Lost Sales

Lost sales are the units of demand a brand turned away because the product was not available where and when the customer wanted it: the size that was out, the shade missing from the ladder, the style not yet delivered, the door never allocated. At the item level they are the difference between unconstrained demand — what customers would have bought with stock in place — and recorded sales; to the brand, only the part that did not substitute to another size, color or style is lost.

Lost sales never appear in sales history, because a transaction that did not happen leaves no record. A POS system logs what sold, not the customer who asked for a medium and left. Every process that learns from sales — a forecast, a size curve, a sensed rate, next season's buy — learns from demand with the lost portion removed, and removes the most from the positions that sold out first. A style that sold out early is recorded as a style that sold less. Any estimate therefore has to separate substitution — the customer who bought another size, color or style — from demand that left.

How lost sales are estimated

No method observes lost sales directly; each infers them from something observed, and each carries an assumption worth recording beside the number.

• →Time in stock. Scale a position's in-stock sales to the whole period. It assumes demand was even across the period.
• →Size or shade share. Infer an out-of-stock size's demand from the sizes still selling and its planned share of the size curve. It assumes the curve was right and the other sizes were not absorbing substitutes.
• →Comparable positions. Compare a door that stocked out with similar doors in its cluster that did not, over the same weeks.
• →Interest signals. Views and clicks on out-of-stock sizes and back-in-stock requests show that demand existed and roughly where, without sizing it on their own.

Illustrative example

The figures below are illustrative, chosen to divide cleanly, and not drawn from any brand.

One door's size medium sells 6 units in the first 3 days of a 7-day week, then stocks out. Scaling to the full week gives 6 ÷ 3 × 7 = 14 units of estimated demand, so 14 − 6 = 8 units were lost — more if the stockout days were the heavier trading days.

In a week when medium is out across the fleet, the other sizes sell 60 units and medium's planned share of the curve is 25%. The selling sizes represent the other 75%, so estimated total demand is 60 ÷ 0.75 = 80 units and medium's lost demand is 80 − 60 = 20 units. That is an upper estimate: medium customers who bought a large instead inflate the 60 and were not lost to the brand.

Where lost sales change decisions

In season, estimated lost sales on a winner are the evidence for a holdback release or a chase, and they correct the rate demand sensing acts on. After the season, the hindsight adds them back — labelled as estimates, with the method and the in-stock rate behind them — before writing next season's buy and curve. Without that step, next season's plan is built on the demand the stockouts left behind. In wholesale, a cut order line is a visible lost sale at sell-in; the account's lost sell-through at its own doors is not visible to the brand at all.

RetailNorthstar keeps the buy, allocation, door-level sell-through and inventory position on a shared data model, so that record is still attached to the plan when the team estimates, in hindsight, what the season's stockouts cost. See how RetailNorthstar handles allocation →