Saima Khan on Nutrition AI, Patient Meal Accuracy, and Safer Healthcare Food Service

Russ Fordyce 🏆 (00:00.984) Welcome back to the Winter Circle. am here today with Saima from Compass Digital and the Compass Group. They are blending AI in the healthcare environment to really go after some unique problems. Problems as I see them and problems that I've experienced when I've been in healthcare situations in the last few months. So Saima, welcome to the show and congratulations on the AI Excellence Award. So yeah, now you're in this digital arm of Compass Group and doing really cool things with AI. Saima Khan (00:22.286) Thank you. Thank you. Appreciate being recognized. Russ Fordyce 🏆 (00:30.732) Walk us through a little bit about Compass Group and this nutrition AI that you built. Saima Khan (00:35.48) Yeah, absolutely. Thanks for the opportunity. So Compass Group, North America, Compass Digital is the technology and innovation arm for that group. So we're responsible for supporting all of our sectors and our business partners with hopefully providing the best there is to offer from a technology, AI and innovation standpoint. So this particularly, we're really proud of this. This is something the team has been working on for several years. It's up and coming technology and we've been working, looking for different partners actually throughout the world to be able to ensure that we leverage the best. this is computer vision technology. So think of a, a kitchen back of the house where trays are being built for patients. There's a line and at the end of the line, that's where you're going to verify from an accuracy standpoint. So super important in a healthcare environment. We want to make sure that there's all kinds of patient safety issues, allergens and whatnot. This helps us identify what's on the tray that's being about to be loaded into the cart. Is that what the patient ordered? Is it matching up with a ticket that's linked into the EHR, so the electronic health record? At a high level view, that's what it does. It takes AI, it scans quickly what's on the ticket, what's on the cart. patient safety, patient accuracy, and that's what it's looking for. Russ Fordyce 🏆 (02:05.666) Now you've been in technology yourself for a couple of years now. Yeah, and it's really interesting because you're now at this intersection of kind of two less than thrilling technology industries, food service and healthcare. Healthcare has got tons of technology, food services seems to be coming along, there seems to be a lot of innovation there. But it's a very unusual combination together. How did you end up there and what keeps you in it? Saima Khan (02:09.678) Just a few. Saima Khan (02:33.464) Yeah, I mean, I've been in healthcare majority of my career. So I, it resonates with me. I like what we do to be able to use technology to, you know, assist with healthcare in those settings. Food service was really interesting because I've always been in clinical technology. So think about technology to enable the providers, the healthcare. workers, clinical staff. So this was a really unique opportunity that came up five years ago to be able to look at everything else that happens in the hospital outside of core clinical care. And that's what we do at Compass Group with food services and support services. So it was just a really unique opportunity for me to be able to holistically have that view across a hospital and run the operations using technology. So. Yeah, it's been a really no day is where I'm like, wow, I'm bored. What are we going to do? There's a lot of cool challenges day in, day out. Russ Fordyce 🏆 (03:37.484) Yeah, and now this technology, this nutrition AI you built, super critical. People don't realize, I think, how important food is in a hospital environment. I witnessed it myself with some family members recently in the hospital. we always joke about hospital food, right? And sometimes it doesn't taste the best, but it really is medicine for folks. And so you're out there kind of looking at that. Saima Khan (03:42.404) Yeah. Russ Fordyce 🏆 (04:03.65) that plate or that tray coming off of the line, making sure that it's correct. How do you explain to people why this particular moment that that final tray check is so critical in this kind of, in the healthcare path? Saima Khan (04:16.92) Yeah, I mean, to your point, food is medicine, right? And I think in a health care environment, we're often taking care of patients that are there that need the food to be able to recuperate from whatever is going on with them. So first and foremost is patient safety, right? We have to make sure that we're always, I mean, that's not something where you can say the target is. you know, 90 % of the time. Now the target is 100 % of the time we have to ensure that we are only serving what we should serve to that patient. There's all kinds of dietary restrictions, right? You know, based on whether you're a diabetic patient, cardiac patient, there's so many different types of diets and this is just that added check outside of the human element, other verifications that we can put in, high-risk trays. you