Leveraging AI To Improve Customer Service In Fastfood Restaurants With Rob Carpenter
The customer service is very important when running a restaurant. If you're ordering drive-thru and the employee misses an order, that's not a good look for the company. Maybe the employee is too tired to upsell because they have been working for eight hours straight. There are a lot of things that limit them from being 100% focused on the job. The CEO of Valyant AI, Rob Carpenter believes he has found the solution to this. Join your host Chad Burmeister as he talks to Rob about his proprietary conversational AI. Learn how AI technology can improve the customer service experience.
Listen to the podcast here
Leveraging AI To Improve Customer Service In Fastfood Restaurants With Rob Carpenter
I'm here with Rob Carpenter. I was asked, “Chad, what type of technology has changed in 2020 that you think is a game-changer on the show?” I couldn't think of anything top of mind that was outstanding and amazing, and then I met Rob. Rob, welcome to the show.
Thank you so much for having me. I'm excited to be here.
Valyant.ai. Why don't we cut to the chase scene because we’ll get to know you in a minute? Usually, I start with getting to know you. Tell us a little bit about Valyant because it's such a neat technology that you're deploying.
We've built a proprietary conversational AI platform that thinks like Siri, Alexa or Google. We're bringing it forward from being a consumer toy or something for the back office. We're sticking this thing front and center inside enterprises and customer-facing roles. You don't see any type of AI that's made big strides in that type of space yet.
Imagine your average quick-serve restaurant where a customer pulls into a drive-through. You have an employee inside the restaurant that has to take that order. They also have to process payment, fill up soft drinks, put food in the bag, check for accuracy and clean up for a spill. All of that pressure on one person leads to 250% to 300% turnover per year. By the way, there are 1.4 million unfilled positions within the restaurant industry.
What we do is automate that crucial task of taking customer orders. The employees still focus on payment processing and the other tasks they're too. By taking 50% to 60% of the work off their plate, it severely reduces the pressure on them and allows them to do everything else quicker and more efficient. We're both a hardware-software company. We have a patented piece of hardware called the NX1A that we install directly inside the restaurant.
It hooks into the employees’ access system, which enables our AI to then talk to customers. Employees can follow along on the entire order. Over the last couple of years, when we've been building this technology, we've developed our proprietary speech-to-text engine that is specifically designed to understand noisy drive-through environments. Our own custom NLP and logic engine, which taken together, completes the trifecta for a conversational AI platform that can automate orders in $865 billion per year industry.
My first job as a teenager was at McDonald's. It was $3.85 an hour in Quebec and County Line right there. I remember in the first month, I messed up every order that came in with the special order. “I'd like to have a hamburger with no ketchup and no mustard, plain.” I was on the headset. I was taking the order. I'm in the back and they would say, “One plain hamburger on twelve.”
If you're familiar in restaurant speak and fast food, that means one plain hamburger. While you're at it, you might as well go ahead and make the other eleven normals. I would build twelve plain hamburgers. Think of all the waste. One order times twelve. They would come back and it's been touched. They give it to you. You have to throw it away. Luckily, I'd only throw away probably half of those orders. The other half they would eat, but then the brand is bad. Think of all the mistakes that get made by taking orders at a drive-through that this form of AI is going to solve.
Imagine too they spent three months training somebody to do that job correctly and that person leaves. That's what we're talking for the average tenure in some of these restaurants. When we first went live, one restaurant chain would tend to staff what you consider all shifts 20 to 23 people. Within six months, we were the sixth longest-serving employee. That's the headache and challenges that a lot of these restaurant owners are trying to deal with.
We would come in and plug a critical hole. We also, from an efficiency standpoint, helped to make everybody's lives a little bit better. Maybe to the point of this conversation, we're also consistent for eight hours. You had ten customers in a row reject you. Are you still going to upsell that apple pie to that eleventh person? Probably not, but the AI is going to be chipper in every single order. It's going to consistently upsell exactly when it's supposed to upsell. As it collects more data, it gets more sophisticated and more intelligent with those upsells. That scene has a big impact on driving the ball.
Allan AI is is an educated citizen.
