Starbucks has decided to stop using a new inventory management system that relied on computer vision technology. The system was introduced in over eleven thousand of its stores just nine months ago. Developed in partnership with a startup called NomadGo, the tool aimed to make inventory counting much faster—cutting the time from one hour to just ten minutes. It used tablet scanning and artificial vision to track items, and during testing, it achieved a 99% accuracy rate. However, after being deployed in stores, the system ran into several problems. One issue was that reflections on refrigerated glass cases caused the system to count some items twice, like milk bricks. Additionally, unstable Wi-Fi networks in stores led to data loss, making the system unreliable. Adjusting the AI models to account for seasonal packaging changes also proved to be time-consuming, requiring up to six weeks of retraining. These challenges made the system harder to manage than expected. Another major problem was the integration of the new system with Starbucks’ existing information technology. The company’s older systems, including an IBM AS 400 platform from the 1990s, were not compatible with the modern computer vision technology. This created delays in processing real-time data from stores, highlighting the need for modernizing legacy systems to support new digital tools effectively. In response, Starbucks has shifted its focus, redirecting $500 million from the abandoned project to support in-store teams and refocusing its AI efforts on improving internal operations and personalizing customer relationships. The startup NomadGo, which had relied heavily on Starbucks as its main client, has now lost its key partner and has had to lay off most of its employees. This outcome underscores the challenges of implementing cutting-edge technology in large, complex organizations.