There is a particular kind of calm that settles over a production floor that has found its rhythm. The changeovers happen in a familiar sequence, the machine operators know without being told which job follows which, and the planner's spreadsheet — or the MRP screen behind it — hums along matching demand to capacity with something approaching grace. Then the sales team lands a new product range. Then another. And the calm, without any single dramatic event to mark its passing, is gone. This article is about what happens in that interval — the period after the SKU count doubles but before anyone has admitted that the scheduling logic built for the old range is no longer fit for purpose. It names five specific assumptions that break, describes what breaking looks like on the shop floor, and offers concrete ways to catch each failure mode before it shows up in a late-delivery report.
Assumption One: That changeover time is essentially constant ¶
When a factory runs a narrow range, changeover times cluster tightly. A line that makes four variants of the same base product will have changeovers that vary by perhaps ten or fifteen minutes — close enough that a planner can use a single average figure and lose little accuracy. The moment the range expands to include genuinely different products — different viscosities, different tooling requirements, different cleaning standards because one product is allergen-bearing and another is not — that average becomes a fiction. A changeover from Product A to Product B might take twenty-two minutes. From Product B to Product C, if C requires a full allergen clean, it might take three hours and forty minutes. The schedule that was built on an average of thirty-five minutes is now, in practice, either wildly optimistic on some sequences or absurdly conservative on others. The failure mode to watch for is not missed changeovers — it is planner drift, the gradual, unacknowledged habit of scheduling the fast sequences and avoiding the slow ones, which quietly distorts which products actually get made and in what order.
Assumption Two: That every machine can make every product ¶
In a simple range, the assumption of interchangeable capacity is often accurate enough. Three presses, four moulding lines, two filling stations — in practice any order can go on any machine, and the scheduler treats the work centre as a pool. When the range doubles and the new SKUs arrive with their own geometric tolerances, their own die sets, their own substrate requirements, the pool begins to fragment into subsets. Line 3 can run the new wide-format products; Lines 1 and 2 cannot. Filler B has the low-shear pump required for the high-viscosity variant; Fillers A and C do not. The scheduling system, if it has not been updated to reflect these constraints, will continue to allocate work as though the pool were whole — and the planner will discover the mismatch only when a job arrives at a machine that cannot run it. The diagnostic sign is a rising rate of last-minute transfers between work centres, jobs being physically moved from one machine to another after scheduling, which the system records as capacity loss but which is actually a routing error.
Assumption Three: That lead times are stable across the range ¶
Standard lead times are among the most load-bearing assumptions in any scheduling system, and they are also among the first to silently degrade. When the original range was designed, someone — probably a planner or a production manager with long institutional memory — set the lead times based on observed cycle times, typical queue lengths, and a reasonable allowance for changeover. Those numbers were roughly right for those products. The new products are different: they may have longer cycle times per unit, they may require intermediate curing or resting stages that the old range did not, and they may queue at different work centres than the existing products do, creating local bottlenecks that the aggregate capacity figures do not reveal. The failure mode here is systematic lateness on the new SKUs specifically — not across the board, but concentrated on the additions to the range — which is often misread as a quality or supplier problem before anyone thinks to question whether the lead times assigned at product launch were ever correct to begin with.
Assumption Four: That the bottleneck is where it used to be ¶
Every production system has a constraint — a single resource, or a small set of resources, that limits throughput for the system as a whole. Goldratt's observation here is not theoretical; it is empirically verifiable on almost any shop floor within a morning's walk. In a stable, narrow-range operation, the constraint tends to be known, managed, and protected. Buffers are positioned in front of it. The scheduler prioritises its utilisation. The maintenance team knows to treat its downtime as the most expensive downtime in the building. When the product range expands, the constraint can migrate — and often does so without anyone noticing for weeks or months. A new product group may require substantially more time at a secondary process — a labelling step, a heat-treatment stage, a quality inspection that the old products passed in two minutes but the new ones require twelve — and that secondary process becomes the new system constraint. Meanwhile, the organisation is still managing, protecting, and buffer-stocking around the old one. The symptom is a well-utilised primary work centre and a growing queue at a process that was previously invisible in the schedule.
Assumption Five: That demand variability is spread evenly across the range ¶
When a planner builds safety stock or safety time into a schedule, that buffer is implicitly sized against a model of demand variability. In a narrow range, the planner often has years of demand history and a reliable intuition for where variability concentrates. The new SKUs arrive with no history and frequently with optimistic forecasts — the sales team, understandably, has sold in the new range with enthusiasm, and the demand signal in the system reflects aspiration as much as evidence. In practice, new SKUs often exhibit far higher demand variability than established ones: customers trial them in small quantities, order patterns are irregular, promotional activity is front-loaded. The scheduling system, if it applies the same buffer logic to new and old SKUs alike, will either over-produce the new products — tying up capacity and working capital in stock that does not move — or under-produce them when a cluster of orders arrives simultaneously. The diagnostic is not stockouts alone; it is the pattern of stockouts paired with excess stock of adjacent new SKUs, a see-saw that reflects a mismatch between the variability model and the actual demand signal.
How to catch these failures before they reach the delivery report ¶
None of these failure modes require sophisticated detection. They require, more than anything, the discipline to look at the right data at the right interval. For assumption one — changeover variability — the intervention is a changeover matrix: a simple table recording actual changeover times between every product pair that shares a work centre, updated quarterly as the range evolves. For assumption two — routing fragmentation — the fix is a routing audit triggered automatically whenever a new SKU is released, with explicit sign-off from the engineering or process team confirming which work centres can run the product and which cannot. For assumption three — lead time drift on new products — the practice is a structured post-launch review at ninety days: compare the planned lead time assigned at launch against the actual average lead time observed across the first twenty production runs. For assumption four — bottleneck migration — the tool is queue-length tracking at every significant work centre, reviewed weekly, with a threshold that triggers investigation when any queue grows beyond a defined multiple of its historical average. For assumption five — demand variability on new SKUs — the approach is a tiered buffering policy that applies higher safety stock coefficients to products with fewer than twelve months of demand history, reviewed and reclassified as history accumulates. These are not technology investments. They are disciplines — habits of observation that keep the scheduling logic honest as the range it serves grows more complex.
A note on pace: why the failures compound ¶
What makes these five assumptions particularly dangerous is not that any one of them is catastrophic in isolation. A planning system can absorb one degraded assumption and compensate — experienced planners do this intuitively, correcting for a bad changeover estimate here, manually adjusting a lead time there. The difficulty is that product range expansion tends to trigger all five simultaneously and gradually. By the time the on-time delivery rate begins to fall, the organisation is typically dealing with compounded errors: changeover sequences that are longer than planned, jobs routed to machines that cannot run them, lead times that were wrong from the start, queues building at an unmanaged constraint, and demand signals that bear little resemblance to actual customer behaviour. At that point, the instinct is to add capacity — more machines, more shifts, more people — when the correct response is to audit assumptions. Capacity additions applied to a system with broken scheduling logic do not improve delivery performance; they increase the volume of work moving through a flawed sequence.
A factory that doubled its SKU count without revisiting its scheduling assumptions has not grown — it has borrowed against its operational future, spending the goodwill that lived in its former simplicity. The debt is not unpayable. But it is collected, reliably, in late deliveries, in planner exhaustion, and in the slow erosion of a production floor's confidence in its own rhythm. The five assumptions named here are the places to start looking — not because they are the only things that break, but because they break first, and fixing them restores the conditions under which everything else can be managed.