Predictive Analytics Improves Sales Forecasts for a Pop-up Retailer

Pop-up retailing involves short bursts of novel product offerings that are quickly withdrawn from the market. We describe an industry/university collaboration designed to improve sales forecasting for an organization that has adopted pop-up retailing as its exclusive business model. Early in the company’s history, the generation of sales forecasts relied heavily on expert opinion, a method that resulted in costly overstock of merchandise inventory. Accordingly, the organization developed a test market protocol in which small numbers of items are sold during a test market period to gauge future demand. Application of least absolute shrinkage and selection operator (lasso) and stochastic gradient boosting methodologies to the test market data along with other information about 508 products resulted in notable improvement in forecast accuracy over expert opinion. Specifically, the percentage of items that went unsold dropped by about 40% when using the predictive analytics tools in place of expert opinion. This striking result reflects, in part, the difficulty of using expert opinion to forecast sales of new, trendy merchandise in the absence of historical time-series sales information. By using the predictive analytics sales forecasts, the company now manufactures fewer products that never sell and, in general, manages its supply chain more effectively.