AI Cuts New-Product Overordering for Beauty Retailers

Beauty and retail merchandising teams are testing AI comparable-product models to avoid excess stock on new launches.

Updated on Nov. 4, 2026 in Retail

Bold vector editorial illustration of stacked cardboard cartons on a shelving run, one open carton showing cosmetic bottles, in coral, teal, and mustard.
Beauty retailers are testing AI models that compare new products to existing ones, after one retailer's new-item demand forecasts came in nearly double actual sales. AI Illustration. Upload story photo >

Beauty retailers face a persistent forecasting problem with new products, and one unnamed retailer's demand forecasts for new items were almost twice actual demand, leaving it with slow-selling and obsolete stock. Retailers are now turning to AI to identify comparable products and model demand curves before committing inventory.

Why it matters

Overforecasting new products ties up capital in excess inventory that eventually has to be marked down or written off. Because beauty retailers introduce new items on a regular cycle, the error compounds across every launch, making demand accuracy a direct driver of margin.

Forecasts for one unnamed beauty retailer's new products were almost twice actual demand. A K-means store clustering project for fashion retailer Zadig & Voltaire improved sell-through by 10 points using 28 numerical features.

The players

Zadig & Voltaire

A French fashion retailer whose store clustering project with AI improved sell-through performance by 10 points.

The details

Retailers identify comparable existing products and analyse product type, market relevance, price tiers and seasonality to estimate how a new item will sell. The model then generates a demand curve for the new product, which informs how much to purchase, where to place it and what order quantity to commit to. Applied to store-level assortments, similar clustering techniques help match products to the locations where they are most likely to sell through.

Timeline

  1. 2026: Remarks on AI in retail forecasting were made at the Retail Asia Summit 2026 in Hong Kong.

Market Landscape

Zadig & Voltaire's store clustering project provides a documented retail precedent for applying K-means models to assortment decisions, with the reported 10-point sell-through gain as its benchmark. Beauty retailers, which launch new items regularly, are now applying comparable AI demand modelling to new-product forecasting.

Merchandising and buying teams should test whether AI comparable-product analysis can reduce overordering on new launches before expanding it across the range. Operators should track slow-moving and obsolete stock as a direct measure of forecast error.

The takeaway

Forecast error on new products is a measurable inventory cost, so operators should compare AI-based demand estimates against actual sell-through on recent launches. Calendar a review of new-product sell-through at the end of each launch cycle to judge whether the model is closing the gap.

Further reading

For broader context on how retailers are reshaping their operations, see the Retail coverage.