In theory, using robots to take drive-thru orders has a lot of benefits for restaurants.
Also in theory, robots can take orders faster and more accurately than humans. That can free up those humans to do other things, or even allow the restaurant to use fewer humans or schedule fewer hours. And a robot never forgets to ask the customer if they want fries with that.
In practice, though, training a machine to work the drive-thru isn’t easy. As more restaurants adopt the technology, they have been faced with a simple question: Is it ready? Today, as some of those efforts move from limited tests to wider rollouts, chains embracing the bots say the answer is yes.
They cite order accuracy rates of 90% or above, faster service, more discrete upsells and happier employees as reasons why they believe the technology is ready for prime time. And while not all customers like the idea of ordering from a bot instead of a human, some operators say they have been surprised at how positive the consumer response has been.
“I see this wave building, where we’re kind of getting past the early adopters phase of AI,” said Ben Bellettini, VP of sales, restaurants, at AI voice company SoundHound. “I think that what we’ll very soon see is a sweeping adoption across dozens and dozens of QSR brands in a very short amount of time.”
Some outside observers have echoed that sentiment. In an analysis of McDonald’s use of drive-thru bots last November, investment bank Cowen wrote, “We view wider roll-out as a matter of when, not if.”
One of the early challenges for drive-thru voice bots has been their ability to take orders accurately and without the need for human help.
If you’re on TikTok, you may have seen videos earlier this year of McDonald’s bots struggling to get orders right. The burger chain has been testing the technology since 2021, following its earlier acquisition of voice technology provider Apprente. At that time, McDonald’s reported an accuracy rate of about 85% in a pilot at 10 Chicago stores.
“We have seen at that 85% order accuracy no falloff in overall restaurant customer satisfaction scores,” CEO Chris Kempczinski said during the Bernstein Strategic Decisions Conference in 2021. “We do believe that, and we've seen, if you go below 80, [satisfaction] really starts to fall off.”
An analysis by investment bank BTIG last summer found that McDonald’s bots’ accuracy rate across 24 locations in Illinois was about 80%.
White Castle CEO Lisa Ingram, meanwhile, said during the FSTEC conference last September that the chain’s drive-thru bot, known as Julia, had an accuracy rate of about 75% in tests at just a few restaurants. “She’s not yet where we want her to be, but she improves every single week,” she said at the time.
Part of the problem is that there are countless ways customers can verbalize their orders, and the technology needs time to learn them.
“What’s been amazing is to listen to how many different ways people order sliders,” said Jamie Richardson, VP of marketing and public relations at White Castle, in an interview. The verbiage can range from “Gimme some sliders” to “Please, I’d like a combo No. 1 with extra pickles,” he said.
All of these variations had to be programmed so Julia could process them accordingly. “Once it was input and part of the learning, it never went away,” Richardson said.
The software from SoundHound now gets about 90% of orders right, without human intervention, he said. And customers always have the option to speak to a human if they want. After testing the system for 15 months, the slider chain now feels confident enough in it to expand it to more locations: It’s aiming to bring Julia to 100 units by the end of next year.
Accuracy will be an ongoing process for voice bots, one that could be sped up with the advancement of sophisticated large language models such as ChatGPT.
“We’re going to get to higher, higher, higher levels of accuracy using these technologies and hopefully at some point reach perfect accuracy,” said Xavier Casanova, CEO of Presto Automation, a provider of AI voice technology for restaurants.
Casanova declined to reveal the level of accuracy Presto’s system has achieved, noting it differs from brand to brand and even from store to store depending on factors like traffic noise.
“Our goal is to have a very high accuracy rate,” he said. “We know that is extremely important to our customers.”
The main catalyst for restaurants’ growing use of voice bots is the industry’s ongoing labor troubles. As of June, restaurants remained about 80,000 jobs short of pre-pandemic levels. Some are turning to automation to fill those gaps and/or ease the strain on existing employees.
So it’s no surprise that voice bots’ biggest impact so far has been on this front. Operations that have adopted the technology say it has indeed made workers’ jobs easier and more satisfying, which has yielded other benefits.
