Wendy’s AI-Powered Drive-Thru: CIO Matt Spessard on FreshAi & Transforming QSR | Technovation 955

We look at it as an additional crew member. The crew looks at it as an additional crew member and in fact refers to the Fresh AI assistant as "she" most of the time. So that is one huge learning for me: that consumers are becoming more and more willing to adopt technology.

Thanks for having me, Peter. That's a great pleasure.

Yeah, if you've somehow managed to not get one of our hamburgers at this point, I would say what you need to know about Wendy's is that we are fanatic about serving fresh, never-frozen beef to our fans all throughout the globe, with more than 7,000 restaurant locations across many, many countries.

Yeah, so as you provided an overview, I've been in this role since February of 2024, and I'm really responsible for all aspects of Wendy's global technology efforts, from restaurant technology, data management, analytics, enterprise technology, software engineering, to information security. And I think as you reflect on those responsibilities, I'd also share that a big part of my job is driving growth across Wendy's digital channels and really working to improve the customer and crew experience through the most strategic implementation of technology solutions and innovation that we can find to be able to effectively drive the business. So whether it's leveraging things like generative AI to reimagine the drive-through experience or the launch of our leading-edge platform for data analytics inside our organization, we're always looking for ways to optimize the business model.

Yeah, well, first I would say we've kind of given our digital footprint a makeover. We launched new CRM and loyalty capabilities in 2024 while also rebuilding our own digital platform and mobile app. So it's been a year full of improvements, and we've seen some really great business results associated with that. We're more than 40% higher in digital revenue in 2024 than we were in 2023, and a lot of that is due to these new technologies that we've leveraged through the channels that we have. So whether it's the new mobile app that we've built, our web experience, or the kiosks that are located in our restaurants, we always try to figure out the right way to employ them for our customers to be able to engage with Wendy's on their terms.

Yeah, I'd say it's a combination of data-driven and hypothesis-driven experiments. We've also over the last couple of years been rebuilding our data state with the development of a customer data platform that we developed in-house. We've been fueling that platform with all of the different experiences that our customers engage with us today, and then we work to understand across a variety of different inputs how they feel about each of those experiences—whether it's how much friction they might encounter, how we work to remove that, whether we're offering the appropriate products to them, and if we need to tune that. So we take all those inputs and use that data to derive a number of different experiments, and we try to do that in an iterative way. We may offer a particular product at a certain time of day for a customer cohort to understand if it's more relevant to them. We may make small changes to the app and deploy that to a small portion of our user base to understand with A/B testing if this is going to work better for this particular cohort or not. Then we effectively just iterate through these feedback loops and try to understand how we can continuously improve across both the experiences and the products we provide.

Yeah, I would say the interesting part is there are two aspects. One is all the existing technology that we have in our restaurants and how we make that better for our crew—easier to use, ensuring it's available and reliable. We've actually been upgrading our restaurant technology and infrastructure inside the restaurants to ensure we are on the latest point-of-sale version and have good network connectivity. We plan to launch more technology capabilities over time, so we want to make sure we have technology that the crew can depend on and that's easy to operate. Then there's the facet of how we look for new ways to innovate and improve the experience. That's really where a technology like our Fresh AI platform comes into play. Initially, we were setting out to understand how to redefine the drive-through experience in a way that resonates with our customers. We have a long relationship with the drive-through, having developed the modern pickup window in 1970, and we've constantly tried to own that space. With Fresh AI, we were trying to understand if we could offer an experience with high-quality interaction, improved accuracy in order-taking, and a conversational experience similar to talking to a crew member, while also improving the crew member experience by freeing them up to focus on making sure the right items go into the right bag and get to the right customer, and providing the smiles and warmth at the end of the interaction. So we're also looking for other areas like that over time. We want to optimize the customer experience and deliver value, but we also want to make it easier for the crew on the other side, so we always look through both lenses.

