Starbucks took a bold step in September 2025. They launched an AI-powered tool to automate inventory counts across North American stores. The promise? Faster stock checks and fewer out-of-stock surprises. But just nine months later, the coffee giant pulled the plug.

When AI Meets Real-World Shelves

The tool, called Automated Counting, used tablet-mounted cameras and LiDAR to scan shelves filled with milks, syrups, and other beverage ingredients. It was supposed to replace tedious manual counts and give store managers instant stock updates. Sounds like a dream, right? But the reality hit hard.

The AI struggled to tell similar products apart. Oat milk got mixed up with dairy milk. Peppermint syrup sometimes vanished from the app’s count. These may sound like small errors, but on store shelves, they matter big time. If the system says you have enough oat milk, but you don’t, customers lose trust. And lost sales pile up.

Even a promotional video Starbucks released at launch showed the AI missing a peppermint syrup bottle. That clip now feels like a red flag ignored. Employees noticed the tool wasn’t reliable. One internal message read, “The thought behind it was great, but the execution was proving difficult.”

The Bigger Picture: Why Inventory Matters

Inventory tracking sounds boring, but it’s a critical business puzzle. Starbucks has faced product shortages for years. CEO Brian Niccol pointed to these gaps as a key reason for lost sales. In early 2024, fewer than a third of deliveries arrived on time and in full. That’s a supply chain nightmare.

Automated Counting was part of Niccol’s “Back to Starbucks” turnaround plan. It aimed to give live, store-level visibility and cut down stock-outs. But the tool’s failure shows how tricky it is to automate real-world retail workflows. The technology struggled with the simple task of counting bottles correctly.

The AI wasn’t generative, but its failure fits a growing pattern. A recent MIT study found that 95% of enterprise AI pilots fail to move beyond testing. Starbucks’ experience adds to the evidence that integrating AI deeply into physical stores is hard. The demos look good, but the real world is messy.

What’s Next for Starbucks and AI?

Starbucks is not quitting technology. The company is ditching the AI counting app but doubling down on supply chain improvements. Stores will return to manual counts for milks and syrups, just like other inventory items. At the same time, Starbucks plans more frequent daily replenishments to keep shelves stocked.

Niccol continues investing in other AI tools. New systems help sequence orders and assist baristas during busy times. These tools aim to improve speed and customer experience without the counting headaches.

The AI vendor, NomadGo, said it’s learning from feedback to make its products better. But for now, the test is whether old-fashioned manual counts and smarter restocking can do what AI could not — keep every bottle and carton where customers expect them.

Lessons Beyond Coffee

Starbucks’ AI inventory saga highlights a key lesson: AI tools must prove themselves in the chaos of real stores. No matter how slick the demo, simple tasks like telling milks apart can break an AI system. Enterprise AI is often harder than it looks.

For companies betting on AI to solve everyday problems, Starbucks’ story is a cautionary tale. Sometimes, human hands still hold the best solution. But the search for smarter, faster inventory management is far from over. The AI revolution in retail is just getting started — and the next success could be just around the corner.