Operating a quick-service restaurant is not for the faint of heart. Whether it’s ensuring inventories are sufficient, kitchens are suitably staffed, fickle customers are satisfied, or meal quality is consistent across multiple locations, food service is one industry in need of a ready solution that can address these challenges.
In this podcast, we discuss why AI may be the ideal answer for overstretched restaurants. Specifically, we dive into how restaurants can use cloud vision AI platforms to guarantee only the freshest food is served. In addition, we explore how AI can help transform restaurants into efficient, data-driven operations capable of ensuring meal order accuracy, reducing kitchen stress, and seamlessly integrating with third-party delivery services.
Our Guest: PreciTaste
Prior to joining PreciTaste, Hauke was Director of Procurement and Supply Chain for the major coffee roaster Tchibo Coffee Service, and Managing Director for global chocolate company Wiebold Confiserie.
Hauke answers our questions about:
- (3:40) Demand for AI in the food service industry
- (6:10) New expectations from customers
- (7:59) Integrating with third-party delivery services
- (11:00) Implementing AI into a quick-service restaurant (QSR)
- (15:23) Taking an edge computing and cloud approach
- (19:20) QSRs that have implemented AI-based solutions
- (24:08) Future use cases of AI in restaurants
To learn more about AI in the food service industry, read QSRs: Want a Side of Vision AI with That? and When the Customer Experience Feels Deeply Human. For the latest innovations from PreciTaste, follow them on LinkedIn.
Christina Cardoza: Hello, and welcome to the IoT Chat, where we explore the latest developments in the Internet of Things. I’m your host, Christina Cardoza, Associate Editorial Director of insight.tech, and today we’ll be serving up AI in the food industry, with Hauke Feddersen from PreciTaste. But before we get started, Hauke, welcome to the show.
Hauke Feddersen: It’s great to be here. Thank you for having me.
Christina Cardoza: What can you tell us about yourself and the company, PreciTaste?
Hauke Feddersen: Well, PreciTaste is first to market with visioning AI management for the chain restaurant industry. Mainly fast food restaurants—who don’t like to be called fast food, because it’s not the food that is fast, it’s the service that is fast. That’s why I’ll be referring to them as the QSRs, the quick service restaurants. And PreciTaste, since 10 years has been developing vision-AI solutions to make sure food reaches customers fresher and faster.
We do that by installing ordinary security cameras or 3D cameras in restaurants to see how much demand is there, how many customers are ready to order, how many cars are in the drive-through. And we’re in the back end of the kitchen in the back of house to see how much inventory is available, how many fries are there, how many burger patties, how many chicken nuggets. And then we blend those two together: we know what they need, we know what they have, blend them together, and that tells them what they should be producing right now in order to serve food fast and fresh to customers in drive-through delivery and in the dining room.
Christina Cardoza: I’m excited to learn more about that. But what’s your role in the company and in making AI happen in the food service industry?
Hauke Feddersen: As VP of Operations, I think my job is one of the coolest in the industry. I go in with my team of project managers, and we identify which processes of the customer’s operation can be digitalized, can be uploaded into our edge AI management platform. Meaning, the employees have the know-how of what to do all in their heads, but in the end of the day they’re leaving. So, how can we integrate most of those processes into the computer brain that remains in the restaurant?
And I like to see my team as the paratroopers: we get into the operations, and we find out—with years and years of experience in the fast food industry—we find out how many processes could potentially go wrong and where the best start for AI management for this particular customer is. And those customers are the largest chains in the business; four of the six largest QSR chains are utilizing our technology.
Christina Cardoza: Well, sounds like a very important and big job. You know, over the last couple of years, I think the food service industry was one of the most disrupted or impacted by everything that was happening in the world. They had a slowdown in customers coming to them, and then they had a rapid increase in people coming into restaurants or going through fast food lanes, things like that.
But also, this idea of bringing AI to help manage how people are coming back to the restaurants, people coming back to the food service, to make operations more efficient is a really interesting idea. When you think of AI, you don’t typically think restaurants or food service. So can you talk a little bit about what’s driving this demand for AI in quick service restaurants, or the food service industry?