know, allergies, all kinds of, there's just so many different variables there. So this, there's the patient safety aspect, right? We must, we have to, right? There's no, there's, you have to get it right. And then there's patient satisfaction. If you're in a hospital, potentially the best part of your day could be the meal period, right? There's not a lot that you have control over. This might be the one thing that you look forward to. So there's the patient satisfaction element of it too. We're a hospitality company as well. So we want to make sure that what you ordered, that's what you get. So when you're getting your meal and you're really looking forward to that meal and it comes and something's not right on it, something's missing, ultimately the workflow ends up, probably call the nurse. We don't want to affect the nurse and their workflow. We want to make sure that they can focus on clinical tasks. They'll call down. to the kitchen staff and then that workflow, then there's the food waste of it. And then just the patient not being satisfied. And then there's the element of potentially they need their meal before they can have some sort of medication or go for a procedure. So there's that whole workflow. So it's not just food, I guess, right? There's the patient safety, there's the healing aspect of it. Saima Khan (06:32.824) there's this might be from a satisfaction standpoint, the number one thing that they look forward to. Hopefully they're not a long-term patient, but as the length of stay increases in the hospital, you know, they might get to know the staff as well. They might get to know their, you know, patient dining associate. So there's so many elements to it that, that we're providing outside of the technology and food services that goes into keeping the patient safe and happy. Russ Fordyce 🏆 (07:00.396) Well, what amazes me is you guys seemingly revolutionized this kind of category and problem with AI almost immediately. I know you've been working on it for a few years, but healthcare food service has been around for decades. Why was this meal tray accuracy problem such a hard problem to crack before AI? Saima Khan (07:20.92) Yeah, I mean, I think you could put in all kinds of processes, right? We have patient dining software, proprietary, best-in-class software that we built in-house that I would say probably from a, you know, not just because my team builds it, it's pretty, the capabilities are strong, but still there's that human element of it at the end of the train line. We can, you know, indicate on the ticket by. color, all kinds of different visuals, that high risk trade, this needs to be double checked, you can have a supervisor come in, great processes and workflows that have been implemented, but just that added element of having the AI assist and in flash just indicate for you, hey, red, that there's something here that doesn't necessarily match the ticket. I think just that added element of having AI. help do the work and then think about just how AI works over time. The models train, it's getting smarter, smarter, smarter, smarter as we're rolling this out and it's just starting to get the speed and the accuracy. And then one of the elements with this, there's always a little bit fear. We don't wanna just introduce technology for the sake of introducing technology. One of the concerns was when we were working through just the whole workflow, is it gonna slow down the time? right from when the tickets printed, they're plating the tray, it's going on the cart, and then it's going up to the room for delivery. Temperature is super important too, because nobody wants cold food. So we're often thinking about timeliness, temperature. So we wanna make sure, and obviously accuracy. Can this help solve that? One of the things that we've noticed, one of the really exciting parts about this technology, it's actually increased. I'm sorry, decreased the time it takes from getting it, the ticket printed to cart up to the room. It actually reduced the amount of time with all the manual checks. So again, from a patient satisfaction standpoint, they're getting the food faster. It's warmer and it's accurate. Russ Fordyce 🏆 (09:31.862) And what I liked about your nomination and all the data that I read was that you haven't completely eliminated the human, right? If the system kind of flags something, it's actually human that comes over and kind of makes the final call. Why not just automate that correction and have it run through again? Saima Khan (09:50.893) Yeah, I think that, you know, at this stage, obviously this is technology that hasn't been used in the setting before. Potentially we'll get to that, but yeah, it'll flag it and it'll have a human come in and verify. I would say most of the times it's right from an accuracy standpoint, there's something missing or there's something that shouldn't be there. But yeah, you still need that human element to come in and, you know, switch out whatever's missing or add what's not missing. But I think. It's continuing to evolve. have more more use cases that we're working to work on this. But yeah, at this time, it's