Croston, and why the forecast is not a black box
Chemical demand is intermittent, which is the case a moving average handles worst. Here is the model behind the days-of-supply figure and the alert it raises.
Intermittent demand breaks the obvious method
A reagent sits untouched for three weeks, then a validation run takes four litres in two days. Average that over a month and the forecast says you use a little every day, which is true of nothing that ever happened.
The consequence is not academic. A smoothed average under-reacts to the burst and over-reacts to the silence, so the reorder point drifts to a number that is wrong in both directions depending on which week you look.
What Croston separates
The Croston method splits the series in two: how much is drawn when a draw happens, and how long the gap between draws tends to be. Each half is smoothed on its own, and the forecast is the ratio of the two.
That structure matches how a laboratory actually consumes. Where a series is regular enough that the split adds nothing, a Holt-Winters fallback takes over.
Days of supply is the number people read
The forecast is not shown as a curve. It is turned into a days-of-supply figure per chemical: at the rate this is going, the shelf runs out in this many days.
When that figure drops under the reorder configuration, the chemical raises a stock alert, in the app and as a push notification, and the alert drops into the order list. The forecast is not a dashboard widget. It is the thing that starts a purchase.