Data-driven product design and assortment optimization

Our study focuses on how real-world data can inform and enhance firms' decisions around product design and assortment, which is critical in logistics, automotive, fast fashion and other industries. This article presents a data-driven analytics study on the challenges of new product design and product assortment. We first implement predictive analytics, utilizing a Multinomial Logit (MNL) model to estimate consumer preferences for both existing and newly designed products. Subsequently, we proceed with assortment optimization, including a deterministic model and a robust model. By applying our data-driven method in the case study based on the historical data of a fast fashion e-retailer, we find that the robust assortment model balances revenue and stability, while performing significantly better in the worst-case than the deterministic assortment model. This demonstrates that the robust assortment model, which accounts for parameter uncertainty, may be more suitable for real-world applications. Furthermore, the numerical results indicate that our data-driven new product design and robust assortment approaches can help the firm achieve a 31% expected revenue improvement. Interestingly, our robust assortment methods based on the MNL model outperform machine learning based assortment methods, despite the latter's more accurate predictive abilities regarding consumer purchasing patterns. These results indicate that accurate predictions of consumer purchasing patterns alone are not sufficient to guarantee good assortment decisions. Firms are advised to adopt the simpler and more comprehensible MNL model as their predictive tool when making assortment decisions.