The Price Of Keeping Something Spare

AFor half a century the dominant instruction to anyone running a factory was to hold less: less stock on the shelf, fewer suppliers to manage, less capital idling in a warehouse. The instruction worked, and the savings were real. Then, between 2020 and 2022, a run of unrelated shocks — a pandemic, a container shortage, a fire at a single semiconductor plant in Japan — showed how little slack was left anywhere in the system. A car maker that had never held more than three days of certain small components found itself unable to finish vehicles that were otherwise complete. The word that entered boardroom vocabulary afterwards was resilience, though what it should mean in practice was, and remains, contested. One school holds that resilience is a quantity: stock, spare capacity, second sources, all of which can be bought. A second holds that it is a speed: the rate at which a firm can find out what has broken and route around it. The distinction sounds academic. It decides where several billion dollars a year is now spent.

BUlrich Takahashi, an operations researcher at the Hallberg Institute in Gothenburg, has tried to settle the first half of the argument with data rather than anecdote. Her team assembled records from 1,430 manufacturers across eleven countries covering the disruptions of 2020 to 2022, and measured, for each firm, the interval between its first missed delivery and its return to normal output. Firms that had entered the period holding more than nine weeks of critical inputs resumed normal output in a median of five weeks; those holding less than three weeks took eleven. That gap, however, held only where the disruption began with the failure of a named supplier. Where the shock came from the demand side — orders collapsing, then returning in an unexpected shape — the well stocked firms did no better than the rest, and a few did slightly worse, having filled their warehouses with the wrong things. Buffers, Takahashi concluded, are insurance against a particular accident rather than against uncertainty in general.

CIf the point of a buffer is to survive the failure of a named supplier, a firm must first know who its suppliers are, and Takahashi's second study suggests that most do not. Almost every large manufacturer can list the companies it buys from directly. Beyond that first tier the picture fades: in a survey of 380 firms conducted in 2022, fewer than one in five could identify the makers of the sub-components inside the parts they purchased. The consequences were measurable. Among firms caught by the semiconductor shortage, those that had mapped their second and third tiers before the event located the bottleneck in a median of four days; those that had not took twenty-seven, by which time the spare capacity had been contracted elsewhere. Mapping, Takahashi notes, is cheap beside inventory, and unlike inventory it does not spoil, drift out of specification or tie up capital.

DLeif Ramaswamy, an industrial economist at the Deerfield School of Management, does not dispute Takahashi's measurements but objects to the comparison they invite. Resilience, he argues, is never weighed against a perfect world in which nothing is stored and nothing goes wrong; it is weighed against what the same capital would have earned elsewhere. Tracking 240 listed manufacturers between 2015 and 2023, he found that those in the top quarter for inventory relative to sales earned, on average, one and a half percentage points less on capital employed each year than those in the bottom quarter — a charge paid annually against shocks that arrive perhaps twice in a decade. Ramaswamy volunteers two weaknesses. The firms chose their own inventory policies, so heavy holders may simply be firms whose demand forecasting is poor; and his figures stop at the factory gate, ignoring the losses that customers and downstream assemblers absorb whenever a plant halts.

EThe dispute has since acquired a legal edge. A directive in force across the European Union since 2023 obliges manufacturers of listed medical supplies to hold ninety days of critical inputs and to file an annual account of where those inputs originate. Takahashi regards the filing requirement, not the stockpile, as the useful half: a firm compelled to describe its second tier will learn things about itself that no quantity of stock can substitute for. Ramaswamy is blunter about the other half. A fixed number of days, he says, is a figure chosen because a committee had to write something down; it imposes the same holding on a producer whose input keeps for years and one whose input is worthless in four months, and it says nothing about whether a substitute exists. Neither researcher defends the sellers of resilience software, whose case studies, both observe, describe only the firms that survived.

FWhat is taking shape is less an answer than a partition of the question. Where an input is storable, standard and cheap beside the product it goes into, the quantity school is broadly right, and a warehouse is a reasonable form of insurance. Where an input is perishable, custom-built or revised every few months, stock is a wasting asset and the only defence is knowing early and moving fast. Most industrial economies contain both kinds of supply chain, in proportions nobody has measured, and the balance is shifting in directions the two researchers describe differently. More data will sharpen each finding without deciding between them, because the choice turns on which kind of input a given firm depends upon rather than on any fact still to be gathered. The framing both reject is the one heard most often in public: that a supply chain is either efficient or resilient, as though every firm faced the same accident.