What Repeated Machine Swaps Tell You About Your Capacity

Every time somebody moves work off one machine and onto another, they have made a capacity judgment under pressure and left a record of it. Counted over a quarter and sorted by which station work escapes and where it lands, those records are the cheapest capacity evidence in the building. EDGEBIC by User Solutions logs each swap with the old machine, the new machine, who did it, and the reason, which is what turns anecdote into a frequency table.

This post covers one return: capacity evidence you already generated and never read. It sits under the EDGEBIC results guide and complements how capacity evidence answers a new machine request.

A Swap Is a Decision, Not an Accident

In most plants a machine swap is invisible as a category. It happens, it works, everyone moves on. Nobody would call it a data point.

But look at what a swap actually is. A person examined a station's queue, concluded the work would not get done there in time, identified another machine that could take it, and moved it. That is a capacity assessment, made by somebody with current information, about a specific station, on a specific day.

Do that four hundred times a year and you have four hundred capacity assessments. They are not on any report anyone reads, they were never aggregated, and they are more current than any monthly average because each one was made at the moment the pressure was real.

The insight is not that swapping is bad. Swapping is usually the right call. The insight is that the pattern of swaps is information, and it is free.

The Mechanism: Swaps Are Logged With Direction and Reason

The Resource Replacement Audit records every work-center swap as an old-machine to new-machine pair, with the actor who made it and the reason if one was entered. It is the companion to the reschedule trail: where that report answers when work moved, this one answers where it moved.

The step-level Work Center Progress report carries the same fact from a different angle, showing the replacement work center on any step that was swapped, alongside net hours, actuals, and percent complete.

So there are two things you can count, and they answer different questions.

That third row is the one most worth pausing on. A from-and-to pair that repeats is an unwritten routing. Somebody has decided, many times, that this operation may legally run on that machine. Your routings do not say so. Your promise dates were computed as though it were not true. And when the person who knows it is on vacation, the substitution stops happening.

Two Kinds of Cost Hiding in the Habit

Repeated swapping is not free, and the costs are of two different types.

The hours cost. A true alternative on a step carries its own setup time and its own required hours rather than a scaled copy of the primary's, so an older or slower substitute genuinely consumes more hours for the same work. Within a work center group, a member's factor multiplies the step's base run hours, where 1.0 is the baseline and a higher number is a slower machine. Either way, if the rescue machine is slower than the primary, every swap spent hours nobody counted, and those hours came out of somebody else's capacity.

The honesty cost. The plan booked the primary. The work ran elsewhere. That means the completion date you quoted was computed against a machine that did not do the job, and the plan the floor followed was not the plan the office believed in. Any alignment between sales and production promises is fictional to the extent that the real routing lives in people's heads.

Both costs are recoverable by the same move, which is to stop making the decision manually.

The Causal Chain to a Decision You Can Actually Make

Here is what the count changes, step by step.

1. You aggregate a quarter of swaps by from-and-to pair, with counts.
2. One or two origin stations dominate. That is your constraint, identified by behavior rather than by a percentage crossing a threshold.
3. One or two destination machines dominate. That is your undocumented safety valve, and its importance to your plant is much higher than your routings imply.
4. You choose the response per pair, not per plant. This is where the money is, and there are only a few options.

For a pair that repeats often and where the machines are genuinely interchangeable for that class of work, pool them as a work center group so the scheduler shops the pool automatically. For a pair specific to one operation, configure it as a true alternative on that step, with the alternate's own setup and required hours entered honestly, so alternates keep a hot job moving without anybody being paged.

For a pair that repeats at high volume with a slower destination, the arithmetic has stopped being a routing question and become a capacity question. You are paying extra hours regularly to escape a station. Price the annual hours against the alternatives.

And for a pair that appears twice in a quarter, do nothing. It is genuinely an exception, and configuring exceptions creates noise.

5. Re-run Drive Schedule so the change takes effect. Configuration edits do not move an existing schedule by themselves, and steps that already have actuals stay locked to the machine that ran them.

Measuring It in Your Own Plant

One hour, one quarter of history.

1. Export the swap audit for the last full quarter. Actor, reason, old machine, new machine.
2. Build a frequency table by from-and-to pair. Count only, no interpretation yet.
3. Rank origins by outbound swaps. The top one or two are candidates for your real constraint. Compare against whichever station you have actually flagged as the bottleneck, and note any disagreement, because that disagreement is worth a meeting on its own.
4. Rank destinations by inbound swaps. Ask what happens to your delivery if the top destination is down for a week. If the honest answer is bad, that machine is load-bearing and is not modeled as such.
5. Price the hours delta on the top three pairs. For the step usually involved, take the difference between running it on the primary and on the alternate, using the alternate's own hours or its member factor, and multiply by the pair's quarterly count. Annualize. That number is the cost of the habit.
6. Check the reason coverage. What share of swaps carry a reason? A low share tells you your next audit will be less useful than this one, which is an argument for entering reasons rather than a defect in the report.

For documented outcomes rather than typical ones, the User Solutions and RMDB lineage includes GE Railcar moving from 30 percent to 90 percent on-time delivery, and Cummins deploying this scheduling approach across 33 locations. Those belong to their engagements and are quoted as heritage, not as forecasts.

Where This Evidence Runs Out

Five honest limits.

Equal machines make the hours cost zero. If the destination runs the step at the same rate with the same setup, the swap cost nothing in hours. The honesty cost remains, because the plan still booked the wrong machine, but do not build an hours case where there is no hours difference.

A swap that was never possible leaves no row. Steps with actuals stay locked to the machine that ran them, so an operation somebody wanted to move mid-run and could not will not appear. The audit records the swaps that happened, not the ones that were wished for, and the frustration is invisible.

Missing reasons cap the diagnosis. The audit shows the reason if one was entered. Without it you know the direction and the frequency but not the cause, and direction plus frequency will get you to the right station without telling you which of three problems it has.

History starts when logging did. Audit rows exist for events captured after audit logging was in place, and older history is not reconstructable. Read long trends with that in mind rather than concluding swapping recently began.

Frequency is not always capacity. A pair that repeats might reflect a tooling constraint, a tolerance only one machine holds, an operator's preference, or a routing whose hours were wrong on the primary all along. The count reliably tells you where to look. It does not tell you what you will find, and checking the routing against variance data is often the cheaper explanation to eliminate first.

The takeaway

Your plant already produces capacity evidence every time somebody keeps a job moving by putting it on a different machine, and that evidence is more current than any monthly average because it was generated under real pressure. Count a quarter of swaps by direction, and three things fall out: the station work keeps escaping, the machine your plan depends on without saying so, and the hours you spend on the habit. Then respond per pair rather than per plant, configuring the substitutions that repeat and leaving the genuine exceptions alone. To see the swap trail against your own stations, book a demo of EDGEBIC, and if you are running the older platform, the move from RMDB to EDGEBIC brings the audit with it. Read this next to how alternative work centers keep a hot job moving and how the audit trail settles schedule change disputes.