Amazon and Panera think conversational AI is the future of easy ordering

How two retail giants are putting large language models to work.
article cover


· 3 min read

Panera Bread is seemingly all-in on implementing emerging technologies at its restaurants. Or maybe it’s just all-in on its partnership with Amazon.

Toward the end of March, the chain became the first national restaurant to use Amazon’s biometric scanning software, Amazon One, rolling out palm scanning for users of the MyPanera loyalty program. Less than two weeks later, Panera announced it had teamed up with Amazon’s Alexa Skills team to offer improved AI-powered voice ordering.

Alexa, Amazon’s voice assistant technology, has long powered voice ordering for retailers (including Domino’s and Dunkin’). But the latest update to Amazon’s Food Skills API offers a smoother, more conversational ordering experience, and only recently became possible, thanks to major advances in artificial intelligence.

“The future is really about going from thought to order in a matter of seconds,” said Mark Yoshitake, GM and director of Alexa Skills. “We had this dream of being able to do this when we first started, many, many, many years ago, but the technology…in terms of being able to handle the language variation, was just not available.”

The updated Food Skills API that Panera recently implemented is distinct from the voice ordering used by other retailers because it uses conversational AI, which means the ordering experience can replicate an in-person scenario.

Before the update, Alexa’s voice-command offering was built like a phone tree: for example, pressing 1 to order a pizza prompts a specific response by the system, which then offers another, predetermined set of actions for the consumer to take (press 4 if you’d like toppings).

Retail news that keeps industry pros in the know

Our free retail newsletter delivers insights to keep retail leaders and decision-makers up to date.

That approach doesn’t allow for the variation in the way humans speak, Yoshitake said, and those experiences didn’t gain traction, in part because users found it more intuitive and convenient to use a mobile app or call in an order.

“[We] realized we needed to start doing something that was much more turnkey, and that meant truly understanding the end-to-end life cycle of the ordering experience,” Yoshitake said.

Creating that experience required mapping Panera’s entire menu into Amazon’s conversational AI facilities, teaching the model the menu and every possible variation.

“We needed to understand, for every one of the food items on the menu, what are the modifications that are possible?” Yoshitake said. “Can you cut it in half? Can you add whipped cream?” On top of that, there’s the variation in human language (do you order “mayo” or “mayonnaise?”)

Building the new ordering system was made easier by the fact that Panera already had detailed menu data to enable ordering on the mobile app or website, Yoshitake said. And Amazon itself has food-specific language data on its retail side, from recipes to food experiences, which also helped to train the model, he added.

For restaurants, the main benefit is an improved loyalty experience through ease of access and friction-free reordering, Yoshitake said.

“It’s really trying to bring the best of the advancements of conversational AI together with the intuitive patterns of customer experiences in a way that delights customers and brings food to the table as quickly as possible,” he said.

Retail news that keeps industry pros in the know

Our free retail newsletter delivers insights to keep retail leaders and decision-makers up to date.