Starbucks withdraws its AI inventory system nine months later due to persistent issues with misidentifying the various types of milk.

      The chain is reverting to manual inventory counts throughout North America, marking the end of one of CEO Brian Niccol’s more prominent technology endeavors and adding another example to the category of “enterprise AI pilots that did not withstand real-world application.” According to an internal newsletter reviewed by Reuters and confirmed by the company, Starbucks has discontinued the AI-driven inventory system it introduced in its North American stores last September.

      “Effective today, Automated Counting will be discontinued,” the memo stated on Monday. “Beverage components and milk will now be counted in the same manner as other inventory categories in your coffeehouse,” which translates to manual counting.

      The tool, developed by Seattle-based NomadGo, utilized tablet-mounted cameras and LiDAR technology to scan shelves of syrups, milk, and other beverage components, providing automatic counts and replacing manual stocktakes for certain categories. It had been under development for several years and was rolled out nationwide after Brian Niccol became CEO in September 2024 as part of his “Back to Starbucks” turnaround effort.

      According to reporting from Reuters in February and internal company documents, the tool struggled with the simple task of distinguishing between similar white liquids. The app often miscounted or mislabeled items, especially products that looked alike, such as oat milk and dairy. In a promotional video released by Starbucks at launch, the system failed to identify a bottle of peppermint syrup on the shelf while counting the bottles next to it, a detail that appears worse in retrospect.

      In a statement to Reuters on Thursday, Starbucks characterized the move as a standardization initiative rather than a step back. The company explained that the decision was made to “standardize how inventory is counted across coffeehouses as we focus on consistency and execution at scale,” noting that it aims for more frequent daily replenishments and ongoing supply chain enhancements. An internal note shared within the company included an employee expressing gratitude for ending the program: “The concept was strong, but the execution was challenging.”

      This decision is significant because inventory management was expected to be a simpler issue. Over five years, four different Starbucks CEOs have attributed lost sales to the company's difficulties in maintaining consistent stock levels. In early 2024, the company admitted that less than a third of deliveries to their distribution centers arrived punctually and in full.

      Automated Counting was intended to provide the chain with the real-time, store-level visibility it lacked and was one of Niccol’s prominent operational improvements. Its termination coincides with a broader trend where the record of enterprise AI is appearing less favorable than initial projections. Last year, MIT’s NANDA initiative reported that 95% of enterprise generative-AI pilots showed no measurable impact on profit and loss, despite about $30 to $40 billion in investment, with just 5% reaching production.

      While the Starbucks tool wasn’t generative AI, the nature of its failure is recognizable: a deeply integrated, store-level workflow proved more challenging to automate effectively than demonstrated. The current financial situation is mixed enough that this decision may be interpreted in different ways. Starbucks experienced its strongest quarterly sales growth in two and a half years last month, and its stock has increased by 24% so far in 2026. However, operating margins in its primary North American market have dropped to 9.9%, down from 18% two years ago.

      Niccol continues to invest in other technological initiatives, including AI tools designed to sequence orders and assist baristas during busy periods. NomadGo, for its part, informed Reuters that it is “constantly learning from customer and user feedback” to enhance its offerings. The next challenge will be whether manual counting and daily replenishments can achieve what the algorithm could not: keeping peppermint syrup readily available on the shelves.