I met the CEO of a chatbot company that's all online chat order processing. At first, the objection of massage envy and those kinds of places were like, “Our people are so good on the phone.” You would try to call the branch and nobody would pick up. It’s like, “Just so you know, they're not picking up.” They deploy from 6:00 until 8:00 at night to say, “Let's try a little bit of load.” The upsell, cross-sell, friendliness and customer sat scores, all of it went off the charts because people will generally mess that up. This is brilliant what you've done over the years.
Let's rewind the tape. I told you we'd cut to the chase scene. Let's go right back to the beginning. I like our audience to understand how you get to where you are. The best way I've found to do that is to go back to when you're a kid. What was intriguing to you? What were you interested in? What were you passionate about?
I'm a space nerd through and through. As much stuff is being thrown at the billionaire space race here, I would love for my opportunity to buy a ticket to go hang out on the international space station or spend a week on the moon. As I looked at that when I was a kid in space camp, I was like, “I can become a PhD, that's also a fighter pilot, that also speaks six languages or I could start a business, build a company and generate wealth.” Even at that time in the mid-'90s, I was like, “I would buy a ticket to space.”
Pretty early on, I knew the trajectory I wanted to take with my life. It was compounded by looking at all the people that were doing the kinds of things that I wished that I could be doing in my life. Who doesn't want to go and fly around on a private jet or go to a private island? You look at people that do that. Almost every one of them either inherited or built and sold companies.
I started to ruminate on wanting to do that. I turned on to Rich Dad Poor Dad when I was sixteen. That was it. The light bulb clicked. I realized how those people made their money and were able to fund things like going to space. Everything lined up for me. I ended up getting an undergrad degree in entrepreneurship and started a student entrepreneurship group at my university campus. I started an advertising business when I was in college and then real estate afterward.
I worked for an internet marketing company for about two years while I was getting my Master's degree. Ultimately, I started a custom software development company that’s been for seven years. I grew it to millions of dollars in revenue. I had a company in India and London. When I was in London closing that acquisition, I came up with the idea for essentially building digital employees. I set out and said, “Where could this technology be useful?”
After analyzing about twenty different industries, I settled on serve restaurants or fast food as we know it. It's a huge market. It's $865 billion per year. They have thin margins, so they're desperate for ways to improve their profitability. There's a huge opportunity to come in and provide a consistent customer service experience.
Selfishly for me, fast food has a far more limited domain set. You might have 150 menu items with another couple of hundred tangential words like napkin and ketchup, but if you look at somebody like one of the big box retailers, they can have ten million product SKUs inside their retail store. That's a huge number of things to be able to understand and converse about intelligence. For us, when you looked at the Venn diagram, fixer restaurants were a nice marriage of market size and potential as well as maybe an easier path to the market than some of the other opportunities.
There are two companies in Denver. If you don't know them, I'm happy to make an introduction. TalentReef is already working with all the big dogs, Check, Wendy's, Taco Bell and all of it. They've got a relationship. They sell a platform that helps hire employees and track that at a low cost. There's OpsAnalitica. Tommy is the CEO and then a former person who used to work with me at TalentReef is also over at OpsAnalitica.
I'll throw a quick plug in. We're trying to hire four engineers, so if they can hire some engineers for me, I'm onboard.
We talked a little bit about AI and all of the different aspects of it. Tell me if AI wasn't a thing, does your product even exist?
No. The quintessential place is we use machine learning as speech-to-text. I'd say we probably have another dozen different machine learning modules throughout our platform, but the core of it is speech-to-text. You can imagine the customer pulling into the drive-through and auditorily, they're saying, “I want two cheeseburgers.” That WAV file then flows into our system. We use machine learning to look at that WAV file to say, “More than likely, this customer is saying, 'I want a cheeseburger.'”
If we don't have any way to convert that WAV file into text, that's the head of the chain, if you will. If you cut that off, nothing else after it works. The reason you've never seen this type of technology in the marketplace before is because AI wasn't advanced enough. The amount of data available wasn't as big as it needed to be. Honestly, the server processing capability wasn't great enough.
All of those forces are coming together and continue to get better year after year. We'll see this technology continue to accelerate in terms of how many edge cases it can handle because as we all know, all AI is these edge cases. As we take on more and more of those edge cases, the accuracy rate goes higher and higher until you get to the point that you don't need any human in the loop in any situation because the AI can handle the vast majority.
