Reasons behind the failure of the Starbucks AI inventory tool at full scale.

      On April 3, Starbucks informed NomadGo that it would be discontinuing the AI inventory tool created by the Redmond startup. Within a few days, the 30-employee company significantly reduced its workforce, as reported by GeekWire. This included the technical team managing the Starbucks account, while baristas were not notified until six weeks later.

      "There’s nothing you can do when leadership and strategy change," said David Greschler, NomadGo's CEO, describing the decision as a total surprise. This account was revealed in a Fast Company investigation released on Monday, which involved insights from numerous Starbucks employees and the startup itself. We previously reported on the cancellation in May, but details from the vendor's perspective had yet to emerge.

      Baristas were informed on May 18 to remove the QR tracking codes from the backroom shelves and revert to manual counting.

      The issues that arose were basic and practical. A shift supervisor near Seattle aimed an iPad at a steel refrigerator, leading the camera to capture a reflection that indicated five cartons of oat milk as ten. In other instances, the app misidentified milk types, swapped syrups, and miscounted food items.

      A store manager in Graham, Texas, faced the opposite dilemma when a loss of Wi-Fi erased her count midway through, and her managers had already deemed hand counts unacceptable, resulting in a total lack of usable data.

      The underlying model itself was not at fault. NomadGo’s computer vision achieved 99% accuracy in controlled settings and had promised counts that were up to eight times faster than manual methods. Greschler provided a more nuanced explanation, stating that computer vision struggles when the inventory frequently changes. Seasonal cups and limited-time packaging could require up to six weeks of retraining, and his developers often learned about new products only after they were available on shelves.

      Another older constraint pointed to the Starbucks backend, which operates on a legacy IBM AS/400 system from the 1990s, complicating the transfer of real-time store data. Insiders estimated that the program might have cost more than $10 million over several years.

      This was not a pilot program. The Starbucks AI inventory tool, named Automated Counting, had been implemented in all 11,300 company-operated cafes in North America by the end of September 2025, but was gone by May 18. Enterprise AI failures are often characterized as pilots that never scaled, but this instance scaled first, making the error particularly costly.

      MIT’s NANDA initiative found that 95% of enterprise generative AI pilots had no measurable profit impact, with most enterprise AI investments still confined to laboratory settings. The UK’s Office for National Statistics observed similar adoption trends, noting that while AI usage broadened, it did not deepen. Conversely, Starbucks adopted the tool deeply across the board simultaneously.

      Starbucks publicly defended the tool shortly before its termination. Reuters reported on the miscounts in February, to which the company responded that adoption had enhanced product availability, as noted by CNBC. About eight weeks later, Starbucks instructed NomadGo to discontinue the tool.

      Greschler attributed the decision to a change in leadership and strategy, although he did not specify which leadership shift was involved. The timeline of events is recorded. Deb Hall Lefevre, then Starbucks CTO, lauded the deployment in NomadGo's launch press release on September 3, 2025. She resigned on September 29, the same month the rollout concluded, and Starbucks appointed a new permanent replacement from Amazon in December. The tool lasted about six months beyond its executive champion.

      As for what Starbucks is saying now, a spokesperson told GeekWire, "We use technology to support human connection, not to replace it," highlighting the $500 million spent on increasing staff in its coffeehouses. "When it fell short, we listened to feedback and changed course." This statement merits consideration, as terminating a tool used across 11,300 locations is a more complex task than allowing a pilot to quietly conclude, and many companies continue to finance software that their employees no longer trust.

      Starbucks has maintained other AI initiatives, including Green Dot Assist for baristas and a ChatGPT integration for customers. However, the statement does not address the supplier aspect, as a tool that creates work instead of eliminating it represents a common pitfall, which researchers have identified. Less often discussed is who bears the cost when a significant client decides against a product. The failure of the Starbucks AI inventory tool impacted 11,300 locations, with the financial repercussions affecting a small company of 30 employees.