Balancing fairness and efficiency in order allocation for quick-commerce applications

Purpose - This study proposes an approach for deciding the optimal allocation of orders in context of last-mile delivery services. We attempt to model and analyze the inherent trade-off between operational efficiency (driving the least to deliver orders while adhering to delivery time guarantees) and fairness (equitable distribution of earning opportunities). Design/Methodology/Approach - The study proposes to convert the dual objective of delivery efficiency and rider fairness into a single formulation. This conversion is done by weighing the two objectives using a tunable parameter λ, that balances fairness and efficiency. The resulting model is a minimum-cost assignment problem, incorporating conventional VRP constraints such as multi-stop pickups, order batching, and rider-specific attributes (e.g., capacity, idle time). Given the computational complexity of the theoretical model, our study proposes a two-stage heuristic which breaks the problem into two sub-problems namely, creation of efficient routes and finding a minimum cost assignment. Finally, the approach is tested on a real-world dataset (NYC yellow taxi data). Findings - The simulations show that there is an average increase of 4-5 percent in delivery time while leading to a 33 percent decrease in inequality as the emphasis on fairness (measured using Gini index) is increased. The use of the approach allows a graded movement across all values of fairness as it not only prioritized even distribution of earnings but also incentivized low performing groups. Research Limitations - Our approach, first of its kind attempts to model the two competing objectives. We demonstrate our approach on synthetic as well as a real world dataset. However, the insights derived from the dataset, might not be immediately applicable to all e-commerce platforms world-wide. Region-specific data would be required for such a conclusion. Practical Implications - We find that prioritizing fairness (measured using Gini coefficient), leads to decrease in delivery efficiency, and increase in average delivery time. Platforms can use our model to incrementally modify the policy, allowing them to manage the trade-off between rider welfare (fairness) and customer satisfaction (efficiency). Originality/Value - The novelty of this work lies in its integrated two-stage model, which combines a VRP route generator with a fairness-aware MILP assignment while allowing a graded movement between the two extremes. This study provides a theoretical approach for balancing the interests of the two major players in an e-commerce setting.