I remember meeting a woman named Sabrina Atienza years ago who was curious with Cue.io. They've since been sold or no longer around because I saw she's a Product Manager at Amazon now. Her entire $1.1 million in seed round funding was to try to collapse the speech-to-text time it takes because, with the old stuff years ago, you could do it into a 98% accuracy, but it might take 10, 15, 30 seconds. If I'm tired of doing a live conversation and you're at the drive-through, you're not going to wait 30 seconds. You're going to peel out of there and be pissed off. It was interesting talking to her over dinner at Sales 3.0 one time and that's where they were years ago, the best of the best, but that's been packed down to seconds or milliseconds, I would expect.
I don't know what our exact timing is on speech-to-text in particular. Probably the 750-millisecond range is where we're at. I feel her pain. That's an extreme challenge, especially if we take that and apply it to our use case. It's not just that one interaction where you say something and you're waiting for a response, but our average interaction with customers can tend to twenty turns back and forth. You're talking about 10 to 30 seconds times 10 to 20 times. You're in a ten-minute order time and that's not viable.
We have probably ten critical pieces of our infrastructure that all have to work perfectly in a row in order to complete an entire turn in a conversation with a customer. Every single one of those components, 750 milliseconds, will allow because it's so critical, but then everything else has to be 10 or 50 milliseconds. Otherwise, all of that time compounds and you've got way too long of a delay.
Now, we average somewhere between a second and a half and two and a half seconds in terms of a response to a customer. I'd still like to get that faster. We are continuing to compress that, but there is a very real push and pull challenge between accuracy and speed of response time. If you sit there for ten seconds, you're likely to have an accurate response, but you have to respond fast enough so that the customer is not getting frustrated, driving away and talking over the AI. The problems can cascade exponentially if you get off the path that the customer expects the conversation.
I'm sure my wife would be an edge case because she would do the no pickles and then at the very end of the meal, she'd throw the stumper. “Cancel the hamburger. Add a cheeseburger. By the way, can you make it without any bun?”
Honestly, modifications aren't that challenging. What we see as the real issue and this goes back to speech-to-text again is, is the customer speaking clearly? Are they direct about what they want? If the customer is going to be mumbling, speaking softly or changing their mind mid-comment, that's where we see the degradation of accuracy in that overtaking process and/or any extraneous things going on behind that. Birds can be loud in those drive-through environments. You have car exhaust, backfire, radio, people talking, rain, wind. All of those types of things can impact that.
We benchmark our speech-to-text software. From an accuracy standpoint, we're right around 90% for speech-to-text. We benchmark against Google, Amazon and noisy drive-through audio. Those guys were about 71%, 72% accurate in relation. You have to train on huge quantities of noisy data to get good at it. If there happens to be a leaf blower going behind the customer at the exact moment, they say something quietly, it's probably not going to be a great experience.
I've worked for Webex. We had single-user orders. I worked for RingCentral. We had a whole team of hundreds of people in the Philippines. I have to believe that as this technology gets proven out on quick-serve, it’s built for enterprise. Do you see this moving into large call centers?
As AI accuracy rate goes higher and higher, you will get to a point where you don't need any kind of human in the loop anymore.
We're not trying to hide everything. From our experience, everybody that we've seen in the industry has human in one fashion or another. It's whether they are upfront about it or not. We do have to make up some percentage of orders with a human jumping in. That's where people either locally or overseas will be involved in helping that process to ensure the customer ultimately has a good experience with it.
Through that process, we've talked to a lot of different call center companies, ones that already have a presence in the food space and others that don't. Everybody unequivocally across the board in the call center space is trying to figure out how to get AI technology to work. They build it in-house, where they're acquiring companies that are doing it and building that technology.
We are pushing it to get to the point where we don't need to have the call center backing up the orders anymore. They are consistently high enough that when we do run into an issue, the onsite employee jumps in and takes over. If they have to be involved in 5% to 10% of orders, you can still effectively automate that task. They just jump in the few use cases that they need.
I was tracking the emails. It's AI-powered with one of our partners out of Israel. It got to a point where I'd have two tasks to do a week on eight virtual assistants. My team's like, “Chad, you don't need to go in. You're good.”
