In a weekly production planning meeting, a proposal to run the same product code four times a month instead of once tends to pass without material objection, supported by three arguments that are each independently sound: faster response to customer pull, less cash immobilized in finished goods, and later commitment against a demand signal that may still move. What the meeting does not discuss is that the month's total setup hours have just been multiplied by four, and the reason it is not discussed is architectural rather than intellectual — setup hours are never aggregated as a single reported line anywhere in the system, dissolving instead into non-productive time inside the operations manager's capacity table and reappearing in cost accounting only as a subordinate allocation within manufacturing overhead. The improvement in inventory turns, by contrast, arrives at month-end on its own slide, named, quantified, and attributed.
On the procurement side the same pattern presents itself from the opposite direction. Splitting an annual requirement into three or four releases rather than committing it in a single order is a defensible preference, since it distributes supplier exposure and defers cash outflow; yet each split converts what could have moved as a full truckload into partial shipments, turns one goods-receipt transaction into three, and replaces a single incoming quality inspection with three separate sampling cycles. None of these incremental transactions surfaces in the purchasing function's performance measures, which are constructed around unit price and payment terms, and all of them settle into the fixed-overhead budgets of logistics and quality — budgets that are, in most organizations, reviewed in aggregate at the annual planning cycle and interrogated line by line almost never.
The common name for both patterns is small-batch inefficiency — the disproportionate rise in unit cost that follows when batch quantities are reduced past the point where per-batch fixed costs can be absorbed — and its mechanism is arithmetically unremarkable. Setup time, first-article scrap, tooling changeover, cleaning and validation cycles, shipping documentation, goods receipt, and incoming inspection all occur once per batch irrespective of batch size, so halving the batch doubles the share of those costs carried by each unit and quartering it quadruples that share. The curve is hyperbolic rather than linear, which explains the characteristic institutional blindness: the first few reductions are genuinely imperceptible in the cost data, while beyond a certain threshold each additional split generates more cost than the sum of every split preceding it.
This tendency is better read as conditional rationality than as error. Where demand variability is high, product life cycles are short, model obsolescence is rapid, or material is perishable, small batches are genuinely the superior configuration, because the dominant risk in those conditions is not setup cost but having produced precisely the correct quantity of the wrong item. The problem therefore lies not in the preference itself but in the preference persisting after the condition that justified it has changed, and most visibly in the case where batches are reduced without any accompanying investment in setup reduction — a configuration that claims the benefit of flexibility without paying its engineering price. To the degree that the measurement architecture renders this distinction invisible, the decision maker is pushed, predictably and rationally, toward further fragmentation.
The productive question is consequently not what the batch size should be. Batch size is a dependent variable of measured setup cost rather than an independent planning choice: where setup consumes an hour, the economic batch is large; where setup has been engineered down to six minutes, it is small; and the distance between those two states is a capital and engineering commitment, not a scheduling preference. Every batch decision taken without a measured setup figure is therefore not a cost decision at all but a cost-transfer decision — a movement of value from a line that is measured, reported, and rewarded, namely inventory, into lines that are neither measured nor reported, namely available capacity and manufacturing overhead. The transfer is real, its direction is consistent, and its accumulation is invisible by construction.
The first institutional bill for that transfer accumulates in the capacity report. Where line utilization reads in the seventies and the lost balance is not decomposed into setup, validation, and changeover, the apparent headroom is systematically overstated; and where a growth plan is built on that reading, the true headroom is found to exhaust itself far earlier than modeled once volume actually arrives, driving the organization toward a new line, a new cell, or a new shift pattern before it has extracted the throughput its existing assets could still deliver. Taken as an order of magnitude rather than a precise figure, the cash impact of a capacity commitment pulled forward by one or two planning cycles will frequently exceed, by several multiples, the entire working-capital saving that the batch reduction was undertaken to produce.
The second bill collects in quality and delivery performance. Defects do not distribute evenly across a production run; first-article scrap, dimensional drift, and operator error concentrate in the interval immediately following changeover, so increasing the number of setups increases the expected defect count as a matter of statistical structure rather than of workmanship. Once rework hours, accelerated tooling and fixture wear, the planning function's repeated schedule reconstruction, and the elevated damage and transit-delay probability of partial shipments are added to that, the speed that small batches were expected to deliver frequently fails to materialize: lead time does not shorten, its variance widens, and the reliability the customer perceives — which is a function of variance, not of average — deteriorates accordingly.
The third bill becomes legible at the diligence table when the company enters a transaction process. Presented with unit costs that fail to decline with volume at the expected rate, overhead variances that cannot be explained between comparable periods, and an overtime line uncorrelated with demand seasonality, financial and operational diligence teams typically converge on one of two conclusions: reported margin sits above sustainable margin, or the growth case requires materially more capital expenditure than the plan discloses. That finding is then priced into the transaction structure as a multiple discount, a pre-closing adjustment condition, or an earn-out that ties a portion of consideration to demonstrated operational improvement — and in none of those configurations does the seller obtain a better outcome for a cost it has never quantified than the buyer's own estimate of it.
The mechanism that neutralizes this tendency is decision architecture rather than individual discipline, and it separates into four components. The first is measurement: a record of setup duration by product family and by machine, established with a stopwatch rather than an estimate, maintained alongside per-batch scrap, freight transaction cost, and goods-receipt burden. The second is a decision rule: batch size bound to an explicit threshold in which measured setup cost is weighed against inventory carrying cost, with the threshold differentiated by product family rather than set as a plant-wide constant. The third is authority design: an express definition of who may split an approved batch and on what grounds, with the rationale recorded at the moment of proposal rather than at the moment of approval. The fourth is cadence: periodic recalculation of the thresholds, since a shorter setup should legitimately produce a smaller optimum.
BEIREK's intervention in a structure of this kind begins not with a recommendation about batch size but with the construction of the ground on which any such recommendation could rest. In the operations workstreams we run for capital-intensive facilities and multi-asset industrial groups, we first build the setup and per-batch fixed-cost record, then reconcile that record against the lost-time decomposition in the capacity report, so that the share of the utilization gap attributable to genuine idleness can be separated from the share attributable to changeover. From there the batch-splitting decision is bound to defined thresholds and named authority, a decision log is operated that captures rationale at proposal, and a review cadence is established in which the thresholds are recalculated rather than inherited.
The second leg of that intervention is the translation of the operational finding into financial language. Once setup-driven lost capacity becomes measurable, the portion of a planned line investment that is driven by real demand growth can be separated from the portion driven by batch fragmentation, and that separation does two things at once: it strengthens the capital expenditure case presented to an investment committee, and it allows questions about margin quality — in a sale process, a partnership negotiation, or a lender's technical review — to be answered from the company's own data rather than from the counterparty's assumptions. What determines the valuation of an operating asset is, in a substantial share of cases, not the performance itself but the demonstrable identification of the mechanism producing it.
A small-batch policy becomes a genuine strategic lever in a facility where setup cost has been engineered down, while in a facility where setup cost has never been measured it amounts to nothing more than the migration of cost from one line item to another; the two situations look identical from the outside, presenting as the same planning decision in the same meeting, and they separate only once a measurement basis exists. The question a manufacturing organization might therefore put to itself is not how small its batches have become, but when the last setup was timed, by whom, and against which product family the resulting figure was recorded.