Reacting versus anticipating
A reorder point triggers when stock falls to a level. That is reactive by design - by the time it fires, the lead time clock starts and a stockout may already be unavoidable.
Predicting the stockout instead gives a window: this line will run out in about eleven days, the supplier takes fourteen, so act now. That difference is where the value is.
What goes into the prediction
- Demand forecast for the line, at the location that matters.
- Current stock, with an honest view of its accuracy.
- Inbound orders and their realistic arrival dates, not the promised ones.
- Lead time distribution for that supplier and item, not the quoted figure.
- Any known upcoming events - a promotion, a large order, a seasonal shift.
Points three and four matter more than people expect. Planning on promised delivery dates when the supplier's actual history says otherwise builds the optimism straight into the projection.
Phantom inventory
The most common reason a stockout prediction fails is that the recorded stock figure is wrong. The system says fourteen; the shelf has three.
| Cause | Detection |
|---|---|
| Theft and shrinkage | Cycle counts, sales stopping while stock shows available |
| Receiving errors | Reconciliation against delivery notes |
| Misplaced stock | Physically present, not where the system says |
| Damage not written off | Counted as sellable, is not |
| Returns processed incorrectly | Stock added that never came back |
A useful and often overlooked signal: a line showing stock available that has sold nothing for an unusually long period, when comparable stores or the previous pattern say it should have. That is frequently phantom stock rather than a demand change.
Prioritise the alerts
Predicting stockouts across a large catalogue produces more alerts than anyone can act on. Ranking by consequence rather than probability is what makes it usable.
Rank by expected lost margin - probability of stockout multiplied by the value at stake - and by whether the item is substitutable. A stockout on a product with a close alternative on the same shelf costs far less than one on a line customers came specifically for.
Measure prevention, not prediction
The awkward thing about a stockout prediction system working well is that the predicted stockouts do not happen, because someone acted. Accuracy measured against outcomes will look poor.
Measure the operational outcome instead: stockout incidents, availability on key lines, and expedited orders avoided. And keep a record of which alerts were acted on, so prediction accuracy can be assessed on the ones where nothing was done.
The stockout you predicted and prevented will look like a false alarm. Measure availability, not accuracy.