Demand forecasting for retail and DTC: the complete guide
A practical guide to demand forecasting for retail and DTC brands, at style, channel and week level, with a confidence range and a way to check accuracy.
Last year plus a percentage is not a forecast. It assumes next year looks like last year, at every size and in every channel, and it quietly bakes last year's stockouts and markdowns into next year's buy. For a brand that is growing, changing channel mix and dropping new products, that assumption breaks in the places that cost the most.
What a real forecast predicts
Demand, not sales. Sales are capped by what you had in stock. Demand is what customers would have bought if the size was on the shelf. A forecast built on sales history alone learns your past constraints. A forecast built to estimate true demand learns the pattern underneath them.
At the level you actually buy
A number for the brand is useless for the buy. You commit orders by style and size, split across channels, phased across the weeks of a season. A forecast has to predict at that level, because that is where the money is committed and where being wrong shows up as broken sizes or dead stock.
New products and seasonality
Most of the risk sits in the styles with no history. Good forecasting models a new style from the ones it resembles before it has sold a unit. Seasonality should be learned from your own selling history, not borrowed from a generic retail curve, because your peak is not the category's peak.
A range, and a way to check it
Every forecast should carry a confidence range, not a single point, so you can plan cover against the risk rather than a false certainty. And it should be backtested on held-out weeks, so accuracy is measured and reported per channel against a simple baseline, not asserted. If a vendor will not show you the error, treat the forecast as a guess.
This guide covers the method and what to ask for, so you can tell a forecast that will hold up from a number that just sounds confident.