a great partnership with our kitchen staff, as well as the technology to be able to ensure that we're having, we're bringing the best of both. Russ Fordyce 🏆 (10:34.21) Well, the other thing that I really liked about what you did and how you did it was the validation that you got. You ran a side-by-side pilot with kind of the old way, the control way, and then this new AI tray line simultaneously. That's a little bit more rigorous than most technologists will do for a pilot. Who pushed you into that level of rigor and why? Saima Khan (10:54.36) Yeah, I we have great business partners. We have great partners at some, you we work with some of the best healthcare systems that you can think of in the country. So we hold ourselves to a high standard. They hold us to a high standard and we're really, really meticulous, I would say about why are we introducing technology? What are we doing with it? It's definitely not for the sake of just saying, you know, hey, cool. factor. So we were all pretty much in line with, if we're going to do this, we're going to pilot, we're going to invest here. This is going to be industry first. We want to be able to prove the ROI, the benefit, and it's resonating with our clients, our partners as game changer, industry first, and where you guys are going with this. yeah, was, I hear you, right, about being able to say, OK, it's cool and it works. But we were definitely all in sync. with we want to make sure that we can prove the ROI. Russ Fordyce 🏆 (11:53.43) Yeah, well, now in that in those initial kind of pilots and trials, and I know all the numbers are there, like you've talked about it short in a cycle, and it's going to increase accuracy, it has to. But outside of that, what did you observe any kind of specific shift or a specific team or a, you know, kind of a staff reaction that told you this was this was real, and it's going to work? Saima Khan (12:14.53) Yeah, I mean, I think a lot of times when something's new and groundbreaking, there's going to be some sort of apprehension. Well, how's this really going to work? it? There was a lot of fear in the beginning. As I said, that it's going to increase the time, right? You're adding this technology, but ultimately, if it's taking longer to get a patient meal on the cart up to the floor and to the patient, then really not sure that we can. find the benefits from it. I think that was pleasantly surprising to me and a lot of other folks as well that we've talked about this that, OK, it's actually reducing the time. And then also, more from a patient satisfaction standpoint, it was nice to see that we were able to catch things early on before it went up there. Some of these are really large campuses, if you think about from a hospital standpoint, for us to be able to. Okay, you know, there's, there's a salt and pepper, right? There's think condiments, things like that missing and it can flag it pretty instantly. So to be able to see the value right away from that standpoint into the three things that I talked about, temperature, timeliness, and accuracy to be able to say, check, check, check. That was satisfying. Russ Fordyce 🏆 (13:33.068) Now, the middle section of that process, getting from the kitchen or from the cook into the patient's room, you have a lot of kitchen staff in there. And generally speaking, I've met a lot of kitchen staff. They're not real technology gurus, and they're not really interested in it. They're interested in serving food and helping people. Did you have any adoption challenge inside of the kitchen, and how did you address any of that? Saima Khan (13:59.725) Yeah, I mean, of course, right? I think that we definitely have champions, I'll say, that helps along the way that you'll find an early adopter, someone who's really interested in the technology. And we're really lucky to have partners in a lot of our health care systems that want to be those champions and they want to try out the new technology. But then, there's of course, hey, I've been doing this for 30 years. We do it well. There's a lot of pride in that. And this was This was more about how do we complement the workflow and the processes and the great just, you know, just these are systems that have been built over time. They work great, right? There's a reason that we're the number one provider for, um, for this in the country. And so, um, it was more about this is going to compliment what you're doing. And, you know, ultimately from a standpoint of giving you the ability to focus on. potential other tasks as well if there's not multiple people at the end that are responsible to make sure that we're getting the right trades out. So yeah, there's always that, but we spend time on site with our operators. We want to make sure that we give them the support, the training, the education. there's a lot of, for the early adopters, there's a lot of that that we did. Russ Fordyce 🏆 (15:21.004) Now you guys are known kind of in this healthcare space, obviously you're SVP of healthcare. So we've got a, you guys are very narrowed in on this one vertical as well as others in your business, but you've got this kind