Hauke Feddersen: I’m a firm believer in the fact that there can only be a solution if there’s a problem. And this industry is facing problems; you’ve mentioned most of them. There is a strong demand for labor that is still unmet in this niche part of the industry where not everybody wants to work, and there is a lot of labor churn of employees leaving from one place to go to the next. And the churn, especially, brings with it the fact that a lot of established best practices, a lot of established know-how, tends to get lost.
But also you mentioned the demand-shift patterns over the last two years, two and a half years since the beginning of the pandemic. They were tremendous, and it is very difficult for operators in the kitchen with very limited data. They basically don’t have windows, they don’t even see what’s happening out there. Their only window into the reality out there is the kitchen information system, the KDS system, that tells them what has been ordered in the past, but they don’t have a system that forecasts what will happen very soon. How many people are there? How many cars are there?
So we are turning restaurants into data-driven operations. AI is incredibly good at solving an equation with almost unlimited variables—whether traffic patterns, historical sales, the sales of the last hour, the sales of the last days, etc.—to really predict that demand better than any human could, and to help the kitchen crew by taking this cognitive load off them and making sure that the individual station that has to hold inventory, hold food, basically just has to do what’s on the screen and follow the screen.
Christina Cardoza: Yeah. In addition to the labor issues that restaurants are facing today, I think there’s also new demands or expectations from customers given the last couple of years—what they expect from their food service or from the quality of their restaurant they go to. And when some of these are changed—you’re talking about, they are having to keep hiring and retrain new staff—I know customers, when they go to one restaurant and then a different location, they expect the quality and the service and the food to all be the same as the first place they went. So, talk about some of the demands that customers have that’s just adding more to the challenges and complexity in the food industry.
Hauke Feddersen: I love one point that you’re raising: customers in this industry segment, they expect quality to be very, very homogeneous. They expect the same quality here that they expected there. That goes for the ingredients, for the food, as well as for the service.
And let’s isolate this one point, the shift in demand pattern. So much more delivery is taking place right now. All of a sudden the customer that used to be standing in front of you is now somewhere, and the food is being delivered to them directly. That puts huge implications on order accuracy. Take McDonald’s for example, the very famous Happy Meal. “Oh sorry, you forgot the Happy Meal for my kids.” “Oh, my apologies. Here’s Happy Meal toy for you.” Everybody happy. But you can’t do that if the customer is 10 miles away and the food has just been delivered. You have to get it right the first time.
That’s why in 2020 we launched a new tool in addition to our QSR brain platform—the auto-accuracy verification, where cameras mounted to the ceiling see what is happening in the restaurant, they see what is being added to that bag, and they can see if the Happy Meal toy is missing or if it has been put into the bag, and they see that the correct bag is handed out the window to the correct customer or delivery driver.
Christina Cardoza: Yeah, that’s a great point. I know multiple times where I’ve experienced going through a drive-through, I’ve had to actually get out because the order was wrong. So anything that makes sure that it’s accurate and gets out the door fast is a win for me. But I’m curious, in addition to the delivery aspect of it, there’s also all of these third-party delivery applications and services out there in the last couple of years—Uber Eats, DoorDash—so many different ways beyond just the restaurant that people can now get their food. So, is that also making things more difficult for restaurant operations, having to integrate with some third-party delivery services?
Hauke Feddersen: Absolutely. There are some great applications that try to harmonize the data streams going into the restaurant. There are some that go exclusively with one delivery company, and there are others that try to play all of them at the same time and use DoorDash, Grubhub, Uber Eats, and all of the others.
One big change that it means for the restaurants that few people see: it changes the customer as it’s perceived by the restaurant. A delivery customer is not necessarily your customer: if you have a restaurant, you don’t know that person. It’s anonymized by the platform; meaning, from knowing your customer and having a direct interaction, owning the entire customer experience from order to delivery, all of a sudden those restaurants are reduced to a part in the middle, and the feedback is prompt and it’s expensive. The refunds to platforms like Uber Eats for inaccurate orders, for “I got the wrong order,” or “Something was missing,” are very, very severe.
And therefore the biggest change that we can make in order to help restaurant crews, stressed restaurant crews—and I’ve spent a good time, all my life now, in the kitchen itself; it is a stressful environment, where it’s sometimes close to magic that the teams are able to churn out the amount of meals in one hour, and have them all delivered, and have them all reach the right customer. The best thing that we can do for them is reducing the stress in the kitchen, reducing the cognitive load, making sure that the processes flow, that inventory is available at all times, and that this very well-oiled machine doesn’t stop.
