Starbucks Retires AI Inventory System After Nine-Month Trial
The "Automated Counting" tool, developed with NomadGo, struggled with accuracy issues including mislabelling milk types, prompting CEO Brian Niccol to scrap the initiative.
Starbucks has discontinued its "Automated Counting" artificial intelligence inventory system in North American stores after a nine-month trial period. The software, developed in partnership with NomadGo and launched in September 2025, was designed to streamline stock tracking using mobile devices. However, the coffee chain has reverted to manual inventory counting for beverage components and milk following the tool's failure to accurately track stock.
The initiative was intended to automate the tedious task of counting items such as milks and syrups, with the goal of increasing accuracy and optimising the supply chain. Employees, referred to internally as "partners," were expected to use handheld tablets to scan shelves, allowing them to instantly view stock levels and spend less time in backrooms. The project was overseen by CEO Brian Niccol.
Despite the strategic intent, the software frequently mislabeled and miscounted items. Reports indicate the system struggled to distinguish between similar milk types, sometimes skipping them altogether. An internal company newsletter, viewed by Reuters, confirmed that beverage components and milk would now be counted in the same manner as other inventory categories, marking a return to manual processes.
Prior to the discontinuation, the tool had been promoted in a September blog post by Chief Technology Officer Deb Hall Lefevre. The post, which has since been deleted, claimed the system would ensure ingredients like cold foam and oat milk were always available. However, an embedded video in the original post inadvertently showed the system failing to detect a bottle of peppermint syrup during a scan, foreshadowing the operational difficulties that followed.
Employee feedback suggests relief at the decision to scrap the programme. One worker noted to Reuters that while the concept behind the automation was sound, the execution proved difficult. The exact number of stores affected by the rollout is not specified, only that the system was deployed across North American locations.
