Plan an ML product to improve store-level inventory replenishment at Walmart

At Walmart, the quantity of an item a store currently has on hand determines when, and how much, new stock is ordered for that store. The company wants to use machine learning to optimize this replenishment process. As the product manager, how would you proceed? ```hint Start from the decision Before choosing any model, name the ordering decision the model would change, who makes it today, and what it costs when that decision goes wrong in each direction. ``` ### Clarifying Questions - What does optimizing mean here: fewer empty shelves, less excess stock, less waste of perishables, lower supply-chain cost, or a balance of these? - Which items and stores are in scope: one category such as fresh groceries or everything, and all stores or a pilot region? - How accurate is the recorded on-hand quantity today, and how is it kept up to date? - Should the system place orders automatically, or recommend quantities that replenishment or store staff approve? ### What a Strong Answer Covers - Framing: the replenishment decision, the people involved, and an objective that weighs stockouts against excess inventory. - Where machine learning helps and where it does not, including the process's dependence on accurate in-store counts. - The data required and its quality problems. - A rollout plan: offline testing, a store-level pilot with comparison stores, success metrics and guardrails. - Operational constraints and risks: shelf space, delivery schedules, staff trust and overrides, and unusual events. ### Follow-up Questions - The pilot reduces empty shelves but raises waste in fresh produce. What do you do? - Store staff override most of the system's orders. How do you find out why, and what do you change? - How should the system behave during a demand spike it has never seen, such as a regional storm warning?