Starbucks rolled out an ambitious automated inventory system across more than 11,000 North American stores last year. Nine months later the coffee giant killed it. The decision blindsided the small Redmond, Wash., company that built the technology. And it exposed the messy realities behind big corporate AI experiments.
The tool, called Automated Counting, promised to slash the time baristas spent tallying stock. Instead of spending an hour each day manually counting syrup bottles, milk jugs and pastries, workers could snap photos with a tablet. LiDAR sensors and cameras would do the rest. But the system never delivered on that vision.
Reports surfaced quickly. Items got miscounted. Shiny refrigerator reflections registered as extra cartons of milk. The app failed to spot bottles on shelves or confused one syrup flavor for another. Wi-Fi hiccups wiped out partial counts and forced restarts. Over-ordering became common. One store reportedly received $700 worth of unwanted croissants in a single day.
Reuters first broke the news of the quiet retirement in May. Its story revealed an internal newsletter that told employees, “Starting today, Automated Counting will be retired.” The move came after months of complaints from store workers who found the tool unreliable.
NomadGo developed the system. The startup had partnered with Starbucks in 2025. Back then executives hailed the project as a step toward fixing persistent product shortages. CEO Brian Niccol had made reducing stockouts a priority. Accurate inventory data was supposed to help.
Yet the technology struggled in real stores. NomadGo claimed 99 percent accuracy during controlled tests. Real-world conditions proved different. Seasonal items changed often. The model required constant retraining. And Starbucks still ran much of its backend on a legacy IBM AS/400 system from the 1990s. That aging infrastructure made real-time data integration tough.
On April 3 Starbucks informed NomadGo the partnership was over. The startup’s CEO, David Greschler, called the news a complete surprise. “There’s nothing you can do when leadership and strategy change,” he later told Fast Company. The publication’s recent investigation painted a picture of a young company caught off guard.
Six weeks after that notification NomadGo laid off a large portion of its roughly 30-person team. The cuts hit the technical staff hardest. A business built around one major client suddenly faced an uncertain future. Greschler declined further comment in some reports. The startup’s website still highlights its computer vision and AI capabilities. But the Starbucks episode clearly stung.
Starbucks defended the reversal. A company statement shared with multiple outlets read, “Human connection is at the core of our business, which is why we have invested $500 million to put more partners in our coffeehouses. We use technology to support human connection, not to replace it. This tool was designed to simplify a routine task and give partners more time with their customers. When it fell short, we listened to feedback and changed course. That is what innovation looks like at Starbucks: listening, learning, and adapting.”
The message emphasized people over machines. It framed the decision as responsive leadership rather than a technical failure. Yet the timing raised eyebrows. The tool had only been in full deployment since August 2025. By May 2026 it was gone. Baristas received instructions to remove QR codes from shelves and return to manual counts on May 18.
Industry observers point to deeper issues. Many large retailers still rely on outdated systems. Integrating modern AI with those platforms demands more than clever algorithms. It requires clean data, reliable connectivity and processes that don’t break under daily pressure. Starbucks stores operate in chaotic environments. Baristas move fast. Lighting varies. Products get moved, spilled or hidden behind others.
Fortune covered the episode in detail. Its reporting highlighted barista complaints and the so-called hallucinations where the system invented inventory that wasn’t there. The piece noted how quickly corporate enthusiasm can fade when frontline workers push back.
GeekWire’s July 27 article brought fresh attention to the startup’s side of the story. The Seattle publication described how NomadGo was left blindsided. It quoted sources close to the company and detailed the layoffs that followed. The report arrived just as conversations about AI overpromising gained new momentum on X, with users sharing the Fast Company piece and debating corporate AI pilots.
This episode fits a broader pattern. Corporations race to adopt AI for efficiency gains. They pilot tools in limited settings and declare success. Then they scale. Problems that stayed hidden in tests emerge at volume. Costs mount. Workers resist. Executives retreat. The cycle repeats.
Starbucks has other AI projects underway. It tested an AI-powered ordering companion. It experimented with ChatGPT integration for its mobile app. Green Dot Assist offers another digital helper for employees. Those initiatives continue. The inventory tool, however, became the cautionary example.
NomadGo isn’t alone. Other startups have watched major clients walk away after short trials. The difference here lies in scale. Eleven thousand stores represent an enormous deployment. The failure carried financial and reputational weight for both parties. Yet Starbucks absorbed the setback and moved on. The smaller partner felt the real pain.
Recent coverage adds context. A FullStack Labs analysis published six days ago examined why the automation effort collapsed. It pointed to repeated miscounts and mislabels before the nationwide retirement. The post suggested forward-deployed engineers might have caught problems earlier if given more authority on site.
Quartz and Sprudge offered shorter takes that echoed the Reuters reporting. Both noted the tool’s focus on beverage components like milk and syrups rather than all store items. That narrow scope was meant to simplify the challenge. It didn’t.
The episode holds lessons for technology buyers and sellers alike. Contracts need clear exit clauses and performance benchmarks. Pilots should stress real conditions, not lab perfection. Leadership must weigh cultural fit as much as technical specs. And companies should prepare for the possibility that listening to employees sometimes means abandoning expensive projects.
Starbucks stock barely flinched on the news. Investors appear to view the write-off as minor in a company of its size. NomadGo’s future remains less certain. The startup may pivot to other retailers or industries where inventory needs differ. Its technology could find success in warehouses or more controlled environments.
Still. The story lingers. A promising AI solution met the messy reality of thousands of busy coffee shops. The result was scrapped code, laid-off engineers and a return to pen and paper. Progress sometimes looks like that. Two steps forward. One expensive step back.
But the pressure to automate hasn’t disappeared. Labor costs rise. Competition intensifies. Customers expect speed and consistency. Companies will keep testing AI tools for inventory, scheduling, ordering and more. Some will work. Others will fail quietly. The difference often comes down to execution, not the sophistication of the model.
NomadGo’s experience offers a stark reminder. Even when a giant like Starbucks says yes, the partnership can end abruptly. Strategy shifts. Budgets tighten. Feedback turns negative. Startups that tie their fate to one large customer accept that risk. The reward is visibility and revenue. The cost can be sudden irrelevance.
Starbucks, for its part, doubled down on its people-first message. The $500 million investment in additional staff signaled priorities. Technology should assist. It should not distract or frustrate. When the AI tool did both, the company acted. That decision may prove wise. Time will tell whether other AI experiments deliver the consistency this one lacked.
WebProNews is an iEntry Publication