Last December's sales are useful, but copying them into this December's purchase order can be risky. Promotions, lead times, returns and stockouts all change what the sales history actually means.
A seasonal item sold out early last year. Historical sales show a sharp drop after the stockout, but demand may have stayed high. A naive forecast will underbuy again. Planners should mark censored sales, look at search and waitlist signals where useful and treat the resulting estimate as uncertain.
Reconstruct demand
Separate observed sales from demand suppressed by unavailable stock. Note promotions, holidays and new product launches. Forecast at a level the business can act on; a category total may hide a critical size or color shortage.
Human overrides are useful evidence. If planners consistently increase a forecast for a festival, the model may lack a calendar signal. If overrides consistently worsen outcomes, the team may need better feedback rather than stricter controls. Record the reason and compare fairly. A planning system should help people learn which judgment adds value.
Include the supply side
Bring supplier lead time, minimum order quantities and warehouse capacity into planning. Show uncertainty rather than one exact number. A planner may choose more buffer for a long-lead item than for a product that can be replenished in two days.
Review decisions after the season
Compare forecast, actual demand, markdowns and lost sales. Record overrides and why they were made. The purpose is not to win a forecasting contest; it is to make stock decisions with a clearer view of risk.
