Data-driven attribute selection for hardware technology products: A multi-criteria framework

Abstract

This paper outlines a multiple-criteria approach for supporting manufacturers in making decisions about tech products' technical, aesthetic and price characteristics. The authors propose a predictive modelling approach that shortlists efficient product designs based on their expected profit margin, consumer rating and demand. The method involves collecting SKU (stock keeping unit)-level data on product features from an online marketplace and estimating regression models. These models include a hedonic pricing model, a demand model and a satisfaction model to identify the factors that drive sales, prices and satisfaction. Analysing the model coefficients and their significance allows for identifying cost-efficient product features that positively impact sales and satisfaction. The models also enable predicting the outcomes for various new specifications making it possible to shortlist Pareto-efficient product designs. The approach uses publicly available data and allows for frequent updates, although it has some limitations, such as omitted variable bias and the use of a demand proxy. The authors suggest ways to extend the framework to account for uncertainty in predictions and include more outcomes of interest.