Dynamic Stochastic Inventory Management in E-Grocery Retailing

TL;DR

This paper develops a stochastic, multi-period inventory management model for e-grocery retailing, using a sequential decision process and Monte Carlo methods to improve replenishment decisions under demand uncertainty.

Contribution

It introduces a novel stochastic lookahead policy that explicitly incorporates probabilistic forecasts and uncertainty propagation for inventory replenishment in e-grocery.

Findings

The proposed policy outperforms myopic and deterministic approaches in simulations.

Explicit uncertainty modeling improves service levels and reduces stock-outs.

Case study demonstrates practical applicability with real-world data.

Abstract

E-grocery retailing enables ordering products online to be delivered at a future time slot chosen by the customer. This emerging field of business provides retailers with large and comprehensive new data sets, yet creates several challenges for the inventory management process. For example, the risk of a single item's stock-out leading to a complete cancellation of the shopping process is higher in e-grocery than in traditional store retailing. As a consequence, retailers aim at very high service level targets to provide satisfactory customer service and to ensure long-term business growth. When determining replenishment order quantities, it is of crucial importance to precisely account for the full uncertainty in the inventory process. This requires predictive and prescriptive analytics to (1) estimate suitable underlying probability distributions to represent the uncertainty caused by non-stationary customer demand, shelf lives, and supply, and to (2) integrate those forecasts into a comprehensive multi-period optimisation framework. In this paper, we model this stochastic dynamic problem by a sequential decision process that allows us to avoid simplifying assumptions commonly made in the literature, such as the focus on a single demand period. As the resulting problem will typically be analytically intractable, we propose a stochastic lookahead policy incorporating Monte Carlo techniques to fully propagate the associated uncertainties in order to derive replenishment order quantities. This policy naturally integrates probabilistic forecasts and allows us to explicitly derive the value of accounting for probabilistic information compared to myopic or deterministic approaches in a simulation-based setting. In addition, we evaluate our policy in a case study based on real-world data where underlying probability distributions are estimated from historical data and explanatory variables.