Our board member James Floyd from Cleveland Avenue, he was 32 years Director of Innovation at McDonald’s, and he stresses this point every time I speak with him. He says, “If you make sure that the job gets easier, then the quality in regards to accuracy will improve automatically.”
Christina Cardoza: I like what you said with the delivery drivers: it now changes who the customer really is. And I think it also changes how the customers perceive their restaurant. If they get a bad delivery they might associate that with the chain itself, when really it may be the delivery driver.
So, going back to what you said in your introduction, “You can’t have a solution without a problem,” I think it’s very evident that there are multiple problems and challenges that the restaurant industry is facing, that AI is coming in to help smooth things out and improve those operations. So let’s talk about how you actually get AI implemented into a restaurant. You mentioned you guys have developed cameras that help with accuracy. What other types of investments or technology does it take to implement AI in the food service industry?
Hauke Feddersen: Important question, because this is a nickel-and-dime business. There is not a lot of money to be wasted, therefore the investments need to be targeted and solution oriented. The KPI improvements must be real for the customers. PreciTaste has been in this business for over 10 years. When we started there were no cloud-AI platforms; there were no cloud-vision AI platforms for a few more years. We had to be extremely frugal about hardware from the beginning on.
And we are strong believers in edge AI; everything that we do runs on small form factor computers like the Intel® NUC in the restaurant. For several reasons: one, because of price. We intend to have the fully fledged solution installed—and the installation in the continental United States, for example, is an important aspect as well, done through our partner network—we try to have it installed for between $2,000 and $5,000 max, including all the cameras and all the edge devices and all the networking kit that is required to make sure that our customers get that KPI improvement much earlier, and that return on invest that they’re seeking when investing in a solution like this. So, very, very little CapEx and software-as-a-service fee to make sure that the AI can continue to learn and can continue to improve.
Christina Cardoza: So, these cameras and equipment being set up in order to keep an eye on everything and alert managers if anything goes wrong, are these existing technology or cameras that the restaurant already has, or to do this do you really need to get a new system, to invest in new technology?
Hauke Feddersen: Very good point. If security cameras are already installed that are TSP compliant, meaning IP cameras, then we absolutely love using their video streams. Vision AI does not need perfect imagery; very few pixels are actually sufficient to run very sophisticated models. Our model is always, what the human eye can see we can teach the computer to see. And then as soon as this data is digitalized, as soon as what we see is transferred into integer values and into bits and bytes, then we upload that into the brain part of our edge AI installation, and that then makes the decisions, that makes the predictions based on what it has seen.
And the role of edge AI in this part is so very important for multiple reasons. One: cost. Cloud-AI platforms tend to be very, very expensive over time. Second: the seamless integration and the low-latency inference that we get from these devices independent of internet; we’re just offline first. Even if you cut the internet, our solution will continue to run exactly like before.
And the third, very important aspect: personal identifiable information, GDPR, and the California Data Protection Act limit what you can do with video signals. And we mount the edge device that captures data from the security camera—ordinary security camera—within a few feet of the device. So the vision data, the PII part, is thrown away immediately, and the only thing that we send over is: there are 6 people waiting to order; there are 12 cars in the drive-through, 2 of them have ordered already. Only these integer values are persistent and transmitted. No PII is in any shape or form saved to make sure that the customers’ and the crews’ data is safe.
Christina Cardoza: And everybody wants this data to make changes as fast as possible in real time. Like you said, if something’s going out with a Happy Meal without a Happy Meal toy, you need to be able to catch that right away so that it doesn’t actually make it out to the customer before putting that in. And so, is that also one of the big advantages of edge computing over the cloud? Is that it gives you these results much faster and much closer to real time than processing it and sending it to the cloud would?
Hauke Feddersen: Absolutely. Super low latency. For everything that has to do with volumetric information of, for example, liquids, or solids that are mixed in liquids—like salsas and sauces, for example.
(On screen: Intel® RealSense™ hardware device)
we rely on 3D sensing like, for example, the Intel RealSense. I’m absolutely amazed by the quality of this very inexpensive piece of hardware. And we can 3D sense volumetric info not for a singular point, but for the entire pan that the inventory is held in. To have scary good inventory information on all ingredients—whether they be pieces, whether they be volumetric, whether they be liquids—we just know exactly how much is present and how much is sold how fast to establish demand patterns just from inventory depletion.