Certainly. I'd like to say it was as well thought out as you just articulated in terms of preparation of the data before we went down the path of determining the right way to layer generative AI on top, but I think it was a bit more iterative and experiment-driven than that. So if I take a step back, about three years ago we started on this path of really trying to master small-scale experiments through our innovation pipeline and understand if there were technologies we could use to solve a particular business problem. We became fanatic about the process of industrializing the idea of proving value early or ejecting things out of the pipeline. That's where what is now Fresh AI started. It began with the hypothesis of whether we could have a conversation with a customer and leverage technology to automate the order-taking process, starting with just one item. There were a couple of different technologies we could use to solve that problem. At that time, generative AI was not as ubiquitous as it is today, and rule-based AI solutions were prevalent. So we dipped our toes in each scenario to understand first from a rule-based scenario if we could bring it to life before going further down the generative AI path. The rule-based solution could answer the question of order-taking, but as we tried to expand the menu, develop the rules and guardrails for conversations, and scale to a restaurant pilot, the complexity was at least an order of magnitude greater than we initially thought. We had developed a partnership with Google Cloud and started working with them on this. As generative AI became more prevalent, we realized the rule-based version wouldn't work, so we moved to leverage generative AI. We used the menu data to train a large language model and taught it to speak 'Wendy's.' We iterated on those interactions, captured feedback from each loop, and made it a little better each time. Once we could go through a full turn of a conversation, place an order, and ask conversational questions, it became easier to see the path to a pilot. We rapidly accelerated into a single restaurant pilot in the Dublin, Ohio area, and we kept hammering the order interface with different transactions, first from the technology teams and crew members, then opened up to customers for limited tests. We expanded the net broader in that single location, with our team spending thousands of hours in the dining room talking to customers and crew about their interactions. In parallel, we demonstrated the technology to franchises and business leaders to show what was working well and the progress we were making, as well as some straits we were running into, so we could showcase improvements next time. We spent a year in location one before feeling comfortable expanding to additional locations. We weren't interested in just blowing up the use case to many locations because we could; it was important to do it right since these are our customers, and the experience needed to be good.

Yeah, well, I'd like to say we invented an industry-leading innovation process, but I would say what we did instead is take a relatively industrialized process around innovation—the pipeline that you would normally see evangelized in a variety of books—and made it a bit better for Wendy's. The process hinges on this idea of a pipeline of ideas. The ideas could be a type of technology that is up and coming that you want to understand if there are business applications against it, but more commonly it's a business challenge we're throwing out there to see if we can solve it in a way that optimizes an experience for either crew or customer. We have several steps in that process, from a lightweight feasibility study to understand if the technology works well enough and if the problem is viable, all the way to a pilot and eventually an incubator before a product launch. At any point, we eject ideas that don't pass the muster. We consider success ejecting six out of seven ideas. I love the idea that the rejection process is a determinative success. It suggests pursuing a lot of ideas, a lot of green-light thinking, but also being brutal and ruthless in terms of what goes further forward and receives broader investment.

No, it's a full-time job, but it's a small team. I think the best innovation teams are like that because a lot of innovation comes from constraint to begin with. Only having a couple of folks helps keep our focus very consistent and continue to evangelize the idea that we have to eject things. We simply don't have the capacity to take many things through that team and into production. This idea of ruthless prioritization and rapid prototyping is incredibly important to us. There's even a bit of a celebratory event when we eject things from the pipeline because we always capture and share the learnings internally at our demo days and other areas. Even for things that were not successful, we learn something, and it's important to radiate that to the rest of the teams so we know how to use those learnings in the future, whether for another run at a similar problem with a different technology or other opportunities.

Yeah, that's a really great question. They are uniquely suited for the role. The team itself is just a handful of people with backgrounds like product leadership, software architecture and engineering, and some expert-level business analyst chops as well. It's really just three to four folks who are kind of a SWAT team of technical depth that can take us rapidly through the analysis and prototyping piece. The idea that we can have somebody build something out and then destroy that infrastructure and code at the end once it's done is also incredibly beneficial because we don't have any dependence on third parties in the mix; we can just roll through it as rapidly as possible.

