Optimizing Walmart’s Outbound Supply Chain from Strategy to Execution – A Grocery Case Study


On July 2, 1962, the first Walmart store opened in Rogers, Arkansas, with a strategy to build it on an unshakeable foundation: the lowest prices anytime, anywhere. In the 1980s, the first Walmart Supercenter opened, combining a supermarket with general merchandise; and later in the 1990s, the supercenter model that redefines convenience and one-stop shopping was rolled out nationwide in the U.S. Now in the 21st century, Walmart is expanding the ecosystem that supports customers, enhancing processes that enable them to shop wherever and however they choose, through an omnichannel retailing transformation. At each step of Walmart’s evolution, one of the critical foundations for success is the applications and systems to support supply chain management across the decision tiers from strategy to execution.

Currently for grocery products, customers can choose shopping in store, pickup at curbside, or delivery from store to home. The introduction of new shopping channels significantly changes the demand patterns faced by the 4,700+ brick-and-mortar stores and creates great challenges for its supply chain to meet the demand. Meanwhile, the recent advancements of warehouse automation technologies, especially in the cold-chain space, have unlocked huge potential to revolutionarily improve productivity. There is a more imperative need than ever before to strategically transform the supply chain network at larger scale in a shorter timeframe.

At the execution level, Walmart has a long history of adopting industry leading supply chain optimization technologies to help continuously drive down operational costs. It has reached a point where any incremental cost reduction on standalone systems becomes very challenging. For example, for dry grocery commodities, Walmart’s trucks were nearly fully utilized in terms of weight and space capacities in the year 2021, providing little room for further utilization increases. A breakthrough in optimization technologies is desired to drive continuous improvement.

Walmart identified opportunities in a virtuous circle between network strategy and execution-level optimization technologies. A supply chain network designed for higher efficiency could unlock greater potential of the optimization applications; on the other hand, faster optimization applications can enable more simulation runs, which allow more scenarios to be evaluated, and thus improve the chances for the recommendations to be adopted by executive decision makers. However, it’s not easy to achieve this virtuous cycle, primarily because of two reasons: 1) considering the most granular level of operating costs at strategic planning level makes the problem intractable and 2) optimization applications need to run much faster to evaluate more scenarios.

Compared to general merchandise, the complexities of outbound grocery product distribution, from network design to daily routing and loading execution, are significantly greater, which creates larger room for improvement on efficiency as well as customer and associate experience. Walmart developed models that are applicable to general merchandising products, but scenario building, recommendations, interpretations, and plan execution are generally kept separated primarily because of the critical difference in temperature requirements.

Walmart built and rolled out an outbound routing and loading planning and optimization system named “Load Planner” to solve truck routing and loading problems in one shot. At its core, Walmart developed a metaheuristic-based framework integrating a suit of algorithms, including various neighborhood searches, heuristics, and mixed-integer programming (MIP) models. At each decision step, the best algorithm and parameter settings are selected based on learning from extensive experimentations with historical data. The framework provided the flexibility to add incremental features, as well as high computational efficiencies, which has been a general challenge when solving NP-hard problems. The optimization system was tested and validated in the past couple years, and is proven to outperform the existing application from both computation time and optimality perspectives.

To support the design of future networks, Walmart developed two MIP models. The first determines the optimal long-term end-state network in terms of distribution center (DC) locations, DC capacities, and DC-to-store alignments. The second multistage model creates the transformation roadmap with time steps, and provides concrete recommendations on how and when to initiate different transformative steps to achieve the end-state network. Both models face challenges to scale up when problem sizes are large. We applied heuristics to considerably reduce the problem size without significant loss of quality and enable running tens or even hundreds of scenarios with different input assumptions and risk levels to arrive at a well-informed set of decisions.

In FY23, with the full network rollout of Load Planner, Walmart was able to avoid 72 million pounds of CO2 and save $75 million by avoiding extra miles and truck loads needed. Given the long range of its planning horizon, the direct effects of this network strategy won’t materialize in the short term, so Walmart will measure its benefits primarily based on business adoption. In the same fiscal year, the network strategy and transformation roadmap were approved for implementation, which will require substantial investment planned over the course of the next several years.  So far, the program has received approval and funding for construction of three perishable DCs. The future is full of uncertainty and the transformation model enables sensitivity analysis across multiple variables, providing executive leadership with confidence that the strategy and roadmap are sound, regardless of unforeseen changes that may come.