Christina Cardoza: Yeah, and that piece of hardware you just showed is so small and compact that I can see getting it up—it’s not a room problem, and it’s not intrusive or aesthetically displeasing in any way.
(On screen: Intel® RealSense™ hardware device)
Hauke Feddersen: Customers don’t even see that we’re installed in those locations. You mentioned the existing security cameras; if we can’t tap into existing security cameras we’ll just buy the same ones and mount them directly next to it, so that nobody can tell which one is actually for this one solution. It’s very, very invisible.
And in the beginning of each project we always have a passive phase, in which we capture—this can only be a few days, or it can be a week or two—where we capture, how well is the restaurant performing without our help? And then as soon as we switch on our software suite, then we compare to benchmark, and we compare to how good was it before and how good is it now with the AI support. And that’s actually our biggest selling argument of all, to just show, “This was before, and this is now.” Always worked.
Christina Cardoza: Yeah, I think that continuous learning and continuous improvement is really important in today’s modern world. So, we’ve been talking about the benefits of edge computing over cloud; does the company use cloud at all? Or is it strictly an edge-computing approach?
Hauke Feddersen: There are a lot of use cases where the cloud just excels. We have single sign-on to our reporting dashboards with the big platforms, Microsoft Azure for example, that is being utilized by our customers. We make extensive use of BI platforms, such as Power BI, to visualize the data that we gather because the dashboards always go—they have three important aspects that they need to supply. One: information that you didn’t have before; it must tell you something about your restaurant that wasn’t there before. Second: it needs to pitch you against you-two-weeks-ago. This benchmarking against yourself—how has something improved?
And the third part, that I like the most: we are distributed through franchise networks; those are companies that do something extremely similar, but do it in a slightly different way. So, where do we see best practices? Where do we see restaurants that excel, and what led them there? How were they able to improve freshness, improve the sell-through numbers, improve customers per hour, etc.? And have that mimicked on the dashboard. And all of that is happening in the cloud, as well as the AI management; so, redeploying new models every week to our solutions, for that we utilize the cloud extensively.
Christina Cardoza: I’d love to get some real-world examples of how this is actually happening, or how PreciTaste is actually helping restaurants in the food service industry. Do you have any use cases or customer examples you can share with us?
Hauke Feddersen: My favorite one is the Chipotle application that was presented at the Intel Innovation Event on the keynote stage, together with Pat Gelsinger, the CEO of Intel, and Ria Cheruvu. It’s about inventory sensing at the front-of-house makeline, as well as the digital makeline in the back of house that is used for delivery orders. Always sensing how much inventory is present, how fast is the inventory depleting, and then advising the crew on what to cook when, next.
For example, the chicken. And Chipotle runs an amazing operation—scratch kitchen at its finest. Raw ingredients being cut, being spiced, be cooked, being marinated in the restaurant itself. They start with raw avocados and raw tomatoes in the morning to make their delicious guacamole. But, for example, the chicken process—just because it’s so artisanal, so scratch kitchen—takes them 25 minutes from the instruction to the crew member, “Please make chicken now,” to the chicken hitting the front-of-house makeline. So you have to know 25 minutes in advance when you need to restock, and the demand patterns throughout the day vary.
So, lunchtime: a full pan means you have to cook right away; an hour later half a pan means you can leave it there for another 10, 15 minutes, it’s still great food. And please cook something else first, and the chicken only comes in another 20 minutes. So, the AI is just very, very good at predicting what will happen, and helping the crew towards getting the point just right, so they never stock out but they can serve the freshest food that is possible to be served.
Christina Cardoza: Yeah, I love that it’s not only improving operations within the restaurant, but it’s helping make sure the highest quality product goes out to the customers. When you go to some of these quick service restaurants, there’s the stigma that you’re going to be getting food that’s been sitting there all day: it’s stale or not fresh. And this is really making sure that people are getting—the restaurant’s using quality ingredients, and people are getting quality food at the end of it.
And you mentioned you guys use Intel NUCs to get some of this done and to make the edge-computing aspect of all this happen. I should mention the “IoT Chat” and insight.tech as a whole, we are sponsored by Intel. But, would love to hear more about the value of working with Intel and its technology.
