Tractable Markdown Optimization for an E-tailer

该奖项的研究目标是设计操作模型，为电子商务降价定价提供决策支持。这些模型将包括在业务规则存在的情况下的战略客户行为，并将有可能在运营中应用。现有的战略客户MDO模型对客户行为做出了很强的假设，这些假设很难用实际收集的数据进行估计或验证。我们的研究重点是可以通过客户访问数据估计的定价模型。这些数据已经通过用户登录和cookie收集。这种信息对于实体店来说是不切实际的，而实体店是MDO问题的传统背景。研究计划是首先试图了解有限的战略（即返回，但短视）客户的影响，最优价格的电子零售商在一个单一的项目设置。接下来，我们将通过额外考虑业务规则来建立这种MDO模型的基础-这些是零售商对价格序列施加的实际重要的硬约束。最后，我们将把我们的模型推广到多个项目的情况。我们将首先从理论的角度分析这些模型，但也将利用它们之间的关系，并在实践中使用真实的数据进行测试。这项研究将采用来自各个领域的方法，其长期目标是加深我们对动态定价问题的理解，因为它们与零售业及其他行业有关。我们将为随机和鲁棒优化在关键定价问题中的应用开发一个集成的框架、模型和方法，如果成功的话，这项研究将填补现有研究中理论和实践之间的空白，从而改变电子商务的定价过程。这将使他们能够从更好地了解战略客户行为中受益。这是至关重要的，因为电子商务是零售业务的一个日益增长的部分。此外，我们相信这项研究的应用超出了定价领域。我们将分享我们的综合框架，模型和方法，以帮助学者和从业者。从教育的角度来看，该项目的成果将作为麻省理工学院教学模块的组成部分。其中包括PI已经分担责任的核心课程模块。这个项目非常适合指导本科生和研究生进行基于易处理实践的优化研究。

The research goal of this award is to design operational models to provide decision support for markdown pricing by e-tailers. The models will include strategic customer behavior in the presence of business rules, and will have the potential to be applied operationally. Existing models for MDO with strategic customers make strong assumptions about customer behavior that are difficult to estimate or validate with the data that can practically be collected. Our research focuses on pricing models that can be estimated with customer visit data. Such data is already being collected by e-tailers through user logins and cookies. Such information would not be practical to collect for brick-and-mortar stores, the traditional context for the MDO problem. The research plan is to first try to understand the impact of limited strategic (i.e. returning but myopic) customers on the optimal prices for an e-tailer in a single-item setting. Next, we will build upon the foundation such MDO models by additionally considering business rules - these are practically important hard constraints that retailers impose on the sequence of prices. Finally, we will generalize our models to the case of multiple items. We will analyze these models first from a theoretical standpoint but also will exploit the relationships between them and test them in practice using real data. This research will employ methodologies from a variety of fields with the long term goal to deepen our understanding on issues in dynamic pricing as they relate to the retail industry and beyond. We will develop an integrated framework, models and methods for the application of stochastic and robust optimization to key pricing problems.If successful this research will fill a gap between theory and practice in the existing research that will transform the pricing processes of e-tailers. It will empower them to benefit from a better understanding of strategic customer behavior. This is vital because e-commerce an increasing segment of retail business. Further, we believe that the applications of this research go beyond the field of pricing. We will share our integrated framework, models and methods, to help both academics and practitioners. From an educational perspective, the results of this project will serve as components in teaching modules at MIT. These include modules in core courses for which the PI already has shared responsibility. This project lends itself ideally to mentoring undergraduate and graduate students in research on tractable practice-based optimization.