Take a minute to think about your grocery list for the week: Whether you’ve got broccoli, bananas, blueberries, or Brussels sprouts on the brain, they’ll likely be waiting in the produce aisle.
Well, just like you, grocery stores have their own grocery lists. And now, some are using AI to help create them.
One grocer that’s signed on to use this type of tech is United Natural Foods, the largest publicly traded grocery wholesaler in the US, and two of the regional supermarket chains it owns: Cub, primarily located in Minnesota, and Shoppers, primarily located in Maryland and Virginia.
The company partnered with Afresh, a San Francisco-based software company that uses deep learning and reinforcement learning to help grocery stores forecast demand for produce, optimize ordering, and ultimately combat food waste.
Luke Anderson, CIO of Cub, recalled the chain’s initial conversation with Afresh taking place about a year ago. He told us the team was interested in Afresh’s specific estimates for how much its tech could reduce Cub’s store “shrink”—an industry metric measuring how much product is shipped to the store versus sold to consumers. So far, the tech’s full rollout has led to an 18% decrease in produce shrink—a significant improvement from even Afresh’s initial savings estimates on the intro call, Anderson said.
In the coming months, United Natural Foods plans to incorporate the tech into its Shoppers chain as well, which has 20 stores compared to Cubs’ 80.
“Thinking about the business model for grocery chains…any shrink, any food waste, is going to seriously cut into their profitability,” Alex Frederick, a senior analyst at PitchBook focused on food tech, told us. “Focusing on the food-waste issue and using tools like AI is going a long way to help grocery stores continue to operate profitably in this challenging market.”
For its part, Afresh has raised $148 million to date, and by year’s end, it’s on track to run the ordering process for 7% of the produce sold in US supermarkets, CEO Matt Schwartz told us. Currently, more than 3,000 grocery stores in the US have signed on with Afresh, Schwartz said—14x growth from the company’s store count at the beginning of 2021. It’s estimated that the US had ~60,000 supermarkets and grocery stores in 2022, per IBISworld.
Connecting Cub
Some of the main reasons that Cub adopted Afresh’s tech: ease of transition, simple user interface, and estimated savings, Anderson said.
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
“They had told us that the IT change would be relatively small—we would get them some files for data that we were probably already sending between systems, and they could take that and incorporate that into their system to get started,” Anderson said, adding that they were also encouraged because the tool seemed “very similar to traditional produce-order guides.”
Produce managers at grocery stores are often tasked with using digital ordering technology for non-perishable items, but for produce, the job can require manual calculations—think: a printed Excel spreadsheet with columns to fill in, Schwartz said—to estimate how much of each item to order.
“Let’s say you’re just going to go in and order…avocados,” Philip Cerles, machine learning engineering manager at Afresh, told us. “Everything in your head when you’re looking at that avocado shelf is based on your historical understanding of how avocados sell.”
That could translate to how ripe the avocados look, how many a grocery store plans to sell before the next order arrives, whether there are any promotions coming up that should help move product, and even how reliable the shipment schedule has proven to be in the past, he added: “Multiply that thinking by, like, 600 items every day—and you get the work experience of a trained produce manager.”
For its part, Afresh’s tool involves an iPad that asks the produce manager questions one-by-one to estimate current store inventory, then merges that manually collected information with the company’s demand forecaster and other machine learning models to reach a conclusion about how much to order.
To do this, Afresh incorporates billions of data points—including historical sales volume, shipment frequency, produce varieties, pricing, past and future promotions, and current inventory estimation—to inform its machine learning models, according to Schwartz and Cerles.
So far, Afresh tech in Cub stores has led to a ~7% reduction in produce inventory held and a 2.5% increase in produce sales, per Anderson. It’s also on track to reduce or save 1,264 tonnes of greenhouse-gas emissions, 43 million gallons of water, and 2.1 million pounds of food waste annually across all Cub stores.