Copy article

Starbucks Ditches AI Inventory Tool After Nine Months of Miscounting Milk

ended 25. May 2026

STARBUCKS ditched its AI stock-counting tool after nine months because it couldn't reliably tell one syrup bottle from another.

Starbucks retired its NomadGo-built inventory AI this week across North America. The vendor claimed 99% accuracy. Reuters reports the tool frequently miscounted and mislabelled items. Its own launch video failed to recognise a peppermint syrup bottle. 

The tool was part of CEO Brian Niccol's turnaround plan to fix product shortages he blamed for hurting sales. In February 2026, Starbucks told Reuters the system had already improved product availability. Three months later, it scrapped it and the company is now telling baristas to count milk and syrups the way they counted everything else before the AI arrived. 

NomadGo says it is "continuously learning from customer and user feedback." Nobody has explained who signed off on deploying a tool across thousands of stores on the strength of a 99% accuracy claim that didn't hold up in practice. 

We'd like your views:

  • Starbucks publicly said in February the tool had improved availability. By May it was scrapped. What obligations should companies have to correct public claims about AI performance when internal evidence contradicts them?
  • Starbucks' operating margins in North America have halved in two years. If AI tools meant to improve efficiency are making operations worse, how should UK retailers measure whether automation is actually saving money or just moving the cost onto frontline workers?
  • The instruction to baristas was simply: go back to counting it yourself. At what point does reverting to manual processes stop being a fallback and start being an indictment of the deployment decision?
  • Is the race to deploy AI always in a business' interests or should caution be factored into the cost-benefit analysis?

3 responses from the Newspage community

Copy all

Star Quote
Copy

Colette Mason
AI Ethics Consultant at Clever Clogs AI
The rush to automate inventory is understandable. Manual counting is tedious, it eats labour hours, and humans make mistakes too. Between a barista miscounting a shelf and a system confidently mislabelling products across an entire estate with no one in the loop to catch it means a dented balance sheet.

Humans miscounting tend to notice when something looks or feels wrong and they check. The AI just kept scanning.

Starbucks didn't need a tool that replaced human judgment. It needed one that supported it with better shelf labelling, smarter logical reorder triggers. How about refreshed simple deterministic rules that flag when stock levels look odd and ask a person to check.

It must be too boring compared with a razzledazzle press release about transforming inventory with spatial intelligence and augmented reality.

Sometimes the answer to a tedious process isn't a clever one. It's a reliable one, with a human still in the loop who knows what spiced pumpkin syrup looks like.
Copy

Editor
Mitali Deypurkaystha
AI Governance Strategist & CMO at AISensa
Workers are now wasting half a workday every week fixing AI mistakes, according to automation platform Zapier. The Starbucks syrup saga shows exactly why blind AI hype is becoming a productivity placebo rather than a business breakthrough. We’re in the same Wild West phase we saw during the early internet boom, and then later with social media, where transformative technology is mixed with aggressive salesmanship and inflated promises that collapse under real-world pressure. AI will absolutely reshape the economy, but too many firms are deploying it without properly testing whether it actually works in messy human workplaces. I’ve seen businesses force staff onto AI systems to make them faster, only for employees to spend longer checking unreliable outputs. When a billion-dollar company ends up telling baristas to ‘go back to counting it yourself,' it’s an indictment of rushing automation before the technology or the training is truly ready.
Copy

Editor
Katrina Young
Futurist | Chief Technology Officer CTO/CDO Ethical AI & Digital Transformation Strategist at KYC Digital
Starbucks publicly stated in February that its AI inventory tool had improved availability. Three months later, the system was retired. The bigger issue is not whether the technology failed, but how closely public confidence reflected live operational reality. In real deployments, vendor accuracy figures are often the first thing tested under pressure. A 99% result in controlled conditions changes quickly once systems meet real stores, inconsistent lighting, damaged packaging, shelf movement, seasonal stock changes, and human workarounds. Human-in-the-loop verification structures usually expose those gaps early. The problem may not have been solely the technology, but the absence of strong validation between supplier claims, operational testing, and public messaging. Enterprise AI is advancing faster than many organisations’ ability to properly verify and govern it in live environments.