The Constraint Map Comes Before the Fix

The task that had to wait

Last week, on a meat processor's project plan, a task sat blocked: "Map current-state process flow, station by station." The predecessor holding it up was not a budget approval or a vendor deliverable. It was "Re-baseline OEE on clean data."

The team refused to draw the map until the measurement underneath it was trustworthy. That ordering looks slow. It is the fastest path in the building. A flow map built on inflated OEE inherits every distortion in the source data: a station that reports 85% but runs 60% shows up as spare capacity, and the map routes your attention, your labor, and eventually your capital toward the wrong seam. You do not get that spend back. Clean the baseline first, then draw.

A map is a hypothesis about interaction

The naive version of a constraint map is a layout drawing with a red circle on the slowest machine. That version is nearly useless, because throughput is not governed by any single station's rate. It is governed by how stations interact: how changeovers on one line steal shared labor from another, how a quality hold upstream turns into starvation two stations down, how a buffer that looks like safety stock is actually the only thing hiding a 40-minute mismatch in cycle times. Under demand variance the governing constraint moves with mix, which is exactly why a circle drawn once a year is fiction by March.

A real constraint map is a hypothesis document. It names the station you believe governs output, states what you should observe if that is true (upstream buffers grow, downstream starves, overtime clusters there), and gets checked against the floor.

That checkability is the point, and it is also the politics. At the same processor, leadership finally walked one of their plants with the outside measurement in hand. For months, what the plant had been reporting up the chain and what the measured data said were two different operations. Standing on the floor, the gap stopped being an argument between people and became a difference between two documents, one of which was checkable. Leadership left willing to push harder, and the conversation moved from whose version to believe to which station to fix.

Build it before you walk the floor

A global protein processor recently asked what preparation a first working session needed. The answer was one file: send the CAD layout ahead of time. With the layout in hand, you can arrive carrying an initial zone map and a set of constraint hypotheses, so the first hour on site tests specific claims about their operation instead of trading generalities about capacity.

That sequence works at any scale, and you can run it this week:

First, get the physical layout and divide it into zones by flow, not by department. Write down, before looking at any performance data, where you believe the constraint sits in each zone and what evidence would confirm it.

Second, re-baseline the measurement at the suspected governing stations. Pull the raw event data, not the monthly rollup. If reported OEE and observed OEE disagree by more than a few points, the map waits until you know which one is real.

Third, walk the flow station by station and mark two things only: where material accumulates and where operators wait. The constraint lives at the seam between an accumulation and a starvation. Everything else is commentary.

Then, and only then, make the labor call. A constraint map turns "we need more people" into a placement question: reallocate first, because moving an operator to the governing station is free throughput, and hire only when the constraint's own math says the station is fully crewed and still short. The processor above ordered its plan the same way: map the current state, then decide where the next hour of labor goes.

What a well-run floor looks like

There is one current-state map, and it lives where scheduling decisions are made, not in a consultant's deck. The constraint is named, and the end-of-shift state confirms it: buffers grow upstream of the named station, downstream stations show starvation minutes, and overtime concentrates there. OEE at the constraint is measured from raw events on data someone has re-baselined inside the last quarter. When the demand mix shifts, the map is redrawn inside a week. Labor requests cite the map, and reallocation is the default answer; a hire request that cannot point to a station on the map does not clear.

The argument is the symptom

Plants that argue about where the constraint is do not have a data problem. They have an unwritten map. Write it down, on clean numbers, and the argument becomes a measurement.