I think what I would say is we have had a lot of confidence and faith in our approach to the solution from the beginning. The pacing particularly early on—of sitting in one restaurant for a year to make sure this was viable before scaling to additional restaurants—is what's given us the confidence to scale to the level we're at now. Today we're in just under 100 restaurants across 18 states, and we're deploying in restaurants every single day because we have the level of confidence to move forward without worrying about needing to take a step back or pause deployment. What we've built we know will be able to achieve the objectives. The other thing we learned a lot about is, thinking back to rule-based solutions, we saw many organizations deploying them, and we kept an eye on that as we considered making a change ourselves. Because it was a pivotal moment—whatever you did from there, you would have to be able to effectively scale, and changing later would be much more complex than committing to the decision up front. So we spent a lot of time analyzing what was in the marketplace, whether solutions were available for sale or built by companies themselves on the rule-based front. When we realized there wasn't really anybody who had figured out the generative AI space, we were comfortable being first, especially with a really good partner in Google.

Yeah, I've been in the industry 26 years, and one thing that's really interesting to me is that while technology has changed quite a bit over that time frame, the expectations of customers and crew really haven't. The bar has been raised in many cases, but the fundamentals of what people are looking for when they come to work at Wendy's or go through the drive-through for some hot and crispy fries have been relatively consistent. That's helpful because as you think about how to apply technology to improve things, the basis for innovation in terms of how many things have been tried over the years is well documented. Consumer trends in terms of folks' willingness to adopt technologies have changed quite a bit over time, and that has had interesting implications for how we think about technology innovation at Wendy's. For example, there was probably a point in time when I'm not sure we could be experimenting in some of the areas we're experimenting in now. But if you think about voice as a technology, voice assistants have existed in homes for many years; people are used to interacting with them. The technology we deployed is tweaked to provide the Wendy's level of quality experience, and I would argue it's many steps above the experience they would be used to with some of those other voice assistants. That's great for us because we look at it as an additional crew member, and the crew refers to the Fresh AI assistant as 'she' most of the time. So that is one huge learning for me: that consumers are becoming more and more willing to adopt technology across the footprint. If I think back earlier in my career, that really held a lot of innovation back. Kiosk adoption 10 years ago was very low; folks would bypass it and go directly to the register to interact with a human. Now folks are very comfortable using those technologies and in fact expect them, particularly with some of the younger generations. So the changing consumer willingness to adopt technology, along with a consistent basis upon which restaurants operate, has helped us start to think about how to innovate more differently over time.

Yeah, this is a really fun conversation to have because there are probably three technologies that I'm really interested in continuing to learn more about, both in terms of the engineering behind them and in terms of consumer and restaurant applications. One is—and it's not really the newest kid on the block—but I'm a big fan of the applications of edge computing. I think as AI becomes more ubiquitous in restaurant and retail in particular, the need for compute that sits closer to the operation, running small models with small data sets at high speed, will become more sought after. Another one I pay a lot of attention to is low Earth orbit satellite technology. It seems like a simple concept—things closer to the Earth's surface are faster than things further out in space like geosynchronous satellites—but even at Wendy's, we've benefited by deploying this technology in some of our more remote locations with really good success. So I appreciate that it is providing connectivity in places that might not otherwise have had access to the internet. Then the third trend I find most fascinating is the idea of spatial computing—the notion that we could virtually engineer things with our hands and abstract some of the technical complexity that could get in the way of creativity. I could definitely see someone learning how to make a Dave's Double that way. So I'm keeping an eye on that one.

Well, spicy nugs for sure is my number one. Even if I get something else, I'm still getting spicy nuggets on every single order. I'm just a huge fan of them. But I'm a little more boring on the other side because I am just a big fan of just a Dave's Single or a Dave's Double, depending on how hungry I am.

I am currently reading a book called 'Neuroeconomics: Decision Making in the Brain,' I think is the title. To me, it's a very interesting book about the role of not just the active parts of your brain that you are managing as part of this interaction, but also what's going on in the background as you think about how to interact with decision-making. The book is primarily about financial things, but there are implications in the book that make me really reflect on a lot of different things day-to-day. Interesting read.

Oh, thank you for having me.