Starbucks revisits an AI-driven inventory counting push it previously struggled to make stick

Starbucks is trying again with a technology approach to inventory counting that it had already tested and, according to a recent report, had not fully delivered on expectations. The company’s effort centers on using inventory artificial intelligence (AI) software to count store stock faster and more consistently than manual methods.

The report points back to a deployment made roughly 10 months earlier, in September 2025, when Starbucks rolled NomadGo’s inventory AI tool across about 11,000 stores. NomadGo’s system was positioned as a high-accuracy way to help determine what is on shelves, with claims that it could reach 99% accuracy and count up to eight times faster than a human doing the same work.

In retail operations, inventory counting is a recurring operational burden that affects everything from replenishment planning to shrink control. When cycle counts or full store inventories are slow or inconsistent, companies often rely on more manual workarounds, or they scale back automation pilots because the benefit does not outweigh the complexity of rollout, training, or error handling.

Starbucks’ earlier decision to bring the NomadGo tool into thousands of locations suggests the company believed the performance upside was meaningful. A system that can reliably measure inventory with less human time could, in theory, reduce labor hours devoted to counting and help improve how quickly stores correct discrepancies. The report’s framing, however, indicates that the first version of the push did not meet the bar the company was seeking.

The same report also describes Starbucks building internal software aimed at reducing reliance on established enterprise tools from Microsoft and IBM. In that context, the inventory AI effort is part of a broader theme: large retailers increasingly try to move operational technology from third-party systems to more company-controlled stacks, which can offer flexibility but require sustained engineering and operational validation in the field.

Notably, the report does not lay out specific performance results from the September 2025 rollout in the excerpt available for this review. It does not quantify what accuracy Starbucks ultimately observed, how often the system missed targets, what percentage of stores remained active on the tool, or whether the company revised its approach before expanding or scaling back. Without those operational metrics, it is not possible to determine whether the obstacle was technical accuracy, workflow integration, inconsistent store conditions, or change-management challenges.

For now, what can be said from the reporting is that Starbucks is not treating the earlier attempt as a one-off. Repeating a store-scale inventory initiative after a prior test failure suggests the company may believe it can close performance gaps through software changes, better store workflows, or updated measurement procedures. Investors and industry observers will likely watch whether future updates include clearer outcome tracking such as accuracy against physical counts, time saved per store, and how reliably the system works across different store formats.