Level the Arrivals Before You Price the Line

The request arrives as a machine

A capacity request usually arrives already shaped as equipment. Someone names the station that visibly backs up, collects a quote, and builds the return on the quoted rate. The decision in front of you is not whether the quote is competitive. It is whether that station is slow or whether it is fed unevenly.

Those two conditions look the same from the aisle. Work piles up in front of the station, operators never stand still, and the shift ends behind plan. One of them is fixed by steel. The other is not, and the purchase that follows will not remove it.

Convergence is an interaction, not a speed

A converging station takes work from more than one source: several cutting or trim stations feeding one packaging line, several product families feeding one scale, two cells feeding a single sealer. Its output depends on whether work arrives at a rate it can absorb, not only on how fast it runs when loaded.

When arrivals are uneven, the station alternates between two losses. It is starved while it waits for the next source to release work, and it is blocked while the downstream buffer clears a surge it just passed on. Both states look like the machine is the problem. Neither is repaired by raising its rate.

Time lost to starvation is not recoverable. A station cannot bank the minutes it spent waiting and spend them later at double speed, so an hour of uneven arrivals costs the same as an hour of downtime in the schedule, while showing up nowhere in a downtime report.

The arithmetic that hides this is simple and common. A team sums the rated rates of the feeding operations, compares that sum to the rated rate of the converging station, and concludes the station is short. That comparison answers a question about averages. The station does not live in the average. It lives in the distribution, and the distribution is set by upstream decisions: how long each product family occupies a feeding station, whether changeovers upstream are staggered or simultaneous, when breaks are taken, and how large a batch is released at one time.

Two feeding operations that each average the same output can produce very different arrival patterns. One releases steadily. One holds work and releases it in bursts. The converging station reads the second as a capacity shortfall, because from where it sits, that is exactly what it is.

Sequence the study before you price the ceiling

My rule is to model the debottleneck and the leveled case at the current configuration before modeling a higher ceiling. This is not an argument against capital. It is that the leveled model is the input the capital case needs, so building it first costs nothing in sequence and changes what the request has to prove.

Start by defining the convergence point and its boundary. Name the station, the sources that feed it, the buffer in front of it and behind it, and the shift window you will observe. A station with three feeding sources and a shared downstream conveyor is a different system than the machine on the quote.

Then record arrivals, not only output. Most plants can report what left a station. Far fewer can report when work reached it, by source and product family. That arrival record is what separates a slow machine from a starved one, and it is usually the piece that has to be built rather than pulled.

Run the leveled scenario on the equipment you already own. Hold the equipment fixed and change only what governs arrivals: the sequence of products through the feeding operations, the size of upstream releases, the staggering of changeovers and breaks, and the rules for when a feeding station starts the next unit. Then compare the delivered rate under that pattern to demand.

The result splits the decision cleanly. If the leveled configuration meets demand, the capital request is no longer a constraint case. It may still be a good decision, for service, growth, redundancy or labor, but it has to be argued on those terms and with its own evidence. If the leveled configuration still falls short, the shortfall is now measured against a real operating pattern instead of a nameplate, and the capital case can state what the new equipment adds at the same demand mix.

Count the cost of leveling as part of the comparison. Smoothing arrivals often means more changeovers upstream, smaller releases, or an extra pair of hands at a feeding station. Those are real costs and they belong next to the avoided capital, not hidden inside the cheaper option.

Name the owner of the arrival pattern before the study closes. The station that fails belongs to operations or engineering. The arrival pattern usually belongs to whoever sets the schedule and the release rules, and that person is often not in the room where the capacity request is being argued. A leveling case with no owner reverts to the old pattern within a few weeks, and the plant concludes that leveling does not work when what it demonstrated is that nobody was accountable for it.

What a well-run convergence point reads

Running, starved and blocked minutes are recorded for a full run on each major product family, not sampled during a walkthrough. Arrival times at the station exist by source, and someone can pull them without building a new report from scratch. The schedule states the sequence intended to feed the station, and deviations from it are explained by a named operating cause on the day they happen. Every rate in a capital request traces to a timed run record rather than a nameplate, and the request shows the delivered rate before and after the purchase at the same demand mix. When the pattern breaks, the person who owns the release rules knows before the shift report does.

The next test

Pick the station your team complains about most. Take one full run and three timers: running, starved, blocked. If starved plus blocked exceeds running, the next dollar belongs to sequencing, not to steel, and you now have the record to prove it either way.