Walmart

Operating over 4,700 U.S. stores and one of the world's largest grocery supply chains, Walmart faced a structural inability to prevent inventory shortfalls before they materialized. Traditional replenishment software operated reactively — flagging out-of-stock conditions only after shelves had already emptied, by which point lost sales and customer trust were unavoidable. The perishables category compounded the challenge: excess stock resulted in direct write-off costs rather than markdowns. Without real-time root-cause visibility across a network spanning tens of thousands of SKUs per store, demand signals from local weather events, social trends, or logistical disruptions went undetected until the damage was done.

Walmart developed 'Wally,' a proprietary Merchant AI Agent built on machine learning and predictive analytics that shifts replenishment from reactive flagging to autonomous intervention. Rather than waiting for stockout triggers, Wally continuously ingests signals — local weather forecasts, social media demand trends, and upstream logistical constraints — and autonomously reroutes stock before gaps occur. This 'Self-Healing Inventory System' was paired with 'Scintilla In-Store,' a platform launched in early 2026 that surfaces live, granular replenishment data for field representatives, closing the loop between algorithmic decisions and store-level execution. The backend is supported by an automated distribution network — built using Microsoft's Azure OpenAI Service for internal agent orchestration — that now services 60% of all U.S. stores, with Ambient IoT tracking via a Wiliot partnership on track to monitor 90 million pallets in real-time by end of 2026.

Walmart's 2025 rollout of the self-healing inventory system delivered measurable outcomes across both waste reduction and top-line efficiency:

• 20–25% reduction in out-of-stock rates, described by analysts as unprecedented for a retailer at this scale
• $55M+ in waste savings within the perishables category during the first year of deployment
• 5% sales growth achieved while holding inventory growth to just 2.6% — a gap that signals material improvement in demand forecasting precision
• 95% of the U.S. population now reachable within sub-three-hour delivery windows, enabled by treating stores as fulfillment nodes

The sales-to-inventory growth spread is a particularly significant indicator: historically, retail sales growth has required proportional inventory build.

• Proactive rerouting outperforms reactive alerting — the measurable gains came from preventing stockouts, not from faster recovery after them.
• Closing the loop between AI decisions and frontline execution (via tools like Scintilla In-Store) is essential; algorithmic precision degrades if store teams lack visibility to act on it.
• Separating sales growth from inventory growth is the clearest ROI signal for supply chain AI — track both before and after deployment.
• IoT-level asset tracking (pallet-by-pallet visibility) amplifies predictive models by providing ground-truth data that ERP systems cannot capture.