of broader platform that's bigger than just nutrition AI. I wonder, is there a design principle that kind of guides you and when you're talking about these frontline healthcare workers and the patients themselves that they depend on, what's the kind of the design principle that you're kind of you've refused to compromise on. Saima Khan (15:51.609) Yeah, ultimately everything we do is about patient satisfaction. We're in healthcare, in hospitals, patient is the reason that the healthcare system is there to be able to serve them. So everything revolves around the patient, every technology that we're looking at, whether we're gonna build it in-house or we're gonna partner with third party, it comes back to patient satisfaction. Our operators. those are our frontline workers, our patient dining associates, or the folks that are on the ground doing this. Everything we built is in partnership with them. We don't sell our software commercially, it's for them. We say everything is built by our operators for our operators. So we spent time on site. So that's one piece of it because they're the ones who are interacting with the patient. And if we can get that part right, that's going to lead to patient satisfaction. Then there's the patient themselves. Then there's people we refer to stakeholders as caregivers. That's you and I. We might be providing support and care for our loved ones in the hospital. So they're interacting with the technology too. So we're considering them. And then there's obviously the hospital employees, clinical care and non-clinical care staff. So all of these people, they're interacting with what we do. But ultimately, they will lead back to the patient satisfaction. So that's kind of our North Star. That's what we look at. Hey, what's going to be the impact here? Whether it's through our operators, caregivers or hospital staff, how are we ultimately affecting the patient satisfaction? Russ Fordyce 🏆 (17:32.408) So you've kind of solved this one problem, which was a huge problem. I think we've all dealt with it. If you've been in a hospital environment, something's missing, something's messed up, you're recalling. And just like you said, you hate to do it, but you're calling the nurse. And that's just a total waste of time. I know you guys are looking to expand this platform into some other areas. Tell us about that. Saima Khan (17:46.403) Yeah. Saima Khan (17:53.251) Yeah, so this was definitely the first use case, but we're not done here. There's a multiple other horizons. Malnutrition, right? That's a huge use case in a health care environment when it comes to patient feeding. That's something that's very prevalent. So one of the next use cases that we're looking at is usually what will happen is a nurse or some other staff is eyeballing. how much food was consumed by a malnutrition patient. And they're just saying, okay, 50 % was consumed, 10%, it's very manual right now. Then one of the next use cases we're working through with our partners is for, there's at the beginning, at the tray line before the meal sent out, obviously we're able to get a view of the tray from an accuracy standpoint, but we're also able to get a view of how much food was on the tray. And then post meal consumption, as we're picking up the trays for the malnutrition patients to quickly able to scan it to say very definitively, 50%, 10%, 30 % was consumed and built a workflow around it to be able to send an alert back to via the EHR, whether it's to nursing staff, dietician, other clinical care workers to say, hey, it's been three meal periods. this patient, they've consumed, you know, whatever the threshold is at that hospital, 50%, someone needs to check in on them, right? Or we could be getting to a point where we have a issue. So accuracy from malnutrition. Russ Fordyce 🏆 (19:29.71) Well, that makes total sense. It's because you're looking at it now from creation to consumption. And the whole idea is to get these people fed. And as you said, the food is medicine. And so you're closing that loop on, was that meal successful? Not only was it successfully deployed, it was successfully consumed. You're doing what you're doing. What really amazed me is you guys are doing this already at massive scale. The example that you provided in the nomination was 600 million meals a year for one of your customers. Walk us through kind of how operating at that scale is kind of, can be constraining, but you guys seem to innovate around it. Saima Khan (20:14.498) Yeah, well, so through our patient dining software that we have that we built in-house, and this is a component of that, this what we could refer to as nutrition AI, through that is where we serve over 70 million meals a day via the tech, not a day, sorry, a year through that. So we have definitely the opportunity to be able to at scale, you know, to be able to make impact. I think it's just been a really great partnership that we have with our partners that are on the ground, our operational teams, and the corporate teams that we work with that help design the workflows, right? So you can have the coolest technology and you can go