Why Did One Extra Order Empty the Factory? Everybody Guessed About Everybody Else

Consider a chain running from shoppers to a shop, then a distributor, then a manufacturer, then a supplier of raw material. A modest and temporary rise in what shoppers buy produces a larger rise in what the shop orders, a larger one again from the distributor, and so on, so the variation at the far end of the chain is several times the variation that started it. The pattern is visible in real data, and the further a business sits from the final customer, the wilder its demand appears and the less it resembles what people actually bought.

Four ordinary and individually sensible behaviours cause it:

• •Each link sees only its immediate customer's orders, not real demand
• •So a rise is read as a trend rather than as noise
• •Orders are batched into economic quantities rather than sent continuously
• •Buyers order extra when they expect prices to rise or discounts to end
• •During shortages buyers inflate orders hoping for a proportional share
• •Each link adds a safety margin on top of its forecast

The effect is taught through a simulation used in business schools since the 1960s, in which teams take the roles of links in a chain and can see only their own orders and stock. Customer demand rises once, modestly, and stays there. Participants invariably produce wild oscillations, accumulating enormous stock and then running out entirely, and the pattern repeats across groups regardless of experience. Debriefing usually reveals that participants blamed each other, and the point of the exercise is that the structure of the system produces the behaviour rather than the people in it.

The same amplification appears in systems that have nothing to do with goods, which suggests the cause is structural. Road traffic does it, where one driver braking produces a wave of increasingly hard braking that travels backwards and can stop cars entirely, kilometres behind, for no visible reason. Electricity grids show it in scheduling. Staffing levels in organisations oscillate the same way, hiring hard after a busy period and freezing after a quiet one. In every case a delay between observing a signal and responding to it, combined with responding to the response rather than to the original, produces the swing.

The fixes all attack information delay or ordering incentives rather than the people involved. Sharing actual sales data with every link removes the need to infer demand from orders, which is why retailers pass sales figures directly to suppliers. Shortening lead times reduces how far ahead anybody must forecast. Smaller and more frequent deliveries remove the batching. Stable pricing removes the incentive to buy ahead. Allocating scarce supply on past sales rather than current orders removes the incentive to inflate. The pandemic years demonstrated the whole phenomenon at scale across many industries at once.

A small change in consumer demand grows at every step back up a supply chain, because each link sees only its own customer's orders, batches them, buys ahead of price changes and adds a safety margin. A long-running business school simulation reproduces it reliably, showing the structure causes it rather than the participants. Sharing real sales data and shortening lead times are the main fixes.