Andrea McCullough, Director/Digital Strategy and Analytics at Dunkin’ Brands, discussed this subject at the Ad Age Next conference in New York.
“I still get a lot of raised eyebrows about how a coffee and donut company is using machine learning,” she said. (For more, read WARC’s in-depth report: How the new (AI) reshapes the old (a cup of coffee) at Dunkin’.)
One application of this technology for Dunkin’ has involved gaining deeper insights regarding the members of DD Perks, the brand’s loyalty program, which is accessed through a mobile app and provides transaction-level data.
“Because we’re collecting that data, we’re learning more about our customers – specifically, what motivates them, and how to get them to visit our stores more often,” McCullough said.
“And, because we’re collecting all this historical data, we’re shifting to the use of machines to learn about [our customers] and predict how we can get them to spend more at Dunkin’.”
This means when a customer uses a mobile phone to order a coconut-crème-pie-flavored iced coffee with a glazed donut on Monday and then checks in with the same order on Wednesday, for instance, AI pays attention.
“Machine learning allows us to know that customer, to understand her behavior, and actually start to recommend products that we know she very likely might want to buy,” McCullough explained.
In taking that one-to-one, brand-to-buyer experience to a mass-marketing platform, she continued, “The power of the machine allows us to [make those pairings] at scale across all customers”.
It also provides the kind of insight to support mobile campaigns that encourage purchase – as evidenced by a rewards program that offered 150 points for trying three menu items.
“This is an offer that we served our loyalty guests back in August  … Because it was all digital, we were actually able to track the results closely,” McCullough said.
In addition to this approach, Dunkin’ tested three new products that the targeted customer had never previously enjoyed. “We were testing the waters of what guests responded to and how smart the machine could be,” McCullough said.
Sourced from WARC