Hit that button and let it fly. It's always the magic and terrifying moment.
Even Breckenridge Brewery in Santa Fe, we toured. My nephew is at the School of Mines also. He goes, “What's that little computer down there sitting on the floor?” They go, “That's our AI that runs the mixers of the hops.” He was smart enough. He goes, “Let me ask another question. How did it do?” They go, “In the first couple of loads or mixes, it’s terrible. Now it outperforms a human ten times out of ten.” I'm like, “That's plugged in with a little power cord and an ethernet cable sitting over there on the floor running transactions that a mainframe couldn't have done a few decades ago.”
That goes to show how important the evolution of hardware is. Every single year, this technology gets better. It goes back to Moore's law. You need better processes, whether it's on edge like that is or up in the cloud like a lot of the voice AI providers out there where they do a lot of other processing. You need enough, generally speaking, GPU's to crunch through the numbers so that the AI can do its job.
As that computing power continues to go up and the costs continue to go down, we'll see these things get smarter and smarter. From an AI perspective, you have to find a way collectively to move away from all of this hand labeling to a point where these AI systems are self-learning. In my opinion, the biggest factor I would say is you hire a server computer power. As that increases exponentially, it's probably not going to be more data or more labeled stuff that's going to make a bigger difference. It’s going to be able to crunch more data faster.
Last question. You're living in this every day. You probably have formed an opinion of what happens years from now. You talked about the self-learning stuff. A lot of people I talk to are like, “That's 20, 30 years out.” I'm like, “It's right there. It's already happening in some places.”
Self-learning is going to be essential. I don't think we know necessarily if Moore's law is going to break down in terms of processing capabilities. These processors are getting more and more desperate. “How do I fit more circuits onto the silicone?” You're seeing a huge amount of market consolidation as these big companies try to achieve on some unit economics here to be able to keep making these things better and better.
A lot of people think Moore's law may start to cater out and potentially start to break down moving forward unless there's some big jump to quantum. In terms of unsupervised learning, I would not state that will be here in a few years. It’s getting better. It will be many years before we'll see high-quality, good, unsupervised learning of AI systems, but that doesn't mean it holds everything back until that happens.
Specific to my industry of restaurants, I would argue in a few years, pending legal issues don't blow it up, you will pull into a quick-serve restaurant. It will read your license plate, probably scan your face. There are biometric issues there. You have to opt into that, but it will then greet you by name. It's going to pull up your favorite items. It's going to know what things you normally buy. If you miss something in your order, it's going to ask you if you want that specific item from a sales standpoint. It's ideally going to have your credit card on file, so it's just going to ask you if you can charge it. If you say yes, you will then drive up, grab your food and drive away. I would expect to see that at some level of adoption in the market.
I'll see you on the spacecraft someday because that sounds a lot of fun. Will you go the Elon route, Virgin Atlantic or have you chosen your vehicle of carry yet?
I'm a big fan of anybody that pushes the industry forward. If it's more engineers that are learning, it makes everybody better. I'd be on Elon Musk's starship if I had my choice. They'll have a viable moon base soon.
Rob, it's been great to have you on the show. Rob Carpenter, CEO of Valyant.ai. All the cool kids are putting AI in their name. If you haven't done it yet, partner with a company like ours. This is an awesome talk. Thank you so much for being on the show, Rob.
No problem. Thank you.
About Rob Carpenter
Rob Carpenter is the CEO and Founder of Valyant AI, an enterprise grade conversational AI platform for the quick serve restaurant industry. Valyant has developed a proprietary software application that integrates within a restaurants existing hardware infrastructure and allows the AI software to take the vast majority of customer orders and insert them directly into the POS for payment.
Rob has a master’s degree in Business Administration with a specialization in Enterprise Technology Management. He spent two years on the board for the Rutt Bridges Venture Capital Fund and in 2013 was named one of the top 25 most influential young professionals in Colorado by ColoradoBiz Magazine and in 2016 he received the Denver Trailblazer award.
Prior to founding Valyant AI, Rob was the CEO of AppIt Ventures, a custom software development company with offices in Denver, Hyderabad and London. Over the first six years in business, AppIt built and launched over 350 custom software applications, including mobile apps on all major platforms, backend web applications and sophisticated databases.