Decide The Line’s Fate Before You Model It

A model nobody could act on yet

A regional lunchmeat and sausage processor had a problem worth real money and a question that made the money imaginary. The line-optimization model was good. A digital twin calibrated to a full year of line data hit 98.2 percent accuracy against facility actuals, ran Monte Carlo scenarios across four operating configurations, and put a number on the prize: $400K to $1.4M in labor efficiency, plus 12 to 18 percent off packaging spend. It even mapped the demand runway, from today's 5 to 6 million pounds a year up past 15 million, with a hard cook bottleneck waiting around 18 million.

Then the gate. Leadership had not decided whether to keep the sausage line in-house or hand it to a contract manufacturer. Until that call lands, every dollar in the model is conditional. Keep the line and the optimization work is the whole game. Send it to a coman and the engineering, the simulation, the scenarios, all of it is worth zero. The team had built a precise answer to a question leadership had not yet agreed to ask.

The decision sits upstream of the model

Here is the mechanism, and it is the one most capacity studies get backwards. The optimal configuration of a line is not a refinement of how it runs today. It flips entirely with the make-or-buy decision sitting above it.

If the line stays, the work is dial-in: scheduling, OEE recovery, selective automation on the front-end load and the back-end pack. If the line goes to a coman, the right move is the opposite, strip it, free the floor, and reallocate the cook and labor hours the line was consuming. Those are not two versions of the same plan with different savings attached. They are different plans. Modeling the first while leadership is privately leaning toward the second does not produce a head start. It produces work you throw away.

This is why decision order is not a nicety. A model inherits all the conditionality of the decisions above it. Optimize before the strategic call and you have not reduced uncertainty, you have spent precision on a branch that may get pruned. The number looks like progress. It reads like progress in a steering meeting. But it is a forecast for a world that may not exist by the time the make-or-buy memo is signed, and the payback windows here run 18 months from equipment arrival, with some equipment carrying a one-year lead time. The clock you are racing does not start until the decision lands.

What to do this week

Before you commission a line study, an automation business case, or a digital twin, run one test on the asset itself. Ask whether the line you are about to model could be outsourced, shut, consolidated, or rebuilt inside the next two quarters. If the honest answer is yes, and no one with authority has actually decided, stop. You are about to model a maybe.

Force the upstream decision first. At the lunchmeat processor, the right sequence was explicit: leadership settles the sausage-line question, then the optimization work proceeds against a real asset with a real future. The same pattern showed up at a multi-plant protein company rolling out a single packaging spec across its network. The technical work, labor and line optimization, was going to happen regardless of which plant went first. But the sequence, where to start and whether to convert specs before going to market or after, was a decision that had to be made deliberately, because starting in the wrong plant meant redoing the conversion and losing the learnings. The optimization was never the hard part. The order was.

Practically: write the make-or-buy or keep-or-kill decision down as a dated, owned item before any modeling budget is released. Name who signs it and by when. If the modeling has to start in parallel for timing reasons, scope it to the work that survives either branch, baseline data capture and OEE truth, the things you need no matter what, and explicitly hold the configuration-specific work until the gate clears. And share the risk honestly: if the asset's fate is genuinely undecided, say so in the proposal, because a precise model delivered against a line that gets outsourced is not a win you get to keep.

What good looks like

A well-run capital and optimization process has a visible decision sequence. The strategic call, keep or outsource, build or refit, comes first and is signed before engineering hours or capital are committed. The model then calibrates within 2 percent of facility actuals on every line, not just in aggregate; when one line lags the others, its data gets audited before its performance is judged, because a single line stuck at 88 percent fit is usually a data problem, not a worse line. Scenario ranges replace point estimates, so the business case survives a variable demand forecast. And every dollar in the model traces to a configuration leadership has actually chosen to pursue. You can tell the difference with one question: if someone asks what changes in this model if we outsource the line, the answer is either "we already decided we are not" or "we held that work." It is never "we would have to start over."

Closing

The expensive modeling error is not a wrong number. It is a precise answer to a question no one has committed to. Decide the line's fate first; then the model is worth what it cost.