Hauke Feddersen: The Intel NUC 12th Generation is a powerhouse in the tiniest imaginable form factor. We’re running two vision-AI streams, inference on two live-AI streams, with 25 frames per second on one of these tiny edge devices in the restaurant. They are extremely reliable. We can mount them everywhere, even if the restaurant doesn’t have a server closet or doesn’t have an a real office, a proper one. And I just really like working with those. Same goes for the Intel RealSense already mentioned. But the best part for us about Intel is it’s not just the hardware house. People think it’s all about chips and silicon, the software aspect, and I’ve heard that there are 20,000 software developers doing a great job at Intel; the software aspect is amazing.
OpenVINO™ has helped us. In the past we were utilizing GPU-heavy edge device systems. So, just one manufacturer making them, very strong GPU force, very little CPU. So, databases, uploads of data—all of that was neglected. And they were quite pricey. Now, through OpenVINO, we can port our models to run on the CPU or the integrated GPU. This just unlocks an abundance of devices that we can potentially use. And, especially in the last two years with the supply chain crisis for digital components, we were able to install restaurants left and right because we weren’t fixed to just one device. We could use every edge device in the market and run our models absolutely independent, whether it has been produced by manufacturer A or manufacturer B.
Christina Cardoza: And talking about OpenVINO, I know in the last year they’ve made huge advancements to the AI toolkit that really just makes these industries become more intelligent, much more accessible, and easier to do. So, PreciTaste—you guys are solving the labor shortages, making sure the food is accurate, all of this stuff helping the restaurant operations. So, where else do you see AI going within the food service industry? Or how do you see this just expanding and improving as time goes on?
Hauke Feddersen: I think we’ll see a lot more implementation: front of house visible to the customer, as well as back of house to optimize processes. It’s about producing more with less: more food with a smaller crew, or with the same size crew that just now does 20% more. And we’ll see the technology being specialized for those kitchen applications that will excel in delivering multiple different food items from one kitchen. So you can have Mexican food paired with Italian food, paired with sushi, paired with wings, delivered with one driver—so that everybody around the table can eat and enjoy the food that they wanted and they won’t be limited to having to pay the delivery fee three times or all get the same.
I think that this industry will get a lot more precise about predictions, and it will work very hard on eliminating what you’ve just stated before—the very stale food that some people still associate the quick service industry with. It will be fresher, food can be more interesting, it doesn’t have to be optimized for shelf life if you have a system that manages the shelf inventory. And it can be optimized for greater taste and greater customer satisfaction. So, I’m—personally as a consumer—very excited about that part as well.
Christina Cardoza: Yeah, absolutely. Well, this has been a great conversation. Unfortunately, we are running out of time, but before we go I just want to throw it back to you one last time. You have any final key thoughts or takeaways you want to leave our listeners with today?
Hauke Feddersen: Well, I’ve listened to a few of your blogs, and a lot of the topics sounded very, very science fiction-y, and the part I like the best is it’s not—this is not science fiction. This is the technology that is out there today, and it’s improving the lives of customers already, unbeknownst to some of them; you don’t even know whether it’s in. What I would love to see in the future is maybe even like a badge outside a restaurant or in your favorite delivery app just saying, “Hey, this is a restaurant that optimizes quality utilizing state-of-the-art vision-AI technology. So don’t hesitate, don’t worry, you can order and you’ll get the best quality from this location every time, because it’s managed by a system that is entirely designed to do just that.”
Christina Cardoza: Yeah, I agree. A badge would be great, because you go to those restaurants and you see they are being powered by AI, you know that you’re going to get quality service, you’re going to get freshest ingredients, it’s going to be a smooth visit. So, absolutely, would love to see that in the future also. So, I want to thank you again for joining the podcast today.
Hauke Feddersen: Thank you very much.
Christina Cardoza: And thanks to our listeners for tuning in. If you liked this podcast, please like, subscribe, rate, review, all of the above, on your favorite streaming platform. And, until next time, this has been the IoT Chat.
The preceding transcript is provided to ensure accessibility and is intended to accurately capture an informal conversation. The transcript may contain improper uses of trademarked terms and as such should not be used for any other purposes. For more information, please see the Intel® trademark information.
This transcript was edited by Erin Noble, copy editor.