Starbucks Ends AI Inventory Tool After Nine Months Over Errors

@HedgieMarkets I'm pretty sure customers just want their coffee. They do not care if AI was involved or not.

Hard to see why CEO's are afraid of "falling behind" if they don't use AI to get coffee to the customer.

@HedgieMarkets The thought of sensor cameras all over a store monitoring everything, including customers, is just too creepy. With any luck there will be a backlash, with customers patronizing low-tech businesses that are "Run by humans, for humans."

@HedgieMarkets Got to give these guys credit, though. It's not easy taking the greatest technological breakthrough in a century and turning it into a capitalist nightmare. Props to corporate.

@HedgieMarkets AI inventory failure at Starbucks

@HedgieMarkets If it was implemented properly and with the right staging, it might have worked. This is all about CEOs measuring up against each other.

@5minute_witness @HedgieMarkets excuse me? if I don't see the AI-Powered sticker on a store window I do not walk in. period.

@HedgieMarkets The rush to integrate in to Anything and everything without due diligence is remarkably stupid and Raises Lots of questions but primarily Why?

@HedgieMarkets That Starbucks story is the real enterprise AI test. Demos pass in clean rooms, deployments fail on two syrup bottles that look the same.

@ForShedeur 🦔Haha! The bar for narrow tasks is genuine though. Counting syrup bottles should be solvable. Counting them while distinguishing between five visually similar variants on a shelf, in variable lighting, across 16,000 stores, is where it broke.

It would be nice to have this kind of technology in the future.

I’d think that this could be accomplished with RFID today, but the cost is likely prohibitive, probably $0.01 per tag which would be per bottle. Still, it might be worth it in the long run if you could have real-time inventory.

Inventory CV has a brutal accuracy bar demos hide.

99% on 10K daily SKUs = 100 errors/day per store. Errors compound — bad orders, bad waste forecasts, stockouts that spread.

A 50-store pilot surfaces the 99% ceiling in 30 days. But piloting is the slow career path. Deploying is the bold one.

@HedgieMarkets AI can't tell you a good joke but you want it to count your syrup inventory? Outrageous!

@HedgieMarkets @UK_Daniel_Card I worked for Starbucks in the 90’s and did many inventories. Happy to talk.

@noiseykid @HedgieMarkets True. Still hard to explain this extreme departure from common sense and practical thinking though.

I remember similar silliness during the "every company must have a mobile app!" years.

@HedgieMarkets AI is nothing more than a statistical data too. Bad data is the achilles heal of AI. AI is just a complex parlor trick, an illusion

@raincoaster @UK_Daniel_Card 🦔Happy to hear the perspective. The institutional knowledge from people who actually did the work is what these AI rollouts always overlook...

@5minute_witness @HedgieMarkets They truly don't. But the CEOs aren't looking at the customer when they launch straight into the sh... They only have eyes for the investors.

@HedgieMarkets @ProfessorTom Why don't the bottles have QR labels?

@HedgieMarkets That's the interesting part, is the market rewarding AI talk or AI walk? A lot of companies are talking the talk on AI implementation during earnings calls but backing that up with productivity improvements is much more difficult.

🦔RFID is the boring answer that would have worked. The technology has been mature for 20 years, the unit economics are well understood, and the failure modes are predictable. Computer vision got chosen here because it tells a better story to investors, not because it was the better tool for the job.