An Integrative Model to Predict Product Replacement Using Deep Learning on Longitudinal Data

Past research on product upgrades has focused either on understanding who and when will upgrade or on figuring out why consumers will upgrade, but seldom on all. It has also neglected the interplay between these matters with decision context and timing. This manuscript depicts a comprehensive approach where, for the first time, product characteristics, individual differences, process, and contextual variables are analyzed on a predictive model of real product upgrades, identified through the systematic collection of primary data from a panel of smartphone consumers. We tested one traditional linear logistic regression model and two types of non-linear, state-of-the-art machine-learning models (extreme gradient boosting and deep learning) to explain upgrading behavior. Results provide an integrative, yet parsimonious, product-upgrade model showing the importance of resources; news about the smartphone brand; sentimental value; predicted, current, and remembered enjoyment; update capacity; and how much the smartphone meets the user’s current needs as the most relevant variables to determine which consumers are more prone to upgrade their smartphones. Our findings advance upgrade decision theory by taking a holistic approach to the phenomenon and bridging different theoretical accounts of the replacement decision literature.