Revenue Management and Policy Design in the Presence of Customer Multi-Item Shopping Behavior

收益管理（RM）已经在许多数十亿美元的行业取得了重大成功，例如航空、酒店和零售行业。成功RM的关键要素是需求建模。将客户选择集成到需求建模中的重要性已经得到了很好的证明。然而，大多数现有的客户选择模型都假设每个客户最多购买一种产品。这个假设在很大程度上偏离了现实。例如，客户经常在同一产品类别中购买多种商品，例如软饮料和谷物。网上购物者通常会购买不同类别的多种商品，事实上，他们这样做往往是为了获得促销活动或免费送货的资格。忽略这种行为可能会导致预测错误，使解决操作问题和推广策略的效果降低。尽管多物品购买行为很重要，但迄今为止，RM对多物品购买行为建模的研究还很少。提出的研究计划旨在弥合这一差距，并推进客户多项目购买行为的建模在RM文献。我们将“多项目”定义为多个项目单位，其中一些可以是相同的，在同一产品类别内或跨不同类别。这些项目可以是互补的、可替代的或相互独立的。也就是说，就像在现实环境中一样，客户选择他们自己的一束或一篮子产品。因此，我们的研究与捆绑文献（客户选择预先确定的配套产品包装）和购物篮购物文献（客户从多个类别中选择不同的产品）有很大的不同，而且比捆绑文献更普遍。本研究项目将积极利用选择建模、非线性优化、博弈论方法和新兴大数据工具等方法来实现以下目标：(1)利用分析和数据驱动的方法，推进分析模型，以捕捉客户的多项目购买行为，并寻求高影响RM问题的解决方案，如分类和价格优化；(2)设计包含这些客户选择行为的最优政策，以帮助公司做出更有效和有利可图的运营决策。RM分析的进步有可能产生重大的经济影响，因为即使是1%的收入适度改善，也可以转化为加拿大零售业超过50亿加元的额外销售额。因此，本研究的结果对RM研究者和行业从业者都很重要。在这个项目中，HQP将获得管理科学和高级数据分析的知识，以及将这些知识应用于现实问题的技能，并为在对这些技能有很高需求的学术或行业职业生涯中取得成功做好准备。

Revenue management (RM) has achieved major success in many multi-billion sectors, such as airline, hotel, and retail industries. The crucial element in successful RM is demand modeling. The importance of integrating customer choice into demand modeling has been well documented. However, most of the extant customer choice models assume that each customer purchases at most one product. This assumption largely deviates from reality. For example, customers often buy multiple items in the same product category, such as soft drinks and cereals. Online shoppers often purchase multiple items across different product categories, and in fact are often incentivized to do so to qualify for a promotional campaign or free shipping. Ignoring such behavior may result in prediction errors, rendering solutions to the operational problems and promotional policies less effective. To date, little work has been done in RM to model the multi-item purchase behavior despite its importance. The proposed research program seeks to bridge this gap and advance the modeling of the customer multi-item purchase behavior in the RM literature. We define "multi-item" as multiple units of items, some of which can be the same, within the same product category or across different categories. The items can be complementary, substitutable or independent of each other. That is, just as in real-world circumstances, customer choose their own bundle or basket of products. Our research is thus vastly different from and much more general than the bundling literature (where a customer chooses a pre-determined package of complementary products) and the basket shopping literature (where a customer picks a different product from multiple categories). This research program will actively leverage methodologies such as choice modeling, non-linear optimizations, game theoretic approaches, and emerging Big Data tools to address the following objectives: (1)Advance analytical models to capture customer multi-item purchase behaviors, as well as seek solutions to high-impact RM problems, such as assortment and price optimizations, using both analytical and data-driven approaches; and (2)Design optimal policies that incorporate such customer choice behaviors to help companies make more efficient and profitable operational decisions. Advances in RM analytics have the potential for significant economic impact, as even a moderate improvement of 1% in revenue as a result of advanced RM analytics could translate to additional sales of over CAD $5B in the Canadian retail industry. Therefore, the outcomes of this research are important for both RM researchers and industry practitioners. HQP in this project will gain knowledge of management science and advanced data analytics and skills in applying the knowledge to real-world problems, and get prepared for success in an academic or industry career where such skills are in high demand.