Demand Sensing
Demand sensing is the in-season practice of estimating the current rate of demand from recent signals and using it to change decisions that can still move.
Demand sensing is the in-season practice of estimating the rate at which demand is running now — from recent signals such as point-of-sale sell-through, e-commerce orders and add-to-cart, returns, the on-hand and in-transit position, and retailer or dealer sell-through — and using that estimate to change the decisions that can still move inside their lead times. It works on a horizon of weeks and complements the pre-season forecast that sized the buy rather than replacing it.
For the concept of demand sensing in full, read Demand Sensing in Retail: the signals and their latency, the grain each one can be trusted at, which decisions a sensed rate can still reach, censored demand, the four failure modes, a worked example and how sensing differs across ten verticals. If you are evaluating demand sensing solutions or software, read What Is Demand Sensing Software?: the category defined, what it ingests and outputs, how it differs from demand planning and merchandise planning software, and where it pays and cannot help by vertical.
Demand sensing, demand forecasting and demand planning are used interchangeably and are not interchangeable. Demand forecasting sizes the buy months ahead, demand planning commits the supply against that size, and sensing reads the rate demand is running at now. The three differ by the question they answer, not by technique: how much should we buy, how do we get it, and how fast is it actually moving.
What counts as a sensing signal
A signal qualifies when it is recent enough to still be acted on and granular enough to attribute to a decision. Point-of-sale sell-through, e-commerce orders, add-to-cart, returns, the on-hand and in-transit position and retailer or dealer sell-through all qualify; a monthly shipment total does not, because by the time it lands the decisions it could inform have closed. Latency is the test, not novelty. Two properties decide whether a given signal is usable: how long it takes to arrive, and the grain it arrives at — a signal reported by style cannot settle a size-level question. The full guide covers each signal's latency and trustworthy grain.
Why sales history is not demand
Sensing reads what sold where stock existed, not what customers wanted. A position with nothing on hand records zero sales, a broken size run records slow sales on the sizes left, and a late receipt records weak early weeks that were really an empty floor. Each reads as low demand unless the rate is adjusted for availability, which is why in-stock rate and a lost sales estimate are prerequisites rather than refinements. Without them, a winner that sold out early looks like a style that slowed down — exactly when it should be chased.
RetailNorthstar reads in-season sell-through against plan and holds it beside allocation and open-to-buy on a shared data model, so a sensed gap reaches the buy-depth and allocation decisions still open. See how RetailNorthstar handles allocation →