Starbucks has killed its AI-powered inventory counting system across all 11,000 company-operated North American stores, just nine months after rolling it out as a centrepiece of CEO Brian Niccol's turnaround strategy. The tool, built by Redmond-based startup NomadGo, was supposed to replace manual stock counts with computer vision that could tally syrups and milks in seconds. Instead, it frequently miscounted and mislabelled items — confusing similar milk types, missing bottles entirely, and forcing baristas to recount every scan by hand.
This isn't just a Starbucks story. It's the clearest case study yet of what happens when an enterprise deploys AI at scale into a messy physical environment without proving it works there first. And for any business owner evaluating AI tools that promise to automate operational processes, the lesson is worth more than the marketing deck that preceded it.
What happened
An internal Starbucks newsletter dated Monday 19 May, reviewed by Reuters and confirmed by two employees, was blunt: "Starting today, Automated Counting will be retired. Beverage components and milk will now be counted the same way you count other inventory categories in your coffeehouse."
The system used tablet-mounted cameras and LiDAR sensors to scan shelves and produce automatic counts of syrups, milks, and other beverage components. At launch in September 2025, NomadGo CEO David Greschler described it as a "unique synthesis of on-device 3D spatial intelligence, computer vision, and augmented reality" that counted inventory "up to 10x faster than manual methods, with 99% accuracy." Starbucks CTO Deb Hall Lefevre said it would let workers "spend more time focusing on what matters: crafting high-quality beverages and connecting with customers."
That announcement has since been deleted from the Starbucks website. So has the promotional video that showed the tool failing to recognise a peppermint syrup bottle on a shelf while counting the bottles next to it.
Starbucks told Reuters the retirement was "a decision to standardise how inventory is counted across coffeehouses as we continue to focus on consistency and execution at scale." NomadGo said it is "continuously learning from customer and user feedback."
The internal employee reaction was less diplomatic. Starbucks shared screenshots of staff feedback, including: "Very grateful our thoughts about AI count were heard," and "Thank you for trusting the partners over unreliable spatial recognition to handle these counts."
The 99% accuracy problem
NomadGo's headline claim — 99% accuracy, up to 10x faster — was never independently verified before deployment across 11,000 stores. The company said in January 2026 that its technology had counted more than 186 million items across those locations. Volume of scans is not the same as accuracy of scans.
The core failure was predictable to anyone who has deployed computer vision in uncontrolled environments. A Starbucks back-of-house during a morning rush — where oat milk, dairy, almond milk, and breve sit inches apart in near-identical jugs, where lighting shifts, shelves get reorganised, and products are moved constantly — is not the same environment as a demo. As The Next Web noted, "the everyday job of telling one white liquid from another" was enough to defeat the system.
The timeline of Starbucks' own statements makes this worse. In February 2026, when Reuters first reported worker complaints about miscounts, Starbucks told Reuters the tool had improved product availability. Three months later, they retired it.
A pattern, not an anomaly
Starbucks is not alone. The food service industry has produced a remarkable string of high-profile AI rollbacks in the past two years.
McDonald's ended its AI drive-thru partnership with IBM in June 2024 after a three-year pilot across 100+ US locations plateaued at roughly 80–85% order accuracy — below the threshold where human workers typically perform — and viral TikTok videos documented the system adding hundreds of dollars in unwanted items. McDonald's promised to select a new vendor by end of 2024. As of May 2026, no announcement has been made.
This month, Pizza Hut franchisee Chaac Pizza Northeast filed a $100 million lawsuit alleging that the chain's Dragontail AI delivery management system caused "cascading operational breakdowns" across 100+ locations. Before the AI rollout, Chaac said more than 90% of deliveries arrived within 30 minutes. After deployment, year-over-year sales growth in New York swung from positive 10.19% to negative 9.78%.
The common thread is not that AI doesn't work. It's that AI accuracy requirements in physical operations are fundamentally different from knowledge work. A chatbot that gives an imperfect answer can be corrected before it causes damage. An inventory count that is off by two units of oat milk feeds directly into restocking decisions, supply orders, and the product availability numbers that executives are using to measure their turnaround.
What Starbucks is still betting on
The NomadGo retirement doesn't mean Starbucks is abandoning AI. Niccol is rolling out Green Dot Assist, a generative AI chatbot built on Microsoft's Azure OpenAI platform, across US and Canadian stores. It helps baristas look up recipes, troubleshoot equipment, and identify shift cover.
The distinction is instructive. Green Dot Assist operates in a domain where an imperfect suggestion can be reviewed and overridden before it causes a downstream problem. A barista who gets a wrong recipe prompt can check and correct it. An inaccurate inventory count that triggers (or fails to trigger) a restock order has consequences that compound silently.
Niccol has also told analysts that Starbucks will move to daily replenishment by end of 2026 — addressing the same stockout problem through logistics rather than surveillance. Whether manual counts and faster deliveries can solve what the AI could not remains an open question.
The financial picture is mixed. Starbucks posted its strongest quarterly sales growth in two and a half years in its most recent report and the stock is up 24% in 2026. But North American operating margins have fallen to 9.9%, down from 18% two years ago. Morningstar analysts had written as recently as April that AI inventory tracking could improve restaurant-level margins — an assessment that now needs revision.
What this means for your business
If you're a business owner evaluating AI tools for operational processes, Starbucks just ran a $100 million experiment on your behalf. Here's what it proved:
Demand proof in your environment, not theirs. NomadGo's 99% accuracy claim may well have been true in a controlled setting. It was not true in 11,000 Starbucks locations with inconsistent lighting, similar-looking products, and constant shelf reorganisation. Before deploying any AI tool at scale, insist on a pilot in your actual operating conditions — not the vendor's demo room.
Watch the verification loop. The moment your team starts checking every AI output, you've lost the efficiency case. You haven't eliminated a task — you've added a second one. If workers don't trust the system's output, you're paying for the tool and the manual process it was supposed to replace.
Match the AI to the error tolerance. Starbucks' Green Dot Assist works because wrong answers can be caught and corrected before they matter. Its inventory AI failed because wrong answers fed silently into supply chain decisions. The question isn't whether AI is accurate enough in general — it's whether it's accurate enough for the specific consequences of being wrong in your process.
Nine months is the window. Enterprise patience is finite. If an AI tool hasn't proven its value within two to three quarters, it will be pulled — regardless of how much was invested in deploying it. Budget accordingly for both the deployment and the proving period.
At Heygentic, we've seen this pattern across industries: the businesses getting real value from AI are the ones that match the tool to the task rather than deploying the most impressive-sounding technology into the hardest possible environment. Starbucks tried to solve a logistics problem with computer vision. The logistics problem remains. The computer vision is gone.
Sources
- Starbucks Scraps AI Inventory Tool That Miscounted Items, Made Errors — HuffPost / Reuters
- Starbucks ditches AI inventory system after just 9 months — Restaurant Dive
- Starbucks Retires NomadGo Inventory AI Across 11,000 Stores — Tech Times
- Starbucks pulls its AI inventory tool nine months in — The Next Web
- NomadGo Transforms Inventory Counting with Groundbreaking On-Device Inventory AI Technology — BusinessWire
- McDonald's to end AI drive-thru test with IBM — CNBC
- Pizza Hut franchisee claims $100 million losses from AI adoption gone wrong — Fortune
- Starbucks Dumps AI-Powered Inventory Tool Due to Errors — PYMNTS
