Judgmental Demand Forecasting for Online Sales of Product Collections: Segmentation by Type or Age?

Abstract

We develop a forecasting method for the manufacturer and online seller of a product collection that changes periodically and radically. The firm, an industry leader in technology and quality, has experienced double-digit annual sales growth. In seeking to minimize supply-demand mismatch costs and satisfy borrowing constraints, management adopted a conservative consensus sales forecast — driven “top down” by the owner’s own estimates — while assuming a minimum sales growth rate across product groups. However, that approach led to low profitability. The firm’s response was to adopt a new forecast technique based mainly on a “bottom up” team estimate of mean sales per stock keeping unit, a reliable measure of the team’s systematic bias across selling seasons, and on two alternative product segmentation approaches. Demand distributions forecast using the new method were extremely accurate (even for new products) and thus enabled profit-maximizing supply order decisions. Company data confirm the validity of this approach: aggregated forecast accuracy increased from 74% to 97%, gross profits increased by 11%, and forgone profits declined nearly by half. The sales of Canyon Bicycles increased by more than 10% after implementation of the new technique. Another benefit was the reduced financing risk of outsourcing production of its specialized frames, which were ordered from overseas and therefore well ahead of the selling season.

Keywords: judgmental demand forecasting, demand uncertainty, de-biasing demand forecasts, forecast accuracy

JEL Classification: C12, C13, C44, C53, D24, D81, E23, E27, M11, M31