Edit ModuleShow Tags
Edit ModuleShow Tags

Holly, the Drive-Thru AI, at Work in Good Times


Published:

A Good Times worker pointing to the little computer with the big job of taking drive-thru orders.

Behind the sliding glass window of a typical fast-food drive-thru is a chaotic place. Employees sitting there have to process orders, hand off food and listen to all manner of customers through background noise, music and passing traffic. 

But the company Valyant AI says it has largely solved the order-taking task with a new digital employee named Holly. Founder and CEO Rob Carpenter said despite the explosion of consumer devices like Alexa, Google Home and Siri, creating a conversational AI for enterprise was “stupid hard,” even for monstrous tech companies. 

“We spent about a year with Google and Amazon and IBM’s Watson trying to get them to work,” said Carpenter. “If it was a matter of adding a few bells and whistles, they would have done it. That’s the reason nobody does enterprise—it’s tricky. It’s one thing to be in a quiet environment asking it to play a song, but much different when you’re in a noisy, confusing environment.” 

When Alexa can’t find "Don’t Bring Me Down" by ELO, it is certainly frustrating, but when there’s revenue attached, it just has to work. With drive-thru orders making up the majority of a restaurant’s business, even a 5 percent error rate can mean thousands and thousands of dollars. 

So starting from scratch, Carpenter said they had to figure out three big things to get Holly to work: getting drive-thru audio to the cloud, interpreting that complex language and what customers mean, and then the “logic engine” that would create the response. And it had to be accurate every single order with every single customer. 

None of those things were easy. The company designed a computer to plug into the microphone base station and it took two years to figure out the language processing necessary to gauge intent. And the decision tree for the AI was a sprawling, complex project. 

“Surprisingly, the logic engine was incredibly complicated, you have to make sure the product is in inventory, it’s the right time of day. And how you get it in the POS system, all that stuff you have to figure out. It became 10 times more complicate than we thought it would be,” said Carpenter. 

It took two years of work while testing at a Good Times restaurant in Colorado, starting with breakfast. 

“We started with breakfast, the very obvious reason being it’s simpler. For the breakfast items, there’s probably only 40 things on the menu when you count drinks, so it’s a smaller subset than lunch or dinner which is double that,” said Carpenter. “Where it gets complicated is the customizations.” 

Today, Holly is taking 95 percent of drive-thru orders, with the remainder being kicked to an employee when something especially complex comes up. He said the company is still in “whack-a-mole” mode fixing minor bugs here and there. 

Beyond the academic novelty of an AI taking drive-thru orders, the test has borne some interesting results. Carpenter said the restaurant has seen 6 percent average increase in spending in the drive-thru in the past year of testing. Holly, it turns out, is pretty good at upselling the right item. 

“We look back at tens of thousands of orders from historic purchases, and we ran those against time of day, season, even historic weather data,” said Carpenter. “Just take something simple like a drink. What are you going to recommend in 20-degree weather versus a 90-degree day.” 

Already, Holly knows that not many people want a iced tea in the middle of winter; they’ll probably prefer a latte.

The AI has also helped some with order accuracy. When employees can focus on bagging food and processing payments, it’s one less distraction. 

But the true beauty of AI is that, like a human order taker, Holly learns. By parsing through a year’s worth of customer orders, it’s gotten right to that eerie edge of technology. 

“What we’re really excited about for the next generation, purely through that audio, we’re starting to get really accurate data around gender, sentiment or mood and their decisiveness. So if they’re more decisive, they’re less apt to take an upsell,” said Carpenter. 

That’s the grand promise of AI, getting closer to the power of ecommerce in a physical location. Amazon, for instance, can probably figure out how old your kids are based on purchases, and can suggest an appropriate book or toy based on that data. At some point, Holly and other AI like “her” can hear that you’re a bit indecisive, in a bad mood and trying to find what you want. She’ll be happy to recommend some of Good Times’ high-margin Hatch Green Chile Fries for some caloric therapy. 

Carpenter said Valyant will continue to scale with Good Times and is in talks with other drive-thru-focused brands.

Edit Module
Edit ModuleShow Tags
Edit ModuleShow Tags
Edit ModuleShow Tags
Edit ModuleShow Tags
Edit ModuleShow Tags
Edit ModuleShow Tags


Covers everything from good news to bad judgment

About This Blog

The latest news, opinions and commentary on what's happening in the franchise arena that could affect your business.

Tom KaiserTom Kaiser is senior editor of Franchise Times. He can be reached at 612.767.3209, or send story ideas to tkaiser@franchisetimes.com.
 
Beth EwenBeth Ewen is senior editor of Franchise Times. She can be reached at 612.767.3212, or send story ideas to bewen@franchisetimes.com.
 
Nicholas UptonNicholas Upton is restaurants editor at Franchise Times. He can be reached at 612.767.3226, or send story ideas to nupton@franchisetimes.com.
 
Laura MichaelsLaura Michaels is editor of Franchise Times. She can be reached at 612.767.3210, or send story ideas to lmichaels@franchisetimes.com.
 
Mary Jo LarsonMary Jo Larson is the publisher of Franchise Times Magazine and the Restaurant Finance Monitor.  You can find her on Twitter at
 twitter.com/mlarson1011.
 

Archives

Categories

Feed

Atom Feed Subscribe to the Franchise Times News Feed »

Recent Posts