A.I. snacks: Potato chip makers are producing new flavors with innovative tech

Poutine, S’mores, and Atlantic Lobster are among the more than 30 flavors visitors to Covered Bridge’s potato chip factory can sprinkle over a fresh bag of crisps. The seasoning session has been the cherry-on-top finale to the tour of the plant’s farmlike facilities, located in Hartland, New Brunswick, but it has also become something of a flavor feedback lab.

“We initially developed a lobster flavor as a bit of a gimmick,” says Brook Dickinson, the company’s national sales director. “But it did so well that first summer that we realized we had something.” Now the Maritime specialty is one of the brand’s summer staples.

In the potato chip business, plain and salted spuds tend to sell best, but new flavors can add zest to sales figures. In 2016, Covered Bridge teamed up with regional pizza chain Greco to create limited-edition donair chips, inspired by the East Coast’s famous sweet and garlicky spin on the Turkish doner. The special batch raked in three years’ worth of sales in a few months.

“Innovation is absolutely key for a percentage of the market,” Dickinson says. “We have to introduce something new every year.” Ideas for the lineup, which has included Sloppy Joe’s, Montreal Steak Spice, and Loaded Hot Dog, come from market trends, trade shows, customers, and social media. Storm Chips, the company’s top-selling winter mix of four flavors, was inspired by a hashtag.

Covered Bridge’s Storm Chips flavor combination.

For leading potato chip maker Frito-Lay, to be innovative at scale requires artificial intelligence. The company’s culinary experts and scientists make decisions informed by a custom-built engine that monitors social media, retail sales, restaurant trends, and news and search analysis. It’s helped churn up flavors like Red Curry Coconut and Champagne Vinaigrette & Shallot for Red Rock Deli, its artisanal Australian chip brand, and the newly released Lay’s Fried Green Tomato.

With the data it collects, the engine can map flavor life cycles and predict staying power based on the depth and quality of conversations. This also lowers the risk of a flavor flop—and the potato chip industry has seen more than a few. The Raspberry Bellini blend put out by British crisp maker Tyrrells did not go down well with consumers when it was introduced two years ago. Jim Beam Bourbon chips were met with disapproval more than 10 years ago, even though they didn’t contain alcohol, but have since made a comeback with Burts. Timing, as it turns out, is everything: According to Frito-Lay, booze infusions are on trend this year.

Red Rock Deli’s Red Curry Coconut chips were popular enough that they were released in the United States.

Artificial intelligence has also helped Frito-Lay trim 10% off its development process. As the database grows, R&D timelines could be cut in half, says Michael Lindsey, chief transformation and strategy officer for Frito-Lay North America. Other snack companies, like Conagra Brands Inc. and McCormick, have adopted similar technology. By 2021, McCormick and its partner IBM expect its database of recipe mixes, extracts, condiments, and more to include more than 1 billion data points and 500,000 formulations.

Lindsey presented on Frito-Lay’s developments in machine learning at MIT Technology Review’s EmTech Digital conference, which was held virtually last week. One of the most significant advancements, he says, has been the technology’s ability to uncover micromarkets. It’s why moviegoers in Frisco, Texas, can now find curry-spiced Kurkure Cheetos—once available only in India— alongside Doritos at concession stands. Micromarket analysis revealed that the city was home to a large population of first- and second-generation Indian consumers, many of whom have a penchant for the snack.

“On a broad level, if you were thinking about what would sell well in Frisco, Texas, you probably wouldn’t land on curry-flavor Cheetos,” says Lindsey. “But when you start diving into the demographics of who lives there and what their preferences are, it lets us bring that variety to life at a micromarket level.” Age, ethnicity, gender, and income profiles are all attributes that can be correlated to snacking preferences, allowing Frito-Lay to target distribution of its 1,200 products to more than 200 North American centers right down to the zip code.

Artificial intelligence also influenced the textural twists in Lay’s latest flavors, including Poppable Sea Salt & Vinegar and Kettle Cooked Flamin’ Hot.

The same logic was used to transform Turbos Flamas from one of the smallest to fastest-growing SKUs. After noticing sales of the tangy, hot lime corn chips spiked in areas with first-generation Mexican-American populations, bags were strategically placed in communities with corresponding demographics. The fiery-purple sacks flew off the shelves, increasing 39% between 2018 and 2019—four times Frito-Lay’s average growth rate.

Strategic distribution doesn’t mean there will be fewer flavors vying for shelf space anytime soon. Frito-Lay peppered its line with more than 50 products last year and is on track to do the same in 2020, though the spread of COVID-19 could slow things down. “Future innovation in salty snacks is vital as brands face intensifying competition for the snacking dollar,” says Lu Ann Williams, director of insights and innovation at global research company Innova Market Insights.

What flavors will land on shelves this year? “Spicy hot is the No. 1 trend, and I think it will stay the No. 1 trend for a while,” says Lindsey. Why? Because there’s almost infinite variety in the genre, from Turbos Flamas to cheddar jalapeño. Williams’s research confirms the trend, adding that pimento chili will stand out, and fruit and citrus will balance the hot and zippy.

Meanwhile, in Canada, Storm Chips will eventually make way for Covered Bridge’s take on a Canadian classic: All Dressed. “We’re hoping we do the Canadian market proud,” Dickinson says.

Your next potato chip may be flavored by artificial intelligence

More must-read stories from Fortune: