Loading...
Loading...
Starbucks has paused and effectively scrapped its North American rollout of an AI-driven inventory and labor tool after about nine months, following persistent miscounts, mislabeling and unstable performance. Deployed on tablets using cameras and LiDAR to auto-tally milks, syrups and predict staffing needs, the system struggled with visually similar items (e.g., oat vs. cow milk), missed items, and inaccurate demand forecasts that frustrated store teams. The move underscores the limitations of computer vision and edge AI in complex retail environments, and signals the need for more rigorous real-world testing, model robustness, integration, and failure-handling before scaling automation.
This matters because it highlights real-world limits of computer vision and edge AI when automating inventory and labor planning in complex retail settings, affecting deployment risk and ROI for tech teams. Engineers and product leaders must reassess testing, model robustness, edge hardware, and integration practices before large-scale rollouts.
Dossier last updated: 2026-05-31 02:02:13
Starbucks has retired an AI-powered inventory system after nine months after the tool repeatedly miscounted and mislabeled stock, Reuters reported. The NomadGo-supplied product, promoted as combining on-device 3D spatial intelligence, computer vision and AR and advertised at 99% accuracy, was rolled out in September 2025 under CEO Brian Niccol’s operational push and CTO Deb Hall Lefevre’s endorsement. Starbucks will revert to manual counting for beverage components and milk to ensure consistency across North American stores. The move highlights risks in operational AI deployments and follows other high-profile retail AI failures; NomadGo says it is learning from feedback while Starbucks continues experimenting with other AI initiatives like Green Dot Assist, Smart Queue and a ChatGPT-powered recommendation feature.
Starbucks has paused use of an AI-powered inventory system across its North American stores after about nine months of deployment. The tool, rolled out starting September 2025 under CEO Brian Niccol’s operational overhaul, relied on tablet cameras and LiDAR to auto-count ingredients like syrups and milks. In practice the system suffered stability issues and frequent miscounts and mislabeling—struggling to distinguish visually similar items such as oat milk and cow’s milk and sometimes missing items entirely. Starbucks says it halted the program to standardize inventory procedures across its cafes and refocus on consistent execution at scale.
Starbucks has stopped using an AI-powered inventory counting tool in its North American stores after nine months of deployment due to frequent errors. The system used tablet cameras to automatically tally materials like milk and syrups, but persistent inaccuracies prompted the company to terminate the program this week. The move highlights challenges in applying computer-vision and automation in operational retail settings, where miscounts can disrupt supply, waste, and staffing. For the broader tech and retail industries, the incident underscores the need for rigorous real-world validation, improved model robustness, better edge-device integration, and clearer failure-handling processes before scaling AI inventory solutions.
Starbucks has discontinued an AI-powered inventory and labor scheduling tool after roughly nine months of limited testing, citing poor performance and store-level frustration. The pilot, rolled out to select stores to predict stock needs and staffing, failed to meet accuracy and usability expectations, prompting the company to halt the program and revert to manual or existing systems. Employees reported the tool often misforecasted demand and suggested inappropriate staffing levels, creating operational headaches and undermining trust in automation. The episode highlights risks in deploying AI for frontline retail operations, the importance of rigorous field validation, and the reputational and operational costs when ML systems don't match real-world complexity.