UC Berkeley

Internet platforms have been growing at a fast pace since the early 2000s and offer now a variety of products and services to their customers through digital devices. E- retailing platforms such as Ebay, Amazon.com, Alibaba and JD.com are changing the way people shop around the planet, while e-hailing platforms such as Lyft, Uber and Didi have reshaped how we commute. Moving the interaction with their customers online allowed internet platforms to gather valuable information for decision making. This data provides internet platforms with the necessary tools to make informed decisions on their key business strategies such as adapting their offer to the demand of regional markets.

An assortment of products is a subset of the products the internet platform offers to its customers in a regional market. To optimize their business metrics, e-retail platform design assortments of products to be held in warehouses while e-hailing platforms decide on which products to offer in what cities. The value of such an assortment may come from the synergies between different products. For instance, an e-retail platform holding products often purchased together in the same warehouse reduces its shipping costs and an e-hailing platform offering a premium service and a discounted one can cover several customer segments.

In this essay, we focus on selecting and studying assortments of products for e-retail and e-hailing platforms. We build data driven algorithms to guide such decisions. We provide thorough numerical analyses build on real world data sets to prove the performance of these methods.

E-retail platforms

E-retail platforms face many logistics challenges such as designing their delivery net- work, selecting assortments of products to be held in these warehouses and optimizing the replenishment policy. These problems are tied together but solved independently due to the intractability of the general problem. In this essay we focus on building the assortment of products for a given delivery network assuming the replenishment policy will minimize stock-outs of the assorted products.

Orders placed on an e-retail platform website may contain one or several products. If such an order can not be entirely fulfilled from a single warehouse, it is said to be split. Split orders typically result in a slower shipping speed and greater fulfillment costs. In Chapters 1 and 2, we develop algorithms to build assortments of products that minimize order split subject to capacity constraints. Acyclic delivery networks are studied in Chapter 1 and we extend these results to cyclic delivery networks in Chapter 2.

E-hailing platforms

E-hailing platforms cover multiple customer segments by offering different products such as X, Pool, Select, etc. for Uber; Lyft, Shared, Lux, etc. for Lyft; and Taxi, Express, Premier, etc. for DiDi. At a high level, these products can be partitioned between classic private rides and shared rides. Classic is a service that provides on demand transportation by dispatching drivers to passengers when a request is made. Shared is a service that matches riders travelling in the same direction with an available shared car. Classic is typically more expensive and Shared is often slower as sharing a vehicle may induce additional travel time. To keep Shared travel time reasonable, e-hailing companies set a detour constraint to place an upper bound on how much detour a rider can experience.

When two riders travelling in the same direction are matched, some supply cost is saved as a single car can serve both requests. The likelihood of such event increases with the demand density. Although a Shared request creates less revenue as the service is sold at a discount, a high density of demand coupled with an efficient matching algorithm, can result in a profitable market for the product. There is an economy of scale. To increase demand density, an e-hailing company can either offer greater discount thus decreasing the revenue per Shared riders or tighten the detour constraint, further constraining the matching algorithm and limiting the amount of feasible matches.

These two types of products are not offered together in every region where the e- hailing platform operates. Although offering Shared captures demand from other transportation solutions, it can also cannibalize demand from Classic. In Chapter 3, we study where and when it is profitable for an e-hailing company to offer both products.