This thesis examines problems faced in the distribution management of e-retailers, in different stages of the supply chain, while accounting for sources of uncertainty. The first problem studies distribution planning, under stochastic customer demand, in a transshipment network. To decide on a transportation schedule that minimizes transportation, inventory and outsourcing costs, the problem is formulated as a two-stage stochastic programming model with recourse. Computational experiments demonstrate the cost-effectiveness of distribution plans generated while considering uncertainty, and provide insights on conditions under which the proposed model achieves significant cost savings.

We then focus our attention on a later phase in the supply chain:​ last-mile same-day delivery. We specifically study crowdsourced delivery, a new delivery system where freelance drivers deliver packages to customers with their own cars. We provide a comprehensive review of this system in terms of academic literature and industry practice. We present a classification of industry platforms based on their matching mechanisms, target markets, and compensation schemes. We also identify new challenges that this delivery system brings about, and highlight open research questions. We then investigate two important research questions faced by crowdsourced delivery platforms.

The second problem in this thesis examines the question of balancing driver capacity and demand in crowdsourced delivery systems when there is randomness in supply and demand. We propose models and test the use of heatmaps as a balancing tool for directing drivers to regions with shortage, with an increased likelihood, but not a guarantee, of a revenue-producing order match. We develop an MDP model to sequentially select matching and heatmap decisions that maximize demand fulfillment. The model is solved using a stochastic look-ahead policy, based on approximate dynamic programming. Computational experiments on a real-world dataset demonstrate the value of heatmaps, and factors that impact the effectiveness of heatmaps in improving demand fulfillment.

The third problem studies the integration of driver welfare considerations within a platform’s dynamic matching decisions. This addresses the common criticism of the lack of protection for workers in the sharing economy, by proposing compensation guarantees to drivers, while maintaining the work hour flexibility of the sharing economy. We propose and model three types of compensation guarantees, either utilization-based or wage-based. We formulate an MDP model, then utilize value function approximation to efficiently solve the problem. Computational experiments are presented to assess the proposed solution approach and evaluate the impact of the different types of guarantees on both the platform and the drivers.