Retail replenishment decisions — Every replenishment system forecasts demand and then applies a rule. The forecast gets the investment. The rule decides the outcome.
Retail replenishment decisions: bu ne anlama geliyor
Retailers spend heavily on demand forecasting and comparatively nothing on what happens to the forecast afterwards. Then they are surprised that better forecasts did not produce better availability.
The forecast produces a number. A replenishment rule turns that number into an order: safety stock policy, minimum order quantity, case pack rounding, delivery schedule, shelf capacity. By the time those constraints have been applied, the forecast’s precision is mostly gone.
The rounding problem nobody models
A forecast of 7.3 units becomes an order of one case of twelve. Whether that is right depends on shelf space, on shelf life, on what else is arriving, and on whether next week’s forecast is rising or falling. Almost no replenishment system considers the last of those.
Where an agent helps
Not by forecasting better. By making the ordering decision with all the constraints visible at once, and by explaining which constraint bound the decision.
“Ordered 12 rather than 6 because the case pack is 12 and the shelf holds 18” is a sentence a category manager can argue with. A number alone is not.
What the costume hides
Framing replenishment as forecasting has a specific consequence: it makes the forecasting team accountable for an outcome they cannot control. They are measured on error, and error is not what produces a stockout. A perfectly forecast SKU stocks out when the reorder point was set for a lead time that no longer holds.
The decision has four inputs, and the forecast is one of them. Lead time and its variability is the second. The cost asymmetry between holding and stocking out is the third. And the review cycle — how often the decision can be revisited — is the fourth, usually unexamined and often the binding constraint.
The asymmetry nobody writes down
Ask what a stockout costs and you will get a number that is the lost margin on the unsold unit. That is the floor, not the cost.
The real figure includes the customer who buys the substitute and keeps buying it, the store that stops trusting the replenishment system and starts over-ordering defensively, and the planner hours spent expediting. None of it is in the ERP, all of it is real, and the ratio between it and holding cost is what determines the correct safety stock.
An organisation that has not written this ratio down has not made the decision; it has inherited whatever ratio the system defaults to.
Where the model actually helps
Not in producing a better point forecast. In producing a distribution, and in connecting it to a lead time that is also a distribution.
A reorder point computed from an average demand and an average lead time is correct for a world in which neither varies. The interesting cases are the ones where both are elevated at once — a promotion during a period when the supplier is already slipping — and those are invisible to a calculation built on averages.
What to change first
Look at the review cycle before touching the model. A weekly review on an item with a four-day lead time is holding safety stock to cover a delay the process itself creates.
Then look at the lead time data. In most systems it is a single number entered when the supplier was onboarded and never revised. The order history contains the truth, and it is usually a wider distribution with a longer tail than anyone assumed.
Only then is a forecasting improvement worth the effort — and by that point it is often no longer the largest available gain.