If it isn't working, stop it
Starbucks has scrapped its AI-powered inventory counting system across North America just nine months after rollout, following persistent accuracy issues and internal pushback.
The 'Automated Counting' tool which used using computer vision and LiDAR (laser scanning) via store tablets was designed to track milk, syrups, and other ingredients in real time.
Instead, it regularly miscounted and mislabeled items (think oat milk vs dairy), or missing products entirely forcing staff to double-check manually.
This was part of CEO Brian Niccol’s turnaround strategy to fix product shortages hurting sales and was rolled out rapidly across thousands of stores. Initially Starbucks claimed it improved availability before quietly reversing it's course.
This is less a verdict on AI, more as a case study in how fragile deployment becomes when hype meets the stockroom shelf
WHY IT MATTERS
The 'AI Failed' headlines are lazy and click bait.
In reality, the system, like many others, struggled with messy, real-world inputs (similar packaging, lighting, store variability), and created more work, not less. We have all been there.
When outputs aren’t trusted, humans naturally revert to the old ways, override, or duplicate the effort obliterating any ROI. Garbage in, garbage out.
WHAT TO WATCH FOR
Early warning signals your AI isn’t augmenting behaviour, it is fighting it.
→ The increase in shadow processes: staff double checking
→ Frontline sentiment flipping from curiosity to quiet resistance
→ Accuracy gaps in edge cases (similar items, messy environments)
→ AI systems that require human verification to function
→ Leadership language shifting from “automation” to “standardisation”
LIMITATIONS OF THE REPORTING
Relies on internal newsletters, employee anecdotes, and earlier Reuters investigations. We lack full technical diagnostics like data quality, model limits, UX issues). Starbucks frames the move as 'standardisation' not failure, and continues investing in tech.
SOURCE
https://www.reuters.com/business/starbucks-scraps-ai-inventory-tool-across-north-america-2026-05-21/
https://www.cnbc.com/2026/05/21/starbucks-scraps-ai-inventory-tool-across-north-america.html
BESCI AI OPINION
This feels very much like so many Automation projects of the past, even before AI was on the scene.
A mixture of garbage in-garbage out, with over reliance and belief in messy data from a messy environment. The edge cases are a problem.
Then there is the human interface at the point of serve. How does this help an overwhelmed and busy barista?
The learnings aren't new, the sad thing is that an organisation the size of Starbucks committed so deeply without realising what they are walking into.
