When Starbucks unveiled its AI inventory-counting tool in early September, it seemed like a great example of how restaurants could benefit from AI.
It was practical, tangible, and, apparently, impactful: Using iPads outfitted with computer vision, employees could check inventory in minutes rather than an hour, which allowed them to take stock eight times more often than before, according to Starbucks.
And, importantly, it was more than just a pilot in a few stores. As of early September, thousands of Starbucks locations were using AI for inventory, and the rest of the chain was set to come aboard by the end of the month.
Starbucks was not shy about the technology, which it developed in partnership with a company called NomadGo. It even produced a video featuring employees talking about how the AI made their jobs easier and freed them up to spend more time in the front-of-house.
So it came as a surprise last week when Starbucks said it was spiking the AI stock-taking system. It said it wants to standardize its inventory process for all products, and the AI was apparently used only for milk and other beverage items.
But Reuters also reported that the AI was not working well. It had problems with counting and labeling, which led to inaccurate figures or forced staff to spend extra time fixing mistakes.
Social media posts from employees in recent months seemed to back that up. One complained on Reddit that using the AI to count took twice as long as doing it by hand.
And Starbucks more or less acknowledged that the old-fashioned pen and paper method works better for inventory. “We continue to use technology across our business,” the company said in a statement to our sibling publication Nation’s Restaurant News. “This change reflects being disciplined about where automation adds value.”
This is not just a Starbucks problem. All over the industry, there are examples of AI falling short of its promises to boost efficiency and lower costs. In a survey of limited-service restaurant chains by POS company Qu in March, just 9% said AI was having a meaningful impact on their business, while 43% said they’d seen only limited value.
And in some cases, AI may be actively making things worse. Earlier this month, for instance, a large Pizza Hut franchisee sued the company over its required use of an AI-based software called Dragontail, which was designed to help restaurants manage their delivery orders more effectively.
According to the lawsuit, the system did the opposite, leading to longer delivery times and a steep decline in sales. The franchisee argued that entering third-party delivery orders into the POS manually, as it had been doing previously, was more effective for the business than the automated system.
When an operator says they prefer “tablet hell” over the AI alternative, something is wrong. (Pizza Hut declined to comment, citing pending litigation.)
These AI challenges aren’t just a restaurant problem, either. AI’s impact on productivity, regardless of industry, has been something of a mixed bag. And some research has shown that, while AI can help some workers save time, it can simultaneously create more work for others. Harvard Business Review calls this “AI workslop,” which it defines as “AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.”
In a survey of 1,150 U.S. workers, HBR found that 40% had dealt with workslop in the past month. (Were any of them Starbucks baristas, I wonder?)
Of course, we’ve heard success stories about AI in restaurants, some of which we’ve covered in this column. AI can forecast how many doughnuts a restaurant is likely to sell or analyze business data that would have taken a long time to do manually. That’s great!
Then again, just a week ago, we would have counted Starbucks as one of those success stories. So it’s hard to know what to believe anymore.
This gap between what AI is supposed to do and what it’s actually doing is probably a result of misaligned priorities. There’s a lot riding on the success of AI right now. Investors have poured billions into AI companies. Tech firms like Meta and Amazon are reshaping their businesses around it. Boards are asking: “What’s our AI strategy?”
All of this has put pressure on AI companies to sell more AI, and on the buyers of AI to get more use out of it.
The result is that square pegs get forced into round holes. So when the technology makes it down to restaurant operators in the trenches, it doesn’t work as intended.
This came up a lot at the National Restaurant Association Show last week. Many operators said they were skeptical of all the AI on display. They felt a lot of it was marketing fluff and doubted whether it would perform as advertised, so much so that some have resorted to building their own AI tools. The Starbucks news just validates why they feel that way.
This doesn’t mean AI can’t or won’t ever work in restaurants. There are pockets where it already probably is. But this business is fundamentally physical and human in nature, two areas where AI still has a lot to learn.
Restaurants got along just fine (relatively speaking) before the AI boom. Like any other technology, it’s a tool that can be helpful. But due to forces beyond restaurants’ control, it has begun to feel like a mandate.
The Starbucks situation is a much-needed reality check. As the chain suggested, restaurants should be thinking about where AI really adds value—and where it just adds more work.