Will Artificial Intelligence be a Positive Food System Disrupter?

The food service industry has long been challenged to waste less, with the stakes high. In the U.S. alone, retailers toss about $18B worth of food a year—about twice their total net profits. Some businesses aspiring to do better—both for their bottom line and the planet—are turning to artificial intelligence (AI) for another set of eyes and an extra shot of wisdom. They aim to get a better handle on what they waste and why, so they can course correct.

Camera-equipped vision systems scope out restaurant trash bins to see what’s inside, and connected AI-enabled computers pick up patterns in what’s getting tossed. Machine learning algorithms automatically discount food as expiration dates inch closer. And “smart” inventory management systems forecast demand, informing ordering.

As these technologies roll out, researchers are brainstorming more ways to put AI to use to reduce food waste and emissions and save money.

Now academics at Cornell University are deep into a multipronged project focused largely on restaurants.

Restaurants are challenging; the waste is massive with commercial kitchens tossing up to 20 percent of what they purchase, which is often equivalent to their net profit, according to Elena Belavina, associate professor at the Cornell SC Johnson College of Business and lead investigator.

“So, if you could unlock opportunities to address the problem you can have a huge impact. But there’s very little data to understand how much is wasted. Even if you have data, what would you do next? Where should you look to reduce food waste?”
Those are questions she’s working to answer.

Partnering with AI technology developer Winnow, she and her team are looking at two models to measure and classify waste. One is fairly basic as technology goes; a scale positioned over a trash can weighs food as it’s tossed. Kitchen workers enter on a tablet what was thrown out and why.

Then the team took it up a notch, introducing a video camera that snaps pictures of food as it’s thrown out, recording the details automatically.

The most basic system, simply leveraging a scale and tablet, yielded a 29 percent reduction in food waste within three months. Adding on computer vision drove another 30 percent reduction.

This came as a pleasant but initially puzzling surprise. A closer look uncovered that when kitchen workers manually entered the data, about three percent of events where they tossed food were uncategorized. That may not sound like much in the big scheme of things, but that three percent corresponded to 26 percent of all the waste by weight. 

“So, I’m throwing out a lot of food and not logging what it is, missing important details. While when we eliminate manual recording, we catch high-impact events and can target additional improvements,” Belavina says.

The Cornell team has gone on to develop what they call a machine learning classifier. It’s trained to recognize waste patterns and to predict the probability of certain biases kitchen staff have that may have led to that waste. Being human they make calls like overordering based on previous demand.

“[The classifier] is a learning system that can tell you what you are doing wrong so you can reassess to be able to make informed production decisions.

But we are also working on an AI copilot that would actually tell you what to do— maybe how much of what to order and when,” Belavina says.

Artificial intelligence can help drive food waste reduction across the value chain, starting from initial sourcing, all the way down through post-consumer waste, notes Danielle Joseph, managing director and head of the Closed Loop Ventures Group at Closed Loop Partners.

“What we are particularly excited about at Closed Loop Ventures is innovations that reduce food waste before it reaches the consumer. For example, solutions that help match order volumes to market demand can deliver operational cost savings alongside food waste reduction,” Joseph says.

With their scale, grocery stores have huge potential to cut their food waste. But they too need better data for more insight.  Grocers have between 15,000 and 60,000 SKUs in each store. While they may have bar codes to enhance inventory management, SKUs don’t have the power of machine learning models that identify demand patterns, help improve forecasting and ordering, and ultimately help decrease the amount of food that ends up unsold.

A pilot conducted a few years ago across more than 1,300 West Coast grocery stores showed the potential for impact using AI-powered technology to improve order accuracy. The end result was a 14.8 percent reduction in food waste at each store, on average, and prevention of 26,705 tons of CO2 equivalent emissions from landfills.

If the entire grocery sector implemented these solutions, an estimated 907,372 tons of food waste could be prevented, representing 13.3M metric tons of avoided CO2 equivalent emissions and more than $2B in financial benefits, says the Pacific Coast Collaborative, who wrote a case study on the projects.

It can seem like a big hurdle for businesses to change out or update their existing legacy systems to add AI solutions, but the efforts can be worth it, says Dana Gunders, executive director, ReFED.

“Just last fall, Albertsons Companies announced the enterprise rollout of the Afresh platform into meat and seafood departments in more than 2,200 stores—the company had been working with Afresh since 2022, and after a two-month pilot testing the software solution in the meat and seafood departments, they began an immediate chainwide rollout last year,” Gunders says.

“These are the types of stories that other food businesses pay attention to. So, we’re expecting to see more and more uptake of AI solutions.”