The Factory Without A Warehouse

AA factory holding a month of components on its shelves is usually said to be buying insurance. The production engineer Kenji Arimoto, working at Nakabayashi Motors from 1958, proposed the opposite reading. Stock, he argued, is not a cushion against trouble but a device for concealing it: a machine that fails once a week, a supplier that delivers the wrong part, an operator absent every Monday — none of these has to be repaired while a buffer stands between the fault and the customer. Arimoto's remedy was to cut the buffer deliberately, letting each stage of assembly draw parts from the stage before it only as it consumed them, and to treat every stoppage that followed as a piece of diagnostic information. He compared inventory to water in a canal and defects to rocks on its bed: lowering the water does not create the rocks, it reveals them. The method was therefore never, in his own account, primarily a way of saving money on warehousing.

BNakabayashi's system travelled slowly. It was described in an English-language engineering journal in 1977 but attracted little attention until the recession of the early 1980s, when Western manufacturers began to examine the cost structures of their competitors in earnest. Adoption then became fashionable, and was often superficial. Plants installed signal cards and fixed delivery windows while leaving intact the practices — long production runs, quality inspection at the end of the line, contracts awarded each year to the cheapest bidder — that made small, frequent deliveries impossible to sustain. By 1990 the average North American assembly plant held 22 days of parts against 4 days in the plants Arimoto had advised, yet the gap in defect rates was narrower than that ratio would suggest, and a series of well-publicised failures established a folklore that the method did not travel.

CJoran Moreau, an industrial economist at the Rovnik Institute, has assembled records from 240 plants across eleven industries covering the years 1995 to 2015, and her conclusion is that the folklore is wrong but so is the enthusiasm. Plants that adopted the full system reduced parts inventory by an average of 61 per cent and defects reaching the customer by 38 per cent. When she separated the elements, however, most of the quality improvement tracked not the thinness of the stock itself but the changes that thin stock compelled: engineers stationed on the line, faults escalated within the hour, suppliers visited rather than telephoned. Her preferred formulation is that low inventory is a diagnosis, not a treatment. She is equally clear about the condition attached to it: the gains appeared only where a plant had first stabilised its production schedule, and in the fifty-one plants that imposed the system on a fluctuating schedule, costs rose.

DRahul Devaraj, who studies supply chains at Marlowe Business School, does not dispute Moreau's plant-level figures and directs his objection instead at the boundary she draws. A single plant is not the right unit of account, he argues, because inventory removed from an assembly line does not cease to exist; it moves upstream. His survey of 380 first-tier suppliers found that 62 per cent held finished-goods buffers specifically in order to meet delivery windows they could not otherwise guarantee, and that these buffers were carried at the supplier's own expense. Counted across the whole chain, he estimates, the reduction in stock is closer to a fifth than to three-fifths. The second half of his case concerns rare events. When fire destroyed a single chemical plant at Ostervik in 2012, four assembly factories stopped within nine days, and the losses of the following quarter exceeded a decade of the inventory savings the method had produced at those sites.

EThe disruptions of 2020 and 2021 were widely reported as the end of the method, and several manufacturers announced a return to holding stock. The measured change was smaller than the announcements. Across the plants Moreau has continued to track, average parts cover rose from 5.2 days in 2019 to 8.6 days in 2023 and has since fallen back, a movement she regards as ordinary insurance against a risk that is now recognised rather than a change of principle. She attributes the shortages instead to single sourcing — buying a component from one supplier in order to obtain a lower price — which she treats as a procurement decision separable from production scheduling. Devaraj accepts that the two are distinct on paper and denies that they can be separated in practice: frequent small deliveries reward proximity and deep coordination, both of which are expensive to maintain with several suppliers at once, so the economics of the method push firms towards single sources whether or not their managers intend it.

FWhat follows is disputed less in principle than in accounting. Both researchers expect the coming decade to bring dual sourcing of critical components, shared regional stockpiles and continuous monitoring of supplier finances, and both observe that such monitoring costs far less than it once did. Moreau's remaining doubt is environmental: replacing one weekly lorry with five daily ones raises transport emissions per unit, and she puts the increase at between 8 and 14 per cent on typical European routes, a figure she believes consolidation centres can largely erase. Devaraj's doubt is about who is counted. A chain optimised at the assembler's gate may simply relocate cost, risk and emissions to firms with less capital to absorb them, and any measure of efficiency that stops at the factory boundary will record that transfer as an improvement. Arimoto himself, interviewed in 1994, remarked that the method had been widely copied and rarely understood; both sides quote the line, and each takes it to be about the other.