Operations.

The framework

1. 01

   Your forecast is built from orders, not from sales.

   This is the first and biggest of the four causes, and it runs without anyone doing anything wrong. You see the orders your immediate customer places, you read a rise in them as a rise in demand, you lift your own forecast, and you add safety stock before ordering from the tier above. So does the tier above you, and so does the one above that. Nobody's lying, and the number arriving at the factory bears little resemblance to the shelf. The amplification tracks the lead time too: the longer the wait, the more cover each tier holds against it.

2. 02

   Batching, forward buying, and the scramble for scarce supply.

   The other three are structural, and each has a cure sitting next to it. Batching: you order monthly because placing an order is expensive, so the tier above sees a pulse instead of a trickle — an ordering cost is a changeover by another name, and the batch morning's rule applies unchanged: attack the cost, not the schedule. Forward buying: your promotions create purchases unrelated to consumption, and the trough afterwards costs as much as the spike. Gaming: allocate scarce supply in proportion to orders and you've taught every customer to ask for more than they need — allocate on past sales instead, and make cancelling an inflated order cost something, so nobody can hold a place in the queue for free.

3. 03

   What would it take to see the shelf?

   The strongest single countermeasure is giving every tier the same end-demand figures, so nobody is forecasting from a signal that has already passed through three sets of hands. In practice that means agreeing to share point-of-sale or consumption data with your supplier, which feels like handing over commercial information and is in fact the cheapest inventory reduction available to either of you. Start small and start asking: one supplier, one customer, and the question of what they'd need in order to plan from real demand rather than from your orders.

4. 04

   Shorten the lead time before you improve the forecast.

   Amplification compounds along the length of the pipeline, so every week you take out of it removes cover from every tier at once — and it does that without asking anyone to forecast better, which is the improvement that never actually arrives. Hunt for the weeks that aren't manufacturing time: the ordering interval, the approval sitting in someone's inbox, the shipping mode somebody chose years ago and hasn't revisited, the monthly planning run. Then look hard at your own conduct in the last shortage, because inflating one order teaches a supplier to discount all of them.