The term generative artificial intelligence, or Gen AI, covers a lot of ground because of all the capabilities it provides to companies across many industries, including restaurants.
To help break down what those capabilities are, and explain what Gen AI actually is, the Food on Demand Conference held a panel with industry insiders to share their knowledge on the technology. On hand for the discussion was David Azose, DoorDash senior director of engineering, and Brian Lichorowic, Vistry AI chief marketing officer.
Moderating their discussion was Steven Elinson, Amazon Web Services director for the services sector, who described Gen AI as a “foundational model.”
“What you’re doing is training these models based on massive sets of data,” Elinson said. “In DoorDash’s case, it’s personalized voice ordering and menu automation, while Vistry, is making pages of standard operating procedures, manuals and training guides interactive for workers.”
Azose said when it comes to online menus, there are times when operators need to make an update on the fly, which can be time consuming. That’s where one of DoorDash’s GenAI options comes into play.
“I think of a sandwich shop selling a variety of sandwiches and all of a sudden, it’s peak dinner rush and they’ve run out of chicken,” Azose said. “Over the years, we’ve made some things to make it easy in the moment for the customer to know that specific menu item is out of stock. But they may have five other items on the menu that also have chicken and there’s no easy way
David Azose, DoorDash Senior Director of Engineering
As a solution, DoorDash is building an AI assistant that, in Azose’s scenario, could instantly disable all chicken items on the menu until the day the restaurant is resupplied.
“It’s a very intuitive way to instantly interact with the software,” Azose said. “It can let a restaurant operator describe what actions need to be taken and ensure it’s done efficiently.”
In Vistry AI’s case, Lichorowic said GenAI can also be effective in building consistency for the restaurant’s back-of-house work.
“We can develop it to have operational questions, manuals and procedural guidance in multiple languages,” Lichorowic said. “Training time is drastically reduced that way and overall consistency because of that training is improved. You can continually build it. As it gets smarter, it better understands how you’re working with it.”
Along with the training and recipe adherence, the technology can also be used to instantly update operators on new rules.
“Health department regulations are constantly changing and you’re responsible if something happens,” Lichorowic said. “Sometimes dishes need to now be washed at 180 degrees, rather than 175. The technology can keep track of that. A lot of smaller operators don’t have a back-office system, either, so this allows them to set one up as they move forward with their place in the market.”
In the front of the house, meanwhile, operators still find themselves getting plenty of phone calls. However, if a person calls into a restaurant during a dinner rush, Azose said there’s a 40 percent chance that those calls go unanswered, and even when they’re not, many are instantly put on hold and there’s a two-minute wait before a customer can engage.
“That’s a massive opportunity that’s lost,” Azose said. “When you solve this problem with human agents, it’s difficult to do sustainably. But now imagine a world where you can remove the human element and we can build a conversational bot that can act on behalf of the human with the same level of eloquence that a real worker would have.”
As a technology that’s still emerging, Azose said it’s still important to think critically about how it’s implemented and the risks to mitigate. He said it happens internally at DoorDash, too.
“From our point of view, it is paramount that we have a solution that you, the merchant, can trust, and by extension, your customer,” Azose said. “We’re not in the business of compromising that. At the same time, we don’t want to miss out on an opportunity to use the technology to do some of the transformative things it can do.
“What we want to do is leverage the tech for what it’s good at, which is generating human-like conversational responses,” said Azose. “Equally, we want to leverage what we’re good at, which is 10 years of experience in building on top of menu data and integrating with restaurant technology systems.”
