Alex is the Co-Founder and CEO of Nabla. Nabla is an all-in-one digital solution with an AI-based assistant for healthcare to help free up healthcare providers to do what they are trained to do. That is spending more time with their patients.
Alex is a pioneer in the AI/ML space. He has had two very successful startups acquired by Nuance (VirtuOz) and Facebook (Wit.ai) respectively. At Facebook, he worked on their AI team developing conversational AI.
If you are a clinic or healthcare startup, check out Nabla or email Alex at email@example.com to set up a meeting and discuss your use case. There is no reason to build something from the ground up when you have such a great resource!
Zain: We got Alex from Nabla here. I'm really excited for this conversation. Alex, thank you for coming on. For those who don't know you, would you mind giving us a little introduction?
Alex: Hi everyone. Yes, sure. So my name is Alex. Alex Lebrun. I started Nabla three years ago for Nabla. I started two startups in the machine learning space. The first one was acquired by Nuance Communications in 2012 and the second one was acquired by Facebook in 2015. So after that I spent three years with Facebook at Facebook AI research, working mostly on conversational AI. And three years ago we left Facebook Research to apply what we've learned to ask care of Naba.
Zain: That's awesome. That's amazing. So what can you tell me about what is Nabla? Can you just give us a little background about what you guys are building with Nabla?
Alex: So we are building AI based assistance for doctors. So we all know that doctors waste a lot of time doing things that they shouldn't be doing and that they don't have enough time for us patients. And so this is a problem we are trying to solve with Nabla. So for instance, if you have a consultation with your physician, let's say it's a remote consultation, they would look at you in the video and on the side, the Nabla assistant would take care of many things that they hate to do. Points on the clinical documentation, writing the report about the consultation, updating your patient records, programming the flood review, maybe writing some prescription, doing some stuff called coding or insurance. And so we are trying to automate all these things so that doctors can spend more time on the actual care and empathy with the patients.
Zain: All of that sounds amazing. I think that's what I mean, what you're bringing up is what I think like AI and ML would be like great for me personally, the connection with the patient is what's the most important in healthcare. And you don't want to replace that, you just want to augment it. And it sounds like that's what you guys are trying to do.
Alex: Yeah, it's funny because I've seen some companies trying to replace doctors with health care professionals with chatbots or things like that. And I spent the first ten years of my life building chatbots for customer service very early on, before it was a pool. And I perfectly know the limits. Chat bots can be very powerful in some context, but I also know the limits and healthcare is typically a setting where I don't think chatbots a good idea. Okay. Anyway, we are far from being able to build this today, and even if we work at all, it is really a good idea. Not sure. So the idea, but actually the technology to build an assistant that makes the Fort life easier, it's very close to technology. Build chatbots. It's the same models that are listing.
Zain: Yeah, no, for sure. What is the back end of novel. From what I understand, it's kind of like a CRM. Is that correct?
Alex: Yeah. So once we say we are building an AIbased assistant, the question is how do you bring these products to the healthcare providers? And so we started by building digital care platform that really looks like a CRM. Like it's like a modern CRM.
Alex: It handles both asynchronous communication with patients, text, and also synchronous communications with video consultations. And it has everything you find in a good CRM system. So to remember important things about your patients, creating some tasks and collaborating as a medical team, because you may have different kind of nurses, specialist nutritionists, working together on one patient. So all these things are obvious if you are in CRMs, but they are not obvious at all in the healthcare provider world today. And so we built that and then added our machine learning assistant in this tool so that it comes with the AI assistant doing the documentation and other tests.
Zain: That's awesome. I even wrote about it. I think CRM, if you have like a back end of the CRM, it's like the perfect tool for health care. Because like you said, it provides you everything, right? It can write all your notes. And the point of a CRM is as many touch points as possible with your customers. But in our case, it's the patient. And it just makes life easier versus what we're doing now. It's like we have all these other systems kind of coming in together and none of us really work well. So I've always wondered why in the beginning, EHRs were just not CRM and then people just built off of that because it would have made our lives so much easier. At least on our end.
