Just in Case

Situation

For forty years, the dominant objective in supply chain management was the elimination of inventory.

The reasoning was sound and the evidence supported it. Inventory consumes working capital that could be deployed elsewhere. It requires warehousing, insurance and handling. It obscures quality problems, since a defective batch discovered three months after production cannot be traced to a cause. Above all, it conceals dysfunction: a process that is unreliable can be made to appear reliable by placing enough stock in front of it, and the buffer removes any incentive to fix the underlying fault.

The methodology that followed from this insight — pull-based replenishment triggered by actual consumption rather than forecast, with suppliers delivering in small frequent lots — was among the most consequential management innovations of the twentieth century. Firms that adopted it rigorously achieved inventory turns several times those of their competitors and released capital on a scale that transformed their balance sheets.

The approach worked because it rested on an assumption that held for four decades: that supply was abundant, transport was cheap and reliable, and disruption was local, brief and insurable.

Complication

Between 2020 and 2023 that assumption was tested across multiple dimensions at once, and it did not hold.

The instructive case is automotive semiconductors, because the failure was not caused by any physical destruction of capacity.

At the onset of the demand shock in early 2020, vehicle manufacturers forecast a prolonged contraction and cancelled chip orders accordingly — an entirely rational response under the operating model they had been running for decades. Foundry capacity, however, is booked months ahead and does not sit idle. The cancelled allocations were immediately taken up by consumer electronics manufacturers, whose demand was rising sharply.

When vehicle demand recovered far faster than forecast, the manufacturers attempted to return to the queue and discovered three things. Lead times for the mature process nodes used in automotive electronics had extended from twelve weeks to over fifty. Those nodes are produced on older, largely fully depreciated equipment that nobody was building more of, because the margins do not justify new capacity. And automotive volumes, though strategically significant to the buyer, represented a small share of foundry revenue, conferring little priority.

The result was the loss of millions of units of vehicle production, on account of components representing a trivial fraction of vehicle cost. Assembly lines with every other part available stood idle for want of a chip worth a few dollars.

Analysis

Three structural weaknesses were exposed, and they are general rather than sector-specific.

Visibility ends at the first tier. Most firms can name their direct suppliers. Comparatively few can identify their suppliers' suppliers, and almost none can map to the third or fourth tier. Risk, however, concentrates in exactly those depths, where specialised inputs are frequently produced by one or two firms serving an entire industry. A fire at a single specialist plant in 2021 disrupted output across the global automotive sector, and many affected manufacturers were initially unable to determine whether they were exposed.

Diversification is often nominal. A procurement function reports three qualified suppliers for a component and records the risk as mitigated. Investigation frequently reveals that all three source the same sub-component from the same plant. Genuine diversification must be assessed at the level of the actual constraint, not at the level of the contracting entity.

Demand signals amplify upstream. Modest variation in end consumption produces progressively larger swings at each successive tier, as every participant adjusts orders for both expected demand and desired buffer. This amplification is a structural property of sequential ordering with lag, well documented for six decades, and it means that the upstream tiers experience far greater volatility than the actual market ever exhibits.

Options

Four responses are available. None is free, and the correct combination is specific to the item.

1. Strategic buffering. Hold inventory selectively where the cost of stockout is high and the holding cost is low — low-value components whose absence halts a line. Cost: working capital, obsolescence risk.
2. Qualified dual sourcing. Maintain a second source in a different region, actually qualified and receiving a meaningful share of volume, since a supplier held at one per cent will not prioritise you in a crisis. Cost: typically five to ten per cent on unit price, plus qualification expense.
3. Design for substitution. Specify components with multiple interchangeable equivalents, and avoid designing around a single vendor's part unless the performance advantage is decisive. Cost: some loss of optimisation; engineering discipline.
4. Postponement. Hold inventory in generic form and differentiate late. A semi-finished item serving twelve final variants requires far less total stock than twelve finished ones, because pooled demand is proportionally less variable than its components.

Recommendation

The error to avoid is reversal. The pendulum in this discipline swings, and the version of resilience that follows a crisis is typically expensive, undifferentiated and abandoned within three years once no further crisis has occurred.

Efficiency should remain the default. It is not a mistake; it generated enormous real value and continues to. What the period exposed was not that the method was wrong but that it was applied uniformly — the same lean treatment to a commodity fastener with fifty suppliers and to a sole-sourced controller with a fifty-week lead time.

The recommendation is therefore segmentation. Classify every input on two axes: the cost of disruption, and the difficulty of replacement. The large majority of items will fall into the low-low quadrant and should be managed as leanly as possible. A small minority — typically five to fifteen per cent of line items, commonly accounting for the overwhelming majority of realised disruption cost — belong in the high-high quadrant and should be managed under an explicitly different policy: multiple genuine sources, strategic buffer, mapped to tier three, reviewed quarterly.

Resilience purchased this way is an insurance premium, and should be presented to a board in exactly those terms. It has a calculable cost, it will appear wasteful in every year in which nothing happens, and the correct question is not whether it can be eliminated but whether the premium is proportionate to the loss it covers.

Firms that answer that question item by item will outperform both those that never ask it and those that, having been surprised once, answer it identically for everything they buy.