plop it into operations and it's not going to do anything. So we're very meticulous about the end-to-end flow, how it's going to impact the operator, the patient, and we work tirelessly on it with our our partners that we have at Morrison Healthcare, who's part of the award that this went to. So yeah, mean, a lot of it is, we're not scared from an innovation standpoint to try different things. We don't see things as failure that you learn from it. And there's always going to be something that you can apply to the next one. So it's a lot of trial and error and working together to make sure that we continue to, it's never perfect, right? We're always modifying it, modifying it, modifying it. Russ Fordyce 🏆 (21:41.997) Yeah. Now, one thing in healthcare, and maybe you can give some insight on this, is we've heard a lot of terms around kind of augmented intelligence. Instead of AI kind of replacing all of our systems and processes, it's kind of AI that supports the human and the human decision-making rather than the AI doing it all. Where do you see the boundaries of that approach? Saima Khan (22:05.292) Yeah, I don't. I mean, I don't see in our industry that AI is going to ever replace the human part of it. I think what it's going to do is surface to them what they need to know as opposed to going and digging for it. Right. That's how that's how I see it. We're to your point, we're augmenting it. Right. We're not we're not trying to replace the human element of it, but we're trying to save them time to be able to, you know, we're in a clinical care environment. There's so many other value add tasks that we can do. So if you can, if you're I'll give an example. mentioned there's the 70 million meals that are going through our software. Think about the data that we have on what people are ordering, but that's one data point, right? This is what people are ordering with this software. One of the use cases that we have in progress is also the consumption part of it, right? So just because someone ordered the meatloaf, I don't know why meatloaf is always a great example, just because they ordered it doesn't mean they consumed it. So this will help us see at the end of it when you take a picture, they ordered it, but they didn't consume it. So the next day or whenever they're looking at the chef is looking at from a plan standpoint, they're looking at what should we put on the menu? We're using the AI and the data to surface to them the recommendations on what should go on the patient meal based on not. just what was ordered, but what was consumed. So it's all about, I think, using AI to surface smart recommendations back to them to be able to make quick decisions rather than sifting through the data. That's how we're looking at it. Russ Fordyce 🏆 (23:44.665) Now we have a very broad audience, obviously spans more than healthcare. You have this unique experience though of a massive AI implementation. So if I'm somebody in logistics or manufacturing or even hospitality, kind of like you're doing, what's the one lesson from your pilot that applies kind of everywhere? Saima Khan (24:01.57) Yeah, I think that we definitely, when we wanted to do this, had some thoughts on what the impact here was going to be, what the workflow was going to look like, how we were going to go about planning it. I think you just need to know that you're going to have to make some shifts in what you thought. And it's little things about even the hardware, right? How we've applied it within the tray line. You might need to, we've had multiple iterations of actually of even the hardware to be able to say, OK, well, this first beta version wasn't exactly the right model. So you're going to go through multiple iterations. And then you're going to come up with use cases that you didn't even think about as you're doing it. And some of that's going to come from the operators, the frontline workers that are using it. One thing we didn't even think about it, and we've rolled this out recently too, Why not when you lift up the tray cover anyway at that point to scan it? Why not check to see what the temperature is? So thermal imaging is something we've added as well in the past several months. that came from, again, that collaboration with the folks that are using, OK, yeah, we should do that. So again, we're going to continue to. iterate and what you thought you were going to do, it's probably going to go well beyond that if you're open to it and you listen to the people using it. Russ Fordyce 🏆 (25:31.234) Well, Saima, this is absolutely thrilling. mean, the amount of complexity that you've minimized with the AI is huge. I can't imagine trying to do this with human people. You had to save time. I really look forward to seeing kind of those next advancements. I think there's so much opportunity there to get down to even like a, know, patient specific menu every day that's, you know, custom designed for them based on their needs and preferences. There's just so much opportunity there for you. So congratulations on the award and all the success that you guys have had. Saima Khan (26:01.093) Thank you. We appreciate it.