What-if decision support for product line extension using conditional deep generative models

Product line extension decisions require anticipating how consumer segments respond to new and unobserved product variants, which is a critical practical problem for marketing strategy and risk management. In this study, we propose a novel Conditional Tabular Variational Autoencoder (CTVAE) that leverages large-scale consumer purchase data to generate synthetic consumer attribute distributions conditioned on product design features such as container type, volume, flavor, and calorie content. By learning the conditional joint distribution of product attributes and consumer characteristics from tabular data, the CTVAE enables systematic what-if analysis for hypothetical product configurations without costly pretests. We demonstrate the effectiveness of our approach using a home-scan panel dataset of over 20,000 consumers and 746 soft drink products, showing that CTVAE outperforms existing tabular generative models in predictive accuracy. Beyond prediction, the model provides actionable insights for marketing decision-making, including evaluation of cannibalization risk and design of product attributes aligned with target consumer segments. These results highlight the potential of conditional deep generative modeling as a practical tool for applying machine learning to solve important real-world business problems.