Insights/Commerce and operations

Data-driven merchandising: better buying and a shorter product lifecycle.

A forecast error of five per cent can lock up cash when trends last only weeks. Linking sales, social and search data turns buying from instinct into a system.

Bespoke ResearchCommerce and operations5 min read

The problem

Many brands still buy on instinct and in disconnected spreadsheets, which leads to repeated overstock and stock-outs. A data-driven pipeline senses demand from point-of-sale, social trends and search intent, then automates the open-to-buy budget, size and colour allocation, and markdown timing.

Demand sensing and buy planning

  • Combine point-of-sale sell-out, social signals and Google Trends in a forecasting model such as Prophet or SARIMAX.
  • Update forecasts weekly, with a target forecast error below 12%.
  • Open-to-buy = planned sales + markdowns + end-of-month stock - beginning-of-month stock.
  • Buy six to eight weeks of cover for the main line and three to four weeks for capsules.

Allocation and lifecycle

  • Cluster stores (urban, suburban, flagship) and use ratio packs. This can reduce size mismatch by up to 15%.
  • Phase-in: if sell-through is below 25% in week one, boost ads and creator unboxing.
  • Mature: at 40 to 70% sell-through, replenish and hold price.
  • Phase-out: at 85% sell-through or more than 90 days of inventory, mark down in steps.
  • Set dynamic minimum and maximum stock with exponential smoothing, and use vendor-managed inventory with advance shipping notices.

How to report it

  • Forecast against actual, with error.
  • Sell-through by size, colour and week.
  • An ABC curve of revenue by SKU.
  • Days of inventory before and after vendor-managed inventory.

Merchandising in the data era should not run on intuition alone. Consistent models, KPIs and an open-to-buy process decide the right range and the right stock turn.

Notes

First published on bespokemarketing.asia and updated for this site. Figures are as cited in the original article; tool defaults are examples unless stated.