Walmart Store and Online Stock: Applying AI to Cut Empty Shelves and Excess Inventory

You are a product manager at Walmart. How would you apply AI to inventory management, covering both stock in stores and stock that serves online orders, so that items run out less often while excess stock that must be marked down or written off also goes down? Describe the problems you would target, the AI capabilities you would build, who would use them and how, and how you would measure success. ```hint Two errors, one trade-off Running out and holding too much pull in opposite directions, so a better forecast alone does not settle how much to stock. Separate the decisions involved, such as predicting demand, knowing what is really on hand, and deciding how much to order and where to put it, then ask where AI changes each one. ``` ### Clarifying Questions - Which decisions does the product own: forecasting only, or also order quantities, allocation between distribution centers and stores, and markdowns? - Who are the primary users: replenishment planners and buyers, store associates, suppliers, or all of them? - Which categories matter most: fresh and perishable food, general merchandise, or seasonal items? - When the two goals conflict, for example for perishables, is avoiding stockouts or avoiding overstock the higher priority? ### What a Strong Answer Covers - A breakdown of why stockouts and overstock happen (forecast error, inaccurate on-hand counts, slow replenishment, poor allocation across stores and online) and which cause the plan targets first - AI capabilities matched to specific decisions and users, including how planners and associates act on recommendations and when they can override them - Data realities: demand hidden by past stockouts, promotions, weather and local events, new items without history, and store stock shared with online orders - Metrics that capture both errors together (on-shelf availability and online fulfillment alongside inventory levels, markdowns and waste), with guardrails - A rollout and testing plan at store or region level, and the risks of automating orders ### Follow-up Questions - The model's forecasts are more accurate, but store managers keep overriding its orders. What do you do? - How would you prove the program reduced stockouts when individual shoppers cannot be randomized? - How would the system handle a sudden demand spike, such as an approaching storm, that the model has rarely seen?