Their model uses a wide variety of data, according to Sam’s Club officials. Things like local temperatures (hot weather often means less pies bought); whether the Sunday football game is home or away (home games can mean more pies are needed); how popular are pecan pies this year (more pecan pies can translate into less pumpkin pie sales).
Those data points, and others, plug into an artificial intelligence model they’ve made. It spits out recommendations to each store leader, such as how many pies need to be on-hand in their shops by the hour. Last year, Sam’s Club sold enough pumpkin pies to fill up 450 football fields, officials said. (They declined to give an exact figure.)
Predicting demand with specificity is necessary, officials added, because the competition to keep customers is cutthroat and profit margins are tight.
“If members aren’t getting what they need, they aren’t going to renew with us,” said Pete Rowe, vice president of technology at Sam’s Club and store member whose family is buying both pumpkin and pecan pie for Thanksgiving this year. “It’s critical for us and our model to make sure.”
Pumpkin pie or pecan pie? With these recipes, you don’t have to choose.
In recent years, sophisticated artificial intelligence models have become commonplace in grocery stores. Spurred on by the pandemic and supply-chain challenges, it’s changing the grocery-buying experience rapidly: from AI-powered shopping carts that recognize the items you’ve picked up automatically to chef bots that generate recipes based on your purchases.
The rise is due to a confluence of factors, according to grocery experts. Stores now have access to mountains of data, including from third-party brokers and shopper loyalty programs. Computer processing power is cheaper and faster. Machine learning models, software that computers use to learn and adapt on their own, have advanced. The pandemic has played a large role.
Gary Hawkins, the chief executive of the Center for Retail and Technology, said in pre-pandemic times, stores used software to help with inventory management, staffing and predicting when goods will be in stock. But after the pandemic hit, “supply chains got blown up, demand went through the roof” and grocery stores were unprepared and needed smarter systems, Hawkins said.
“It literally blew up all the models, because they simply were not sophisticated enough,” he added. “So very quickly, especially the big guys said: ‘We need something better here.’ ”
In April of 2019, Walmart launched an Intelligence Research Laboratory where cameras and sensors are wired into algorithms to monitor how stocked shelves are. In March, Kroger launched an AI lab where technology can keep track of vegetable freshness. Ketchup maker Kraft Heinz now uses machine learning to track demand for its products leading up to events such as the Super Bowl. Amazon opened a fully automated Whole Foods this year that uses deep-learning software to let customers shop and walk out without needing a cashier. (Amazon founder Jeff Bezos owns The Washington Post).
Start-ups have also proliferated. New York-based Caper Cart makes AI-fueled shopping carts that automatically recognize what customers pick up and check them out. Seattle’s Shelf Engine tells stores how many items it needs daily. Hivery, based in Australia, has a model to advise grocers on where to put products on shelves.
“AI is making its way into nearly every tech-related capability,” Hawkins said.
This World Cup is wired and fueled by AI
Dominic D’Agostino, a 30-year-old Sam’s Club member in Dayton, Ohio, said he had no idea the company used such sophisticated technology to predict pumpkin pie demand.
Though he isn’t a fan of the dish, and likely won’t bring any to his sister’s house for the holidays — “the only pie I really like is pizza,” he said — D’Agostino is intrigued, and somewhat concerned, that artificial intelligence is used this way.
“It’s creepy,” he said in an interview. “It’s also fascinating.”
Sam’s Club made the decision to use AI shortly before the pandemic, Rowe said. The chain used software to guide its operations, but felt it could be better.
In years past, for example, Rowe said “we would produce too many pumpkin pies, too many croissants and that [would lead] to our associates wasting their time and also us having to throw away inventory.”
Now, the company uses machine learning to predict inventory for everything they make in-house, such as pies and rotisserie chicken. They also have “autonomous floor scrubbers” — or self-driving robots — to scan shelves and send alerts to staff prioritizing which items need to be restocked first when delivery trucks arrive.
Rowe said it’s helped the store become over 90 percent accurate in predicting demand, and wants it to be higher.
The Google engineer who thinks the company’s AI has come to life
Despite AI’s appeal, it has risks. Algorithms run off troves of customer data, fueling risks about privacy, researchers from the University of Arkansas said. It can also lead to bias.
“Even if race or gender is not a formal input into an AI algorithm,” they wrote, “an AI application may impute race/gender from other data and use this to ‘price higher’ to specific demographics.”
Others note AI is not a universal solution, and stores might waste money buying fancy software just to keep up with the hype.
“You can’t be overly enamored with the shiny object element of AI,” Mike Hanrahan, formerly the chief executive of Walmart’s Intelligence Research Lab, said in a tech publication. “There are a lot of shiny objects out there that are doing things we think are unrealistic to scale and probably, long-term, not beneficial for the consumer.”