After becoming CEO, Brian Niccol made fixing those supply problems a key priority. One initiative was an AI-driven inventory system designed to give stores faster and more accurate visibility into what they actually had on hand.

The technology—developed with Seattle-based startup NomadGo—used computer vision, 3D spatial intelligence, and augmented reality to identify and count products when employees scanned shelves, refrigerators, and storage areas with a smartphone or tablet.

In theory, the system could:

• Replace manual counting of inventory
• Speed up stock checks dramatically
• Flag low supplies automatically
• Improve replenishment and reduce out-of-stock items

Starbucks rolled the system out widely, planning to use it across more than 11,000 company-operated stores in North America.

The Types of Errors the System Made

In practice, store workers reported that the system struggled with basic tasks required for reliable inventory tracking.

According to reports cited by Reuters, the AI frequently:

• Miscounted products on shelves
• Misread or mislabeled items
• Confused visually similar products

One widely cited example involved milk containers. In some stores the system reportedly struggled to distinguish between similar-looking milk cartons, which led to incorrect inventory counts.

Even small inaccuracies matter in inventory systems. When counts are wrong, ordering software can generate incorrect replenishment requests, causing stores to order too little or too much of key ingredients.

Why the Tool Was Ultimately Scrapped

The errors eroded trust in the system among store workers, who relied on the counts to manage daily operations. Instead of reducing shortages, incorrect data could potentially worsen them if stores ordered supplies based on inaccurate inventory records.

As a result, Starbucks decided to discontinue the program about nine months after it was deployed widely in North America.

Internal communication to staff said the company was retiring the “Automated Counting” tool and moving toward a more consistent approach to inventory management while continuing broader efforts to improve supply chains and replenishment systems.

What the Experiment Reveals About AI in Retail Operations

The short-lived rollout highlights a common challenge with AI in real-world retail environments. Systems that work well in controlled demos can struggle with the messy conditions of actual stores—changing lighting, cluttered shelves, similar packaging, and inconsistent placement of items.

For Starbucks, the technology was meant to eliminate a tedious task and help solve a costly supply-chain problem. But when accuracy fell short, the company chose to abandon the tool rather than rely on flawed data.

The episode underscores a practical lesson for retailers experimenting with automation: when the task depends on precise counting and labeling, even small error rates can quickly make AI systems unusable in daily operations.

After becoming CEO, Brian Niccol made fixing those supply problems a key priority. One initiative was an AI-driven inventory system designed to give stores faster and more accurate visibility into what they actually had on hand.

The technology—developed with Seattle-based startup NomadGo—used computer vision, 3D spatial intelligence, and augmented reality to identify and count products when employees scanned shelves, refrigerators, and storage areas with a smartphone or tablet.

In theory, the system could:

• Replace manual counting of inventory
• Speed up stock checks dramatically
• Flag low supplies automatically
• Improve replenishment and reduce out-of-stock items

Starbucks rolled the system out widely, planning to use it across more than 11,000 company-operated stores in North America.

The Types of Errors the System Made

In practice, store workers reported that the system struggled with basic tasks required for reliable inventory tracking.

According to reports cited by Reuters, the AI frequently:

• Miscounted products on shelves
• Misread or mislabeled items
• Confused visually similar products

One widely cited example involved milk containers. In some stores the system reportedly struggled to distinguish between similar-looking milk cartons, which led to incorrect inventory counts.

Even small inaccuracies matter in inventory systems. When counts are wrong, ordering software can generate incorrect replenishment requests, causing stores to order too little or too much of key ingredients.

Why the Tool Was Ultimately Scrapped

The errors eroded trust in the system among store workers, who relied on the counts to manage daily operations. Instead of reducing shortages, incorrect data could potentially worsen them if stores ordered supplies based on inaccurate inventory records.

As a result, Starbucks decided to discontinue the program about nine months after it was deployed widely in North America.

Internal communication to staff said the company was retiring the “Automated Counting” tool and moving toward a more consistent approach to inventory management while continuing broader efforts to improve supply chains and replenishment systems.

What the Experiment Reveals About AI in Retail Operations

The short-lived rollout highlights a common challenge with AI in real-world retail environments. Systems that work well in controlled demos can struggle with the messy conditions of actual stores—changing lighting, cluttered shelves, similar packaging, and inconsistent placement of items.

For Starbucks, the technology was meant to eliminate a tedious task and help solve a costly supply-chain problem. But when accuracy fell short, the company chose to abandon the tool rather than rely on flawed data.

The episode underscores a practical lesson for retailers experimenting with automation: when the task depends on precise counting and labeling, even small error rates can quickly make AI systems unusable in daily operations.