually come off the line. When a plant that spent a year saying there was no opportunity watches five people walk off the floor, the reduction gets reclassified as something they always planned to do. In this case, the people who authorized the work are personally compensated on savings being credited, which helps, but the only durable defense is written scope, widely copied, dated in advance.

One more constraint deserves its own line. This site is about to absorb a funded 33 million dollar expansion and a set of October volume moves. A baseline with an expiration date is still a baseline, but it obligates you to measure and remove before the reset, because after it your before-state no longer exists and every number you carry forward becomes an argument instead of a record.

What a well-run capital gate looks like

Every headcount reduction in the request is tagged free, bought, or to be bought before the document goes out, and the free ones already have execution dates on them. Equipment is quoted against the reduced line, never the original. Baseline is expressed as output per employee hour across the real mix, and it holds three consecutive weeks at the new level before anyone calls it proven. Any monitoring number above 100 percent stops the gate, because a denominator error upstream of a capital decision is a capital decision made blind. Every removed position has a name, a date, and a person who agreed to it in writing.

Closing

The machine was never the risky part of this decision. The risky part was the number in the current-state column, which nobody had counted, which moved the moment anyone touched the line, and which the plant would have revised the instant the savings became someone else's credit.