A Data-Driven and Real-time Approach to Personalized Bundle Recommendation and Pricing; from Theory to Practice

一个好的定价策略在任何零售商的业务流程中都起着至关重要的作用。这在很大程度上是因为定价方法对客户和竞争对手都非常明显，因此对利润有直接和戏剧性的影响。在过去的十年里，在线零售商品市场发展迅速。开发复杂的个性化产品和价格推荐系统可以为任何互联网零售商提供必要的竞争优势，带来数十亿美元的利润差异。新零售模式的挑战是设计捆绑和个性化定价模型，能够从电子零售商可以访问的新客户数据流中提取价值。这项研究的目标将是开发一种个性化的捆绑和定价模式，向消费者推荐一系列相关产品，但在他们的在线会话期间以激励价格(低于单个项目的价格总和)。该模型的创新之处将包括同时在捆绑和定价方面平衡几个关键因素，同时开发一种新的个性化需求评估方法。这项工作的重点包括四个考虑：(I)开发有意义的个性化模型(在产品捆绑方面，但也在这些产品的定价方面)，(Ii)开发库存平衡方法，该方法将在未来降价造成的损失和同时面临未来可能的库存缺货之间进行权衡，(Iii)开发有趣的结构结果和关于个性化如何在定价策略方面与捆绑销售进行权衡的见解。最后，也是非常重要的一点，(Iv)开发能够影响实践的有效的解决方法(理论上，但在计算上，后者使用实际的行业数据)。我们的目标是在建模中结合这些不同的目标：个性化、库存平衡、定价洞察力和效率/实用性。这项研究将设计创新的模型，首先从理论和应用的角度对它们进行分析，然后探索它们之间的关系，并首先用实际数据进行检验。此外，本研究将构建一个从机器学习、整数优化、随机稳健优化到关键的个性化捆绑定价问题的集成框架、模型和求解技术，以帮助定价领域的学者和实践者。

A good pricing strategy plays a crucial role in any retailer's business process. This is largely due to the fact that pricing approaches are highly visible to both customers and competitors, and thus have an immediate and dramatic impact on the bottom line. The online market for retail goods has grown enormously over the past decade. The development of a sophisticated personalized product and price recommendation system can provide the necessary competitive edge for any Internet retailer, making the difference on the order of billions in profits. The challenge of the new retailing paradigm is to design models for bundling and personalized pricing that are able to extract value from the new streams of customer data that an e-tailer has access to. The goal of this research will be the development of a personalized bundling and pricing model that recommends a bundle of related products to a consumer, but at an incentivized price (lower than the sum of the prices of the individual items) during their online session. The innovation in this model will consist of balancing several key factors in terms of bundle and pricing offerings simultaneously, while developing a new approach to personalized demand estimation. The focal points of the work consist of four considerations: (i) developing meaningful personalization models (in terms of bundling of products but also in terms of pricing these products), (ii) developing inventory balancing methods that will balance the tradeoff between losing money due to future markdowns and at the same time facing potentially future inventory stock-outs, (iii) developing interesting structural results and insights on how personalization can tradeoff with bundling in terms of the pricing strategies. Finally, and very importantly, (iv) developing efficient solution approaches (theoretically but also computationally, the latter using actual industry data) that could impact practice. The goal will be to combine these distinct goals in the modeling: personalization, inventory balancing, pricing insights and efficiency/practicality. This research will design innovative models, analyzing them first from a theoretical as well as an applied standpoint but also in exploring the relationships between them and testing them first with real data. Furthermore, this research will build an integrated framework, models and solution techniques from machine learning, integer optimization, stochastic and robust optimization to key personalized bundle pricing problems that are accessible to all, in order to help academics and practitioners in the area of pricing.