|
16 June 2026 (San Francisco, CA) - The “AI has failed” information stories continue to appear in my newsfeed. Most I ignore, but due to the ubiquity of an outfit called Starbucks (and the “bucks” word is germane to the business model for hot water sales) jumped into the output of a Thermoplan AG device.
How did Starbuck's dip into the warm bath of AI marketing work out? Well, that vendor of hot water is about to toss out the AI.
The article in State Of Brand explains the details in “Starbucks Sold Its AI On A 99% Accuracy Number. Nine Months Later, 11,000 Stores Are Counting Milk By Hand".
Starbucks installed an AI tool called "Automated Counting", designed by NomadGo, in September 2025. It was described as a way to relieve employees from mundane tasks:
A “unique synthesis” of on-device computer vision, 3D spatial intelligence, and augmented reality. Wave a tablet at a shelf of milk jugs and syrups, and the count appears, validated in AR. Starbucks’ chief technology officer framed it as freeing partners from a time-intensive task so they could get back to making drinks and connecting with customers.
It was removed in May 2026. The job for AI was “simple” like most work processes analyzed by “let’s increase margin” crazed MBAs and flexible bean counters. The system was supposed to count milk cartons.
Eliminating manual counting would free up Excel projected hours. Presumably the time would be used for more productive activities like turning out more expensive hot water and writing in physician scrawl a customer’s name on a cup with a mermaid image.
How did that AI handle the simple job?
Not very well. As anyone who has worked in a busy foody type of environment can tell you (for me, every summer during my university days, at 4 different restaurants), when the rush arrives, life is pretty crazy.
The Starbucks AI system worked great in a store with a tidy, clean shelf, decent lighting, controlled SKUs.
But when caffeine semi- withdrawal "Type As" crowd the counter, the smart software did not work too well.
But no modern corporate “leadership” ever admits mistakes. Starbucks didn’t consider it a failure. Instead Automated Counting is labeled as "more of a learning experience than anything". It was a carefully crafted statement from Starbucks HQ to output what I call smarmy talk.
The reality? The computer-vision technology - designed to track milk and syrup levels via 3D spatial intelligence - could not scale due to frequent identification errors, product mislabeling, and inability to handle the variable conditions of a busy store. Worse, the AI frequently confused similar products (like mistaking oat milk for almond milk) and struggled to detect certain syrup bottles on crowded, poorly lit shelves.
And the baristas were getting pissed off. Because the AI system often hallucinated counts or miscounted stock, store replenishment orders were thrown off, sending the wrong amount of product. This forced baristas to go back into the stock rooms, double-check outputs and often revert to manual counting - rendering the tool counterproductive.
Gee, and it worked so well in the demos the company gave Starbucks management.
Does this AI moment prove anything? Yep. Real world actions are different from those demonstrated in a controlled setting. And there are so many similar stories I could relate.
Modern leadership believes in Tooth Fairies, silver bullets, and the outputs of BAIT acolytes (BAIT being the lingo for big AI tech outfits). Probability is good for some things - but not for all applications.
My view is that modern management is gullible and believes "the next big thing" will wave a magic wand over some business problems.
News flash: There is no tooth fairy.
| 