“Oh my gosh, every restaurant I visit, they just can’t thank me enough for putting in the technology,” said Minh Le, CIO for Checkers and Rally’s. “On the employee side, they just can’t live without it.”
That has translated to lower turnover at the twin drive-thru chains, which now have voice technology from Hi Auto in about 350 of their 800-plus locations.
The technology has allowed the brands to cut four hours from their schedules each day, just enough to cover the costs of the system, Le said.
The automation also gives employees more time to do other things. At Carl’s Jr. and Hardee’s, the introduction of voice bots at about 90 restaurants has freed up drive-thru staff to interact with guests, prepare food and do inventory. But its overall staffing levels have not changed as a result, said CTO Phil Crawford.
“Our team members are our No. 1 asset,” he said. “This tool is to help them be more efficient and more effective at their role.”
At White Castle, the number of people working in its restaurants has actually increased as hiring has improved, Richardson said. Though it’s too early to say whether Julia is helping with retention, the bot is well-liked by staff.
“We have surveyed team members, and the overwhelming [response is] ‘Please don’t ever take this away,’” he said. “Just the comments alone make you feel good.”
In addition to taking orders, voice bots can act as salespeople in the drive-thru. And proponents say they’re even better than humans at upselling, because they can be programmed to make the right pitch at the right time, and they never get tired of asking.
The Cowen report highlighted automatic upselling as voice bots’ greatest asset. It estimated they could increase a restaurant’s sales by 10% to 15% with suggestive selling.
At Carl’s Jr. and Hardee’s, the most common upsell is asking guests if they would like to upgrade to a larger size. And it has been effective.
“We’re seeing a check average increase, we’re seeing an upsell increase rate,” Crawford said. “And all those are impactful to the overall bottom line of the restaurant.”
Still, brands have learned to fine-tune this function to avoid irking customers.
“We’re learning to listen for better cues about when to upsell,” said Richardson of White Castle. For instance, if someone has already ordered fries and cheese sticks, that might prompt the system to ask if they want a dessert rather than another fried side. “The last thing you want to do is have a good opportunity turn into a negative,” he said.
Checkers and Rally’s has similarly found that there is a time and place to upsell, and has dialed in its strategy accordingly. When the chain first began rolling out bots, everyone got the same suggestions. “As we scaled, we found quickly that one-size-fits-all does not work for everyone,” Le said.
For instance, the bots had been asking every customer who ordered a burger if they would like to add bacon or add cheese. During the busy lunch hour, the extra questions were slowing things down.
“A human wouldn’t do that,” Le said. So the chain has adjusted the flow of upsells based on the daypart. “That change to reduce upsells when there’s volume is one of the biggest changes that we’ve done,” he said.
Overall, the bots have sped up service at Checkers and Rally’s, Le said. The chain did a study comparing its automated drive-thrus to its traditional ones and found that service times were about 7% faster when bots were taking orders.
A big question hanging over the adoption of drive-thru bots has been how customers will respond to the change.
According to a recent consumer survey by researcher Intouch Insight, 47% said they didn’t like the idea of a robot taking their drive-thru order. “Many times AI can't understand certain words and it's often difficult and frustrating to correct errors,” one respondent said. Another raised concerns about what sort of data was being collected by the bots and how secure it would be.
Chains using the technology acknowledged the experience isn’t for everyone, particularly older folks. But they insisted that most people are OK with it. Crawford of Carl’s/Hardee’s said he was surprised by how little pushback there has been.
“We’re seeing a time in our lives post-COVID where more and more people are dependent on technology,” he said, arguing that innovations like voice bots are now expected by consumers.
Richardson compared the bots to self-checkout kiosks at grocery stores. While the machines have now become commonplace, there will always be some people who prefer to pay for their groceries the old-fashioned way.
“We realize that when it comes to acceptance of something new, it’s not gonna be for everybody,” he said. “What we’ve heard from so many of our consumers is, ‘Oh, this is pretty cool.’”