Using consumer reviews for demand planning: Case of configurable products

Short-term demand forecasting plays a pivotal role in a company’s success as it affects several decisions in the supply chain from procurement to stockpiling, and purchasing to distribution. The related literature is very rich in terms of demand estimation models; however, as in the case of configurable products, they usually suffer from the curse of dimensionality when there is an abundance of parameters to estimate. Configurable products allow consumers to create the product variant they have in mind by choosing predefined product attributes from a drop-down list. Basically, a configuration refers to a combination of (binary) product attributes offered by the manufacturer. A large number of attributes can easily complicate the demand estimation process. Besides, if these attributes have too many levels, then such a product, in theory, can have millions of different configurations. This would, in turn, lead to biased coefficient estimates since in this case, experiencing shortage of customers in the marketplace is inevitable. In this study, we utilize various text mining techniques on online consumer reviews of a configurable product to ease its demand estimation process. We identify the most important attributes considering their frequency of mentioning in the reviews and the way they were mentioned (positive, negative, neutral).