Fast Fashion Pre-Season Forecasting

The Problem

One of Europe’s largest fast-fashion retailers, trading from more than 1,000 stores in over 50 countries and designing close to 90,000 new products a year, places the bulk of its seasonal buy roughly 10 months before the season opens. Every miss in purchase quantity shows up twice: as markdowns on the lines that over-ran, and as lost sales on the lines that sold out early.

Brand-new style-colors were the largest part of the problem. With no selling history, the existing process relied on picking a lookalike product and copying its sales curve. That carries the lookalike’s quirks into the new line and keeps distorting it for months after launch.

Results

Dataviva Demand Forecasting produced the forecast for a full selling season 10 months before it opened, using only history from before the season, with no sight of actual sales. The forecast was then measured against the retailer’s actual sales across the full February–December period.

New style-colors are forecast from product and location attributes. Each one joins a demand group of similar products, so there is no lookalike to pick. New lines tracked as closely as carry-over lines with years of trading history. Accuracy landed near the best end of the fashion industry’s range at the hardest level to forecast, style-color by store, and was much better at the level where purchases are made.

• Forecast error, style-color by store ~36%
• Industry benchmark, same level 35–50%
• Forecast error, week by category 16.8%
• Forecast horizon 10 months