Explainable AI based interventions for pre-season decision making in fashion retail

Abstract

New product demand forecasts are essential for fashion retail because of high volume of new products introduced every season. Explainability and interpretability of the forecasts are important for better adoption by designers and buyers. We built an explainable new product demand forecasting system which can explain a forecast by attributing relative credit or blame to different features of the product. These forecasts were further made interactive through what-if analysis for each feature of the product and their counterfactual explanations. Deployment of this system for a leading fashion retailer has lead to higher efficiency and rapid turn-around time for fashion designers and buyers.