Alex: I have the answer to this question, I think. So EHR are not designed it around because EHRs were designed to build insurance payers. And this was the only goal of EHR. It's around coding because documentation to support your claims and how you code them. And so it's really a financial goal or anything. And then you try the plug eventually a few things for patient relationship, but it's not the main this is not the main goal, the original goal. So somebody should build an EHR. Certainly many people tried, but then it's difficult to have people change their EHR process.
Zain: Yeah, I can tell you using, EHR? I mean, they are. That's why they were built, right? That's the genesis of the HR epic. All them were built for billing. And I can tell you using it, using them, it definitely feels like they were not built for patient care. We're built for billing. And there's a lot of issues that come along with it, which we don't have to really get into that, but that's amazing. So where do you see AI and ML in healthcare kind of going like what do you see as like the biggest roadblocks and then how would you get across those roadblocks?
Alex: There are lots of applications. There is maybe the first real world problem that AI solves in health care was around imaging, analyzing x rays or ECGs or other because this is very close to the original playground problems that researchers worked on in the it's a surprise that it started with these kind of pattern matching problems and now it's totally in production. There are FDA clearance for the augmented Xray. And so this was I think the first wave of ML impacted Escare. And now we are entering the second wave where it's less obvious. In health care, I think 80% of the product is communication between the patients and the medical professionals. And so if you really want to impact healthcare delivery, you need to be involved in this conversation. And so this is where conversational AI comes into play. How can you be helpful in this conversation and automated parts of it without replacing the doctor? And there are many challenges in that. I think the first one is doctors at very high expectations. If you say I will help you with my software, it really works. You don't have three shots at the problem. If you're wrong twice, then you're out. And so this is one example of but on the other hand, you need data to train your models, to improve your models. How do you break this chicken and eggs problem?
Zain: Yeah, no, for sure. I think one of the reasons why healthcare has just been so resistant to technology in general is that every iteration of new technology that comes in and is kind of presented to us or in some cases shoved down our throat, it ends up adding more work to our pile. Right? So for example, like EHRs, right, they were supposed to revolutionize everything. They were supposed to make everything easier for us. But what ended up happening was most places kept the way that we were doing it before with paper, like all this insurance stuff, all that stuff. So you're doing the paperwork and then you're also documenting inside the EHR. And on top of that, you're doing other things that you can't do. Then you have to figure out workarounds because you're getting stopped and this and that. So I think that's one of the biggest things in healthcare is that I think healthcare providers at least I mean, just for me personally, I love technology and I love things that you guys are making. I love AI ML. I think there's a huge future in it. But to your point, not everyone is like me. Not everyone is. We've been burned so many times. You've heard that saying pull me once, shame on me. Shoot. Shame on you. Pull me twice, shame on me. I think that's where health care is. And I think eventually it's going to take some time to get to that point. But we need good people like you, with great intentions that are building a good product to stand above all the other. Kind of like a couple of years ago, people are getting paid hundreds of millions of dollars for an idea rather than an actual physical thing. And I think what we have to kind of go through the trash and kind of lift people up like you guys, no, blend them. And hopefully we can get to that point and get gained that trust. But gaining that trust initially is going to be the hardest part. Like you said.
Alex: Yeah, as you said, we are not starting from zero, we are starting from ten because it's burned so many times and the first 50 years of computers has been bad thing for healthcare professionals. I agree with you. The other day I was visiting a clinic and the staff was really happy. It was like a special day, like Christmas. And then they say yeah, the system is down today, we are just using paper. And they were celebrating. It brought many new constraints without solving their problem. Again, it solves, I think, more like the payoff problem than not the patient problem, not the provider problem. And we hope to reverse this trend just as many startups now I think are. Maybe the level of machine learning has achieved recently makes this finally possible. But we have to be careful when we when our machine learning models make some suggestions to the healthcare professionals. We have a threshold below like 0.5 threshold of confidence. This suggestion will not be displayed. But we realized that because at the office we work with positions and same desk. So we see what they do and we learn from that. If a suggestion is wrong after just three or three or four times, just ignore all the tools, any suggestion so you don't have so many shots to be good. And we had to raise the threshold of confidence to 0.9% otherwise we would lose the user after just 24 hours. And then there is no chance, no opportunity to come back because they just ignore and they are trying to have so many tools on the EHR every time they do something there are like ten red windows popping up everywhere. This medication just ignore everything. They're trying to do that. And so you don't want to become the 11th window, but you should be very careful. So we learnt many things like that at the beginning.
