Google Cloud wants AI to audit store shelves

Empty store shelves cost U.S. retailers a staggering $82 billion in missed sales in 2021, according to NielsenIQ. To make matters worse, tracking inventory on the floor is so notoriously painful that most store managers essentially rely on educated guesses.

Google Cloud wants to change that. Just ahead of the National Retail Federation's NRF 2023 conference in New York, the cloud giant unveiled a suite of new retail AI tools, led by a shelf inventory AI that aims to fix computer vision's long-standing retail curse.

Eyes on the aisle

Powered by Vertex AI, Google's new shelf-checking tool is trained on an image database of more than a billion products. It combines a product recognition algorithm with a tag recognizer, built to identify items in real-world messy conditions like harsh lighting, odd angles, and seasonal packaging tweaks.

Crucially, it isn't married to specific hardware. Retailers can feed it footage from existing ceiling cameras, store-roaming robots, or just an associate walking down the aisle with a smartphone.

Supermarket chain Giant Eagle has already been putting it through the paces in an innovation lab. Graham Watkins, the company's EVP of supply chain transformation, noted that accuracy topped 90%. While promising, he cautioned that it is not quite ready for chain-wide rollout. Extreme camera angles still throw the model off, and hardware deployment costs remain steep.

History justifies the caution. Back in 2020, Walmart famously pulled the plug on roaming inventory robots in favor of simpler alternatives. Gartner retail analyst Robert Hetu estimates it could still take three to six years before computer-vision shelf tracking hits the mainstream.

Reinventing the digital storefront

Google is also revamping the digital side of retail with Browse AI, a tool available in 72 languages that automatically optimizes product sorting across e-commerce categories without manual rules.

Its existing Retail Search now pulls in customer behavior patterns to personalize query results, while Recommendations AI adds page-level dynamic optimization. With global online retail projected to hit $8 trillion by 2026, Google is making a direct bid for every layer of the retail tech stack.

Why it matters

Google's approach here is clever because it tackles the hardware bottleneck head-on. Instead of forcing retailers to buy proprietary camera rigs, it leverages a massive 1-billion-product database so stores can just use a phone or existing overhead cameras.

However, a 90% accuracy rate in a lab setting is a double-edged sword. In the thin-margin world of brick-and-mortar retail, that remaining margin of error causes replenishment chaos. As Walmart discovered, high-tech computer vision does not always beat a trained human walking the floor when capital expenses are factored in.

We think this move shows Google Cloud pivoting from raw compute competition to specialized, workflow-specific AI. By tying physical shelf visibility to online search and recommendation systems, Google is attempting to become the invisible operating system for commerce both in-store and online.