Advancing product innovativeness under market uncertainty through machine learning-driven models

Product innovativeness, the ability to introduce novelty across a firm’s offerings, is crucial for value creation, but its relationship with market uncertainty remains contested. Prior studies rely on statistical methods that assume linearity and fixed model structures, so whether non-linear interactions among antecedents matter has not been examined. Machine learning (ML), especially neural networks, can uncover such patterns, yet its use in this area remains limited by data and interpretability challenges. This study re-analyses survey data from 313 technology-intensive manufacturers using neural networks tuned with a Convolutional Gated Recurrent Neural Network (CGRNN)-based optimisation method, and interprets them with three complementary feature-importance techniques. The results suggest that internal capabilities, particularly interfunctional coordination and the assimilation and exploitation of knowledge, are more closely associated with product innovativeness than market uncertainty itself. A paired comparison on identical data partitions shows that regularised linear regression predicts both outcomes at least as accurately as the networks, while the leading predictors recur across both model classes, suggesting that differences from earlier findings reflect how capabilities are specified rather than non-linearity. Given the modest sample, these findings are presented as propositions for confirmatory testing. The study proposes tentative priorities for further managerial investigation and a framework that can be adapted to other industries and contexts.