Zain: Yeah, alert fatigue is a huge thing. We get so many pop ups here and there and a lot of times we're just so used to blowing through them and unfortunately, sometimes that leads to bad things. But when you were talking about the confidence thing, 0509, what do you mean by that? Is that linked to a percentage point or is that like in your AI model?
Alex: It's in the AI model, so it doesn't mean much actually, as you know, machine learning is very empirical. Things work, but we don't know how they work. Exactly. And so I'm very compatible with the health care mindset where you shouldn't kill someone and if it's not proven that there is a risk, you shouldn't use it. So at the opposite of how machine learning works, where the model tells you something, it's very hard to we have a confidence score, but the reality is we don't know, we are not confident about this confidence score. And so it's very difficult to reconcile that with the requirements of health care where you shouldn't do random things.
Zain: Yeah, no, for sure. But like you mentioned, for machine learning to be better, you need the data. But in health care, it's hard to get the data because the door isn't wide open. Right. You can't just walk into a clinic and be like, hey, can we just use your data? There's privacy laws on top of. And then also just this apprehension from the physician or whoever it happens to be. For me, machine learning and AI, I think your vision of machine learning and AI I think resonates with me because that's what I tell people to. Like, I want robot. I wrote something about robots. How would I replace myself with a robot? But the point of that was I want robots, machine learning, AI, all these algorithms to replace all my mundane tasks, all the things that are so algorithmic that it's a yes or no.
Zain: That's not a gray area, it's just going through. And then it allows me to freeze me up, to talk to my patients and really do the things that I was trained to do. Right. And that's what I see. And it seems like that is where you guys are going. Is that accurate?
Alex: Yes. Usually we don't do things in the background automatically because if there is the slightest risk of error, it could be very bad. And so we work a lot on the user experience in the UI so that everything the AI assistant does is to be validated by the healthcare professional, by the user. And of course, we use these validations or non validations as a signal to better train about. But we don't make an update in a patient record silently. And the other thing is, we keep the source of every update we do. So if it's something and you want to know why this information is in the record, or why this sentence is in the consultation documentation or summary, and you can click somewhere and get the excerpt of what happened really during the consultation, where this information comes from. And this is very important to build trust with the users. And actually we observe physicians working and even in the old school paper record about the patient, when they see something, they are always suspicious, like, who did that four years ago by another hospital? And I don't trust these guys. So I will ask them this question or check again. So even in the old world they are always very suspicious of information that is already there. So you need to trace to the source.
Zain: Yeah, no, 100%. I mean I can tell you why because a lot of times people will. Well, at least in the modern EHRs I don't have too much with the paper records but in the modern EHRs and stuff, everything is usually copy and pasted over and things get overlooked because there's so much documentation that needs to be done. So I like to call it healthy skepticism. It's not that we don't believe you, we don't trust you. It's just that we need to make sure that what is there is actually happening because we all understand that we're all really busy and so on and so forth. Let's say I'm a clinic or something and I'm excited about Navala. How do I get the process started? How long does the process take and how does that work?
Alex: Right now it's used for remote communications. It's very simple. You will go to Nada.com, sign up there, it's automatically free of charge. And then you would have this little care tool that's supposed to be very easy to use. If you find it's not easy, call me with a layout that is very much like a CRM for like Intercom for instance would be or other than that. And so from there you could for instance create a link share with your patient remote configuration link, like a zoom in or you can write a message to your patients. You can use this as a self service tool. So as a physician or product practice you could just use that. And then for digital health startups or digital providers, we already have a patient at the website, we have a set of APIs and SDK. You can embed this experience into your product. Of course you don't want to see nabla or to use something on the side. So we have SDK for iOS, Android, Web, React, native and so on so that you can add messaging or video