A queue in front of a busy machine looks like a clear message. The machine is full, orders are late, so we need another machine.
My recent M365.FM episode, Should You Buy Another Machine, or Change the Production System?, argues that this reading is often wrong. The queue tells you that work is waiting. It does not tell you why. This week’s note is about the order of operations: look at the system and test your options before you commit money to capacity.
The busy machine is not the whole story
Utilization shows that a resource is working. It does not show whether the work was the right work, or whether the output can go anywhere. A machine can run all day because planners keep feeding it, while the parts it makes block or pile up downstream.
The episode’s example is a hypothetical plant with a specialist work center in the middle of the route. Late orders grow, overtime rises, and someone proposes a second machine. On closer inspection, inspection downstream has a backlog and one technician. A shared fixture is not always free. An operator is covering another area. With a parallel machine, the queue would simply move to inspection, and shipments would not rise in proportion.
That is the Lean idea of the constraint, seen from the buying side. Throughput is what ships in acceptable form, not what a single machine produces. Add capacity at one point and the constraint moves to the next limit.
Why the simple math misleads
The usual business case divides planned hours by cycle time. A related M365.FM article, Why Simple Capacity Math Fails, explains the problem: rated speed describes the equipment under ideal conditions, not the factory. The formula leaves out setups, missing material, quality holds, operator cover, maintenance and product mix.
ERP planning can add to the false comfort. Another article from the same series, Why ERP Production Dates Are Fake, describes how infinite-capacity planning places work into time buckets without checking whether machines, operators or tools can handle it. Its example is 20 hours of work in a day that has 8.
None of this means ERP is bad. It records demand, due dates and inventory. As the episode describes it, ERP does not model detailed machine sequence, setups or shared resources on its own. MES, in the episode, is the system for execution and traceability.
Describe the system, then test it
The episode lays out a sequence that I find useful as a checklist before any purchase:
Ask whether you have a machine problem or a system problem, and define the commercial result you want to increase: finished, shippable product.
Challenge the investment case. What does the capacity estimate assume about setups, downtime, scrap, demand mix and labor?
Map the flow from release to shipment and find where work starves or blocks.
Describe the system through product (demand, due dates, batch size), process (operations, routing, inspection and rework paths) and resource (machines, labor, fixtures, tools, calendars).
Run a discrete event simulation as a decision test, starting from a baseline and changing only what the option affects.
The episode suggests testing demand ranges, not a single forecast, and measuring shipped throughput, lateness, queue movement, labor load and downstream strain. It also compares alternatives. A new machine is one experiment. A different shift pattern for the constrained product family is another, and it needs maintenance windows, handovers and sustainable staffing. A third is changing release rules and batch sizes, meaning when and how much work enters the system.
The episode is clear that it is not against machine investment. The point is that the decision should rest on a tested model of the whole system.
Where Microsoft tools fit
The episode describes Microsoft Fabric as a possible governed shared data layer, with Power BI for shared measures and reports. Azure supports integration, IoT paths, identity and security. It also says plainly that Fabric does not replace MES or finite scheduling, and that machine control and safety functions stay on the shop floor. Those boundaries matter when someone proposes a dashboard as the answer to a capacity question.
Planning data is the same story. The related urgent order episode shows how a rush request collides with a schedule that is already committed, and why options should be simulated before work is moved.
The takeaway
Before you buy, write down what you expect the new capacity to do to shipped orders. Then check that claim against the route, the people and the data. If you cannot test it, you are not ready to spend on it.


Writing down what the new machine should do to shipped orders is the best line here. I'd add one cheap test first. Track one week of late orders and note where each one waited the longest. If most of them waited at inspection, a second machine won't ship them any sooner.