The idea behind Automated Counting was simple: instead of employees manually tallying inventory in refrigerators, shelves, and storage areas, they could scan the area with a handheld tablet or mobile device.

NomadGo’s software analyzed the camera feed using several technologies:

• Computer vision to recognize products
• 3D spatial intelligence and augmented reality overlays
• Depth sensing and LiDAR-style scanning to map shelves and containers

Employees could move the tablet across a refrigerator or shelf and the system would automatically identify items—such as milk jugs, syrups, or coffee bags—and count them almost instantly.

The companies promoted the tool as dramatically faster than manual counts, claiming it could deliver results up to eight times faster while improving accuracy.

For a company with thousands of locations and high‑volume beverage ingredients, even small improvements in inventory visibility could help reduce shortages and streamline supply‑chain planning.

What Actually Happened in Stores

Once deployed in real store environments, the system struggled with accuracy.

Employees reported several recurring problems:

• Miscounted items when the camera scan missed products or counted incorrectly
• Mislabeled products, especially when items looked similar
• Unrecognized items that the system failed to detect at all

A particularly common issue involved different milk varieties, such as whole milk, nonfat milk, and alternative milks. Workers said the system sometimes confused these visually similar containers, producing incorrect inventory counts.

Because these ingredients are critical for drink preparation, inaccurate counts could lead to flawed restocking decisions and product shortages.

Why Starbucks Ultimately Shut It Down

Starbucks ended the program in May 2026, about nine months after deployment, according to internal communications reviewed by reporters and confirmed by employees.

The company said it would retire Automated Counting and return those categories to manual inventory processes. Milk and other beverage components would be counted the same way as other store inventory going forward.

The decision was tied to a broader effort to:

• standardize inventory practices
• improve store‑level consistency
• support daily replenishment and supply‑chain execution

In short, the automated system did not deliver inventory data reliable enough for operational decisions.

A Lesson From a High‑Profile Retail AI Pilot

The rollout was notable because it happened at enormous scale: tens of thousands of employees were using the technology across thousands of stores.

But the experiment highlighted a common challenge for enterprise AI systems. Technologies that work well in controlled demonstrations can struggle in messy real‑world environments where lighting varies, shelves are cluttered, packaging changes, and products look nearly identical.

For Starbucks, those small visual ambiguities were enough to make automated counts less trustworthy than traditional manual checks—at least for now.

The company has not ruled out future automation in inventory management, but the short life of Automated Counting shows how difficult it can be to deploy computer‑vision systems reliably across large retail networks.

The idea behind Automated Counting was simple: instead of employees manually tallying inventory in refrigerators, shelves, and storage areas, they could scan the area with a handheld tablet or mobile device.

NomadGo’s software analyzed the camera feed using several technologies:

• Computer vision to recognize products
• 3D spatial intelligence and augmented reality overlays
• Depth sensing and LiDAR-style scanning to map shelves and containers

Employees could move the tablet across a refrigerator or shelf and the system would automatically identify items—such as milk jugs, syrups, or coffee bags—and count them almost instantly.

The companies promoted the tool as dramatically faster than manual counts, claiming it could deliver results up to eight times faster while improving accuracy.

For a company with thousands of locations and high‑volume beverage ingredients, even small improvements in inventory visibility could help reduce shortages and streamline supply‑chain planning.

What Actually Happened in Stores

Once deployed in real store environments, the system struggled with accuracy.

Employees reported several recurring problems:

• Miscounted items when the camera scan missed products or counted incorrectly
• Mislabeled products, especially when items looked similar
• Unrecognized items that the system failed to detect at all

A particularly common issue involved different milk varieties, such as whole milk, nonfat milk, and alternative milks. Workers said the system sometimes confused these visually similar containers, producing incorrect inventory counts.

Because these ingredients are critical for drink preparation, inaccurate counts could lead to flawed restocking decisions and product shortages.

Why Starbucks Ultimately Shut It Down

Starbucks ended the program in May 2026, about nine months after deployment, according to internal communications reviewed by reporters and confirmed by employees.

The company said it would retire Automated Counting and return those categories to manual inventory processes. Milk and other beverage components would be counted the same way as other store inventory going forward.

The decision was tied to a broader effort to:

• standardize inventory practices
• improve store‑level consistency
• support daily replenishment and supply‑chain execution

In short, the automated system did not deliver inventory data reliable enough for operational decisions.

A Lesson From a High‑Profile Retail AI Pilot

The rollout was notable because it happened at enormous scale: tens of thousands of employees were using the technology across thousands of stores.

But the experiment highlighted a common challenge for enterprise AI systems. Technologies that work well in controlled demonstrations can struggle in messy real‑world environments where lighting varies, shelves are cluttered, packaging changes, and products look nearly identical.

For Starbucks, those small visual ambiguities were enough to make automated counts less trustworthy than traditional manual checks—at least for now.

The company has not ruled out future automation in inventory management, but the short life of Automated Counting shows how difficult it can be to deploy computer‑vision systems reliably across large retail networks.