Retailers are spending heavily on AI right now — demand forecasting engines, personalization models, AI-powered merchandising, copilots for planners and store associates. Boards want to see AI in the roadmap, and for good reason: the tools genuinely help retailers make faster, sharper calls.

But ask a store operations leader a simpler question — do you know, right now, exactly what's in this store versus what the system says is in this store — and the confidence usually drops. That gap between recorded inventory and physical reality is exactly where sophisticated retail AI quietly stops being useful. An algorithm can only optimize the data it's given, and in most retail environments, that data is stale before it even reaches the model.

This is the piece of supply chain AI in retail that doesn't get enough attention. Not the model. The ground truth underneath it.

‍

What Supply Chain AI Actually Means in Retail

‍

Supply chain AI isn't one product — it's a stack. A sensing layer that generates real-world signal, a data layer that aggregates and structures it, and an intelligence layer that reasons over it to forecast, flag, or act. Most retail AI investment so far has gone into that top layer: better models, better forecasting, better recommendations.

Retail adds a wrinkle that other industries don't deal with at the same intensity: omnichannel. A single SKU has to serve the sales floor, the e-commerce site, buy-online-pickup-in-store, and ship-from-store — all off one shared inventory record. When that record is even a little bit wrong, it doesn't just cause one bad outcome. It cascades across every channel drawing from it at once.

That's why retail inventory accuracy isn't a back-office metric anymore. It's the input every other AI initiative in the building depends on.

‍

Where Retail's Data Breaks Down

‍

Most retail systems were built around periodic, manual events — a cycle count, a receiving scan, a shipment check. Between those events, the system is essentially guessing. A few places where that guess goes wrong most often:

Omnichannel inventory tracking gaps. A traditional cycle count only reflects the moment it was taken. By the next morning, the sales floor has shifted, and "phantom inventory" — stock the system believes exists but that isn't actually there, or isn't where it's supposed to be — quietly breaks BOPIS and ship-from-store promises. Retailers relying on manual counts typically operate around 60–80% inventory accuracy, which is a rough foundation to build any automated decision-making on top of.

Receiving delays. Manually counting and reconciling inbound goods against purchase orders and ASNs commonly takes 24–48 hours before inventory is usable in the system. That's a day or two where product is physically on-site but functionally invisible to every downstream system, including the AI trying to plan around it.

Mis-shipments. Wrong pallets, incomplete loads, or goods routed to the wrong store are typically caught after the truck has already left — which turns a fixable dock-side error into chargebacks, emergency re-picks, and frustrated store teams.

Shrink. Loss that happens in the gaps between scan events — in transit, in the backroom, on the floor — is exactly the kind of anomaly retail AI is supposed to catch. It can't catch what it never sensed in the first place.

None of these are AI problems. They're sensing problems that AI inherits.

‍

Physical AI: The Sensing Layer Retail AI Needs

‍

This is the gap that Physical AI is built to close. Instead of waiting for a scan or a manual count, battery-free IoT Pixels attach to cases, pallets, or products and continuously broadcast signal — location, movement, condition — using ambient IoT: existing Bluetooth infrastructure that's already in stores, backrooms, and distribution centers. No new scanning hardware, no batteries to replace, no human required to trigger a read.

Wiliot's Physical AI platform turns that constant stream of signal into structured, AI-ready data, synced directly into the ERP or WMS retailers already run. It's not a better dashboard. It's the continuous, item-level ground truth that every retail AI initiative — forecasting, replenishment, agentic workflows — is currently missing.

‍

What This Looks Like Across the Retail Supply Chain

‍

On Wiliot's retail applications page, this shows up as three connected capabilities, each targeting one of the blind spots above:

Inventory Intelligence. Replaces manual cycle counts with continuous, scan-free tracking across backrooms, coolers, and storage zones, typically pushing accuracy to 99%+ and cutting excess safety stock by roughly 20%, since planners no longer have to buffer against data they don't trust.

Automated Receiving. Detects inbound goods automatically as they enter the dock and matches them against open POs and ASNs in real time, taking dock-to-stock time from 24–48 hours down to roughly 2–6 hours, with 99%+ receipt accuracy and 30–50% lower receiving labor.

Automated Shipment Verification. Continuously monitors pallets staged at the dock and flags mismatches before the truck pulls away, cutting mis-shipments by up to 90% and typically paying for itself in under a year.

Together, these don't just fix three operational headaches. They give every retail AI system downstream — from replenishment engines to agentic store operations — a data foundation that actually reflects the store floor.

‍

Proof at Scale: What Walmart's Deployment Shows

‍

The clearest evidence that this works at retail scale is Wiliot's collaboration with Walmart. What launched in October 2025 across roughly 500 stores is expanding through 2026 to more than 4,600 Walmart Supercenters and Neighborhood Markets and over 40 distribution centers, with a target of 90 million IoT Pixels deployed by the end of the year.

The retailers moving fastest on AI aren't skipping the sensing layer to get there faster. They're building it first.

‍

The Next Conversation in Retail AI

‍

Retail's AI conversation is shifting from "which model should we use" to "can these systems act autonomously on what's actually happening in the store." That shift — toward agentic replenishment, autonomous fulfillment routing, self-correcting inventory — only works if the data feeding it is trustworthy in real time, not reconciled after the fact.

Supply chain AI in retail was never really a modeling problem. It's a visibility problem, and it starts on the floor, not in the algorithm.

‍

Ready to see what continuous, item-level visibility could do for your retail operations? Get in touch with Wiliot.