Starbucks scraps AI inventory tool after nine months
https://finance.yahoo.com/sectors/technology/articles/starbucks-scraps-ai-inventory-tool-165449189.html
Starbucks scraps AI inventory tool after nine months
https://finance.yahoo.com/sectors/technology/articles/starbucks-scraps-ai-inventory-tool-165449189.html
42 Comments
>The program struggled with basic product identification. Errors were common, Reuters found, with the system routinely confusing visually similar products like different milk varieties or overlooking stocked items altogether. A Starbucks promotional video from the launch period captured the malfunction plainly: a peppermint syrup bottle sitting on the shelf went unregistered as the system scanned the surrounding bottles on either side of it.
Why do AI tech people love to show off demos of their products actually failing if you pay attention
You’re right! We are out of Milk! I have added 200 pounds of sugar to the next inventory order!
Counting inventory is one of these menial tasks an AI could probably be quite good at. Maybe in a warehouse setting or with clearly labeled products.
But why Starbucks rolled that out to all their stores without rigorous testing is beyond me.
The worst part of all this, AI does not even fookin work!
Did the AI hear “do you have coffee flavored coffee” too many times and quit working? Because I get it.
Not working well uh?
A close acquaintance of mine is on a health plan through her insurance and it comes with a fitness/food tracker app. It used to be like many other apps of its type where you would be able to search for entries of food products and add them to what you ate that day, or fill in a meal manually if it wasn’t in their database. Not perfect or anything, but it got the job done.
Recently the app has shifted to an “AI-assisted” model, where you *describe* what you ate in writing, and the AI LLM tries to figure out what the nutritional data for that would be. It’s always highly inaccurate with the macros, has no idea what certain things are (I think avocado mayo became an avocado and some unconnected mayonaise), the portions of each item provided in the prompt would often be ignored and come out incorrect in the results, and sometimes would just fail to fill in info for an item and leave the data blank. She’s had to get a second app that works the normal way just so she can keep accurately tracking her meals on the side for herself since the official app she’s supposed to use is now slop.
That, the Pizza Hut story, and now this, is showing that companies are really rolling out these AI systems and using them in the real world, and they’re failing hard.
>Getting the right products onto shelves has frustrated Starbucks leadership for years, with multiple chief executives citing stock gaps as a factor weighing on revenue.
I prefer coffee shops that make everything on site.
I sat in on a demo that Figma was presenting to the client I currently work with.
They were demoing Figma Make. They had already created a prompt for some new landing pages.
They showed us the results and it got everything about 80% correct. Well that doesn’t help us. Having some clickable prototype is cool when it works 100% of the time correctly. Otherwise you spend the rest of your day fixing it.
I will never forget the time I walked into a Starbucks to order a hot coffee and the cashier told me, *”we’re out of coffee.* How in the world do you run out of the one thing you are known for making? This is like walking into a McDonalds and having them tell you they are out of hamburgers.
The hilarious thing is that it was next to a grocery store that sold Starbucks coffee. I explained this as a solution, go to the store and buy coffee, but the cashier was not amused.
oh it’s a mess. our backroom is not nearly big enough for the amount and variety of product we have on hand and one slip up in ordering can throw everything in disarray for a week or two. why they ever thought this was going to work plainly showed the complete disconnect from the corporate level and the store level. they have no clue what it’s like to work in a store how day to operations go. starbucks corporate has been living in a fantasy land for a few decades now
This is not even real AI. This is just Apple’s LIDAR sensor with some cute detection counting. A moving scan like that is gonna be definitely inaccurate.
The issue is not the technology, it’s the implementation of it.
What are you talking about, it correctly identified “Not a hotdog”
As someone who does inventory… no shit.
The only way to make sure it’s not hallucinating is to do the inventory yourself and check, making it pointless.
Imagine spending billions on AI just to lose a fight against a bottle of milk.
