Real-time Dynamic Optimization for Omnichannel Retailers

这个项目寻求使用数学模型和分析来确定在全渠道零售中出现的动态优化问题的接近最优的算法。中心主题是帮助全渠道零售商了解当订单履行(即向客户实际交付产品)成本高昂时应如何做出各种决策(例如定价、分类和库存)。全渠道零售是指一种完全集成的方式，为客户提供跨多个渠道的统一购物体验。它正在迅速崛起，成为零售业务的默认模式。为了实现具有多个接触点的无缝购物体验，全渠道零售商承受着订单履行活动带来的巨额运营成本。此外，相关决策需要共同优化，以实现利润最大化。这一领域日益增长的实际重要性和学术文献的稀缺性提供了令人振奋的研究机会，这些机会将在广泛意义上产生直接影响。在最近全渠道零售创新的激励下，我们将调查以下三个密切相关的研究问题，这些问题在文献中尚未得到解决：1.延迟订单履行的价值是什么？2.零售商如何联合优化交付选项产品和订单履行决策？3.履行过程中的不灵活性如何影响零售商的定价和库存决策？我们的研究将提供使理论与商业实践保持一致的解决方案，并突出联合动态优化的好处和可行性。从理论上讲，我们的问题被建模为随机控制问题，很难以最优方式解决。我们计划提出易于实时实现并具有近乎最佳性能保证的算法。由于这些问题涉及以前从未研究过的新动力学，该项目广泛地贡献了随机控制和在线算法文献。五名HQP将获得运营和数据分析领域非常受欢迎的技能，这对于解决相关的现实问题以及在学术或行业职业生涯中取得成功至关重要。从实用的角度来看，这个项目及时解决了全渠道零售商面临的共同挑战，他们依赖复杂的物流网络来服务于有高服务要求的客户。我们的算法将使用真实数据进行测试，并有可能被零售业的行业合作伙伴(如亚马逊、甲骨文实验室和Insta)实施。在Z世代崛起和新冠肺炎疫情的影响下，消费者的购物习惯发生了根本性的变化。研究结果将有利于更广泛的零售业接受这些变化。这对加拿大零售业的潜在经济影响将是巨大的，因为即使收入略有改善1%，也可能转化为超过5B加元的额外销售额。

This project seeks to use mathematical models and analysis to identify near-optimal algorithms for dynamic optimization problems arising in omnichannel retail. The central theme is to help omnichannel retailers understand how various decisions (e.g. pricing, assortment and inventory) should be made when order fulfillment (i.e. physical delivery of products to customers) is costly. Omnichannel retail refers to a fully integrated approach where customers are provided with a unified shopping experience across multiple channels. It is fast emerging as the default mode for retail operations. To enable a seamless shopping experience with multiple touchpoints, omnichannel retailers endure significant operational costs from order fulfillment activities. Moreover, correlated decisions need to be optimized jointly to maximize profits. The increasing practical importance of this area and the sparsity of academic literature present exciting research opportunities that will be directly impactful in a broad sense. Motivated by recent innovations in omnichannel retail, we will investigate the following three closely related research questions that have not been addressed in the literature: 1. What is the value of delayed order fulfillment? 2. How can retailers jointly optimize delivery option offerings and order fulfillment decisions? 3. How do inflexibilities in the fulfillment process affect retailer's pricing and inventory decisions? Our research will provide solutions that align theory with business practices, and highlight the benefit and feasibility of joint dynamic optimization. From a theoretical perspective, our problems are modelled as stochastic control problems that are intractable to solve optimally. We plan to propose algorithms that are easily implementable in real-time and have near-optimal performance guarantees. As the problems involve novel dynamics that have not been studied before, this project broadly contributes to stochastic control and online algorithm literature. Five HQPs will gain a skillset that is highly in demand in Operations and Data Analytics that is essential to address relevant real-world problems and for success in an academic or industry career. From a practical perspective, this project timely addresses common challenges faced by omnichannel retailers, who rely on complicated logistic networks to serve customers with high service requirements. Our algorithms will be tested using real data and have the potential of being implemented by industry partners in the retail sector (e.g. Amazon, Oracle Labs, and Instacart). Driven by the rise of Generation Z and the impact of the COVID-19 pandemic, consumer shopping habits have changed drastically. The research outcomes will benefit the broader retail industry in embracing these changes. The potential economic impact on Canadian retail sectors will be significant, since even a moderate improvement of 1% in revenue could translate to over CAD 5B of additional sales.