I was standing behind the counter when the tablet flashed the count: ten gallons of milk, but there were three empty crates on the shelf. The room tightened—the rush was coming and the numbers were wrong. It felt like a broken scale at a grocer, promising accuracy while tipping everything off-balance.
I want you to keep a clear head here. I’ll walk you through what went wrong, who pushed the button, and why a promise of near-perfect accuracy crashed into the everyday chaos of a busy coffeehouse.
A barista on a Saturday morning discovered the tool labeled items that weren’t there.
Starbucks quietly retired an AI-powered inventory tool after nine months of tests across North America, Reuters reported. The internal memo read, “Starting today, Automated Counting will be retired,” and milk and beverage components will revert to the old counting method. I know how that reads to you: a tech pilot that sounded promising but failed when the lights went on.
Executives framed the rollout as a fix for inventory shortages that were hurting sales.
CEO Brian Niccol, who arrived from Chipotle in 2024, pushed the effort to fix gaps that executives said were costing revenue. Deb Hall Lefevre, Starbucks’ chief technology officer, initially presented the NomadGo system as the automation answer. NomadGo itself boasted 99% accuracy and a mission to “count everything of value in the world,” a claim that looks fragile when a store can’t trust its counts during a morning rush.
Why did Starbucks abandon the AI inventory tool?
It kept miscounting and mislabeling items, according to Reuters. NomadGo’s device paired 3D spatial intelligence, computer vision, and augmented reality—technology that sounds precise but often failed in messy, human-controlled environments. When you run a café, a single mislabel can cascade: orders slow, waste rises, and managers start second-guessing every scan.
A taste-test of the tech showed promise in demos but floundered in practice.
The company released videos and press statements. NomadGo CEO David Greschler described the product in a release, and the firm showed how shelves could be scanned in under 30 seconds. But the scans were “frequently” wrong, Reuters said. That gap—between a polished demo and the unpredictability of real stores—is exactly where your trust erodes.
How accurate was NomadGo’s inventory AI?
NomadGo advertised near-perfect numbers. In the controlled demo, you see 99% accuracy. In the field, Starbucks partners reported frequent errors. The machine vision that excels in clean, static sets struggles with cartons, condensation, and oddly stacked supplies that retail staff handle every day.
A franchisee lawsuit and other misfires made this more than a one-off embarrassment.
Across the industry, AI inventory systems have ruffled more than feathers. A Pizza Hut franchisee sued after an efficiency system allegedly cost $100 million (€92 million) in lost revenue. That’s real money and a real warning: models can compound mistakes into catastrophic outcomes.
Will Starbucks try AI inventory again?
Starbucks hasn’t said it will stop experimenting with AI. The company is running Green Dot Assist, Smart Queue, and a ChatGPT-powered discovery feature in its app. Its spokeswoman told Gizmodo they “test ideas in our coffeehouses, listen closely to partner feedback, and make changes to deliver a better, more consistent experience.” I take that to mean they’ll go back to the lab, but you should expect caution before another full rollout.
NomadGo declined to answer Reuters’ emailed questions but said it’s “continuously learning from customer and user feedback.” I read that as signals and calibration—technology iterating on its errors, not miracles arriving fully formed.
Here’s what I want you to carry away: automation can speed routine work, but in stores where a rush is a storm, the system has to be as forgiving as the humans using it. The promise of flawless counts is seductive, yet reality often pulls the rug out from under those claims—like a compass that points south when you need north.
So after all this, will you trust AI to count your coffee?