“We could only find so many places for all the vanilla syrup it kept ordering.”
Salesman got bank selling that, congrats
does anyone know how much of a money suck this failed system was? how many millions? not including the wasted inventory or failed sales due to lack of inventory?
It’s kind of hoping for a funny story like Starbucks AI inventory erases all records and backups.
I guess I should market a barcode scanner and basic ERP as ‘AI’
Businesses are in for a rough time when they realise not only is AI garbage for most tasks (I’ll give it a pass for things like language translation), but these AI companies are all running at crazy losses, and ultimately it’s going to cost way more than the human workers that got sacked.
Why pay humans to correctly inventory when you can pay and AI to just guess?
“What if we rewrote all our inventory software with ‘fuzzy matching’?”
Huh thats strange, inventory managements actually something AIs really pretty good at because a lot of its identifying patterns in numbers. Wait what the fuck do you mean they were trying to use machine vision to automate inventory checking?
Its shocking to me how stupid a lot of companies are about using AI. Identifying products which are visually similar via AI is actually really fucking hard and requires a lot of setup and it just doesnt take that long to have a minimum wage employee just go count bottles. This application was doomed from the start because its just a really stupid use of the tools
LLM AI is the latest tech grift after NFTs, crypto, the blockchain, segways, google glass, 3d printing, theranos… you know, all the other stuff that was going to *CHANGE THE WORLDDD*
It could been useful as a read only system, save it for prediction and analysis but ultimately leave it to humans as usual for the best results.
I’m with AI as a tool in a toolkit. I’m so sick of it being an employee displacement force.
They forgot to tell the ‘AI’ to make no mistakes, didn’t they? Rookie error right there.
It’s scary how much the people with significant resources overestimate the capabilities of this janky tech. It’s very neat, and certainly inspiring, but it’s not fucking good for technical stuff where accuracy is required. Not yet, and probably not ever, no matter how much you scale computation.
They put alpha software into production and discoverd it’s shit? Yeah I see a theme here
I work in construction for a very large company and my bosses have been pressuring me to prove that I’ve been using ai. So today I was doing a very simple task marking up and overlaying plans and he told me to use AI it took me 20 minutes to get a bullshit answer and I went back to my bluebeam markup app and completed the task in 2 minutes
the promo vid thing is wild. seems like they either didn’t actually test it or knew it was broken and shipped it anyway hoping nobody’d notice until after launch hype.
I knew there was a reason Starbucks started selling bags of rocks to eat.
Lemme guess, it sucked?
Palantir time!
A friend of mine manages the Starbucks near where I live. She’s been complaining about this for a while now.
And so it begins
My company tried ai. One operator just knew it was the future and we should get in front of it.
Immediately it made errors. I stopped using it and carried on. Well, the company still uses the ai in that operators market. But no where else. Even the ceo saw it was nonsense.
Wow so guessing isn’t all that useful for actual business application I suppose
AI is just a glorified search engine. It cannot replace actual intelligence.
No way! AI doesn’t actually work? Who woulda thought
So, literally, Stabucks employed a machine to essentially do the exact thing CAPTCHA relies on to distinguish people from machines and, surprise-surprise, the machine failed in exactly the way CAPTCHA said it would.
This is my shocked face. LOL.
I’m an accountant. I cannot see AI being a threat to my profession whatsoever.
First of all, in most companies, the finance department is a skeleton crew. There won’t be layoffs.
Second of all, there can be a high number of transactions going through an accounting system, with each transaction having multiple attribute values attached to it. This has to be deterministic to be accurate.
Third, once transaction volume has reached a certain size, accounting systems have well-tested workflow automations that already take a lot of manual entry out of the equation – and they are all deterministic.
Inventory tracking is done by ERP systems (accounting software but full-featured). Starbucks is discovering that accounting and inventory tracking has to be deterministic.
The only place for AI is high level analysis, or looking for obscure patterns or information.