Sometime in the early 1990s, logistics people at Procter & Gamble pulled the order history for Pampers. Sales in the shops wobbled a little, as sales do. Then they looked at what their distributors were ordering, and the wobble had become a swing. Then they looked at their own orders to suppliers like 3M, and the swing had become a lurch.
Nothing had happened at the far end. Babies are the steadiest customers in the economy: they get through diapers at a rate indifferent to Christmas, interest rates and the news. The demand never moved. The orders went wild. P&G named it the bullwhip effect, and Hewlett-Packard soon found the same shape in printers — reseller, then printer division, then the division making the chips, each hop wider.
Set that against every other wave here. A ripple flattens; a sound fades. Amplitude is what a travelling wave spends on the journey. This one arrives larger, and nothing is added to it — no new customers, no extra appetite, nothing pumping away upstream. The gain belongs to the medium, and the medium is people deciding sensibly.
Start with the delay, because it is the whole difference from last time. A nephron regulates itself with a wire from load back to tap that is one cell long: the tubule brushes its own glomerulus and the correction lands instantly. A supply chain runs the same corrective loop with the wire four tiers and several weeks long. That length is not a detail — it is a change of kind. You order to cover a shortfall; the goods that fix it are already in transit and invisible to you; so you order again. When both loads land you are swimming in stock, so you order nothing, and the tier above watches demand fall off a cliff. A correction delivered late is not a weaker correction. It is a push, and pushes at the wrong moment build waves instead of damping them.
Jay Forrester produced these swings in simulation in 1958, and at book length in 1961, from a model with no human players in it at all. John Sterman put people inside the loop and measured them. His beer distribution game — retailer, wholesaler, distributor, factory, no talking, nothing to go on but the order slip from below — reliably produced the oscillation in 1989, and his diagnosis was human: players suffer misperceptions of feedback. Above all they discount the supply line, ordering again for beer already on its way to them.
Which is where it gets good, because Lee, Padmanabhan and Whang set out in 1997 to show the opposite — the same amplification from players who make no mistake at all. They found four causes, and any one of them is enough on its own.
Forecast updating. Every tier forecasts from the orders below it rather than from the customer, and an order is need plus safety stock scaled to that forecast. So each desk re-forecasts a number that was already a forecast. Longer lead times, bigger cushion, bigger distortion.
Order batching. Nobody orders continuously. You wait for a full truckload, or for the monthly planning run. That alone converts a smooth trickle into a monthly spike — and the runs synchronise, because everyone’s month ends on the same day.
Price fluctuation. Promotions make buying ahead the correct move. Lee and colleagues report that around eighty per cent of grocery trade between manufacturers and distributors was forward buying, one industry observer calling the practice the dumbest marketing ploy ever. The buying pattern stops resembling the consumption pattern, deliberately.
Shortage gaming. When a maker rations pro rata, asking for more than you need is how you get what you need. HP could not meet demand for the LaserJet III, rationed it, watched orders surge, then watched resellers cancel the moment the constraint lifted — millions of dollars, and no way to tell a real order from a phantom.
So: two complete explanations of one pattern, and both are right. Remove the people and the structure still generates it. Remove the four structural causes and the people still generate it. The bullwhip is not held up by a mechanism but by a committee of them, any one of which will do — which is why it survives every remedy. Kill one leg and the rest take the weight. Systems with redundant causes are the hardest things in the world to switch off.
Worth saying what this is not. A queue also turns variability into cost, but a queue absorbs: it converts variability into waiting, and what accumulates is a stock of people or parts. This does the opposite: it converts variability into more variability and hands it upstream with gain. One is a shock absorber; the other an amplifier.
And it is not everywhere, which took the literature a decade to admit. Cachon, Randall and Schmidt went looking in US industry data: wholesalers show it, most retailers and most manufacturers do not, and seasonality explains a fair share of what looks like amplification from a distance. A property of particular chains, not of chains as such.
Then 2020: carmakers read a collapse in demand and cancel their chip orders; the foundries hand that capacity to laptops and consoles; demand returns early; everyone orders multiples of what they need from every supplier at once. Thirty years of remedies, and the shape came straight back.
≈
One loop I’m watching
Next: a layer of the sky that is torn down every night and rebuilt every morning, and has never once failed to be there. Sunlight strips electrons off the upper atmosphere; the electrons find their way back to their atoms in the dark; and the ionosphere is nothing but the running difference between those two rates. The lowest part of it vanishes after sunset, which is precisely why a distant AM station arrives so cleanly at midnight and not at noon. A hundred years of radio, bounced off something with no container, no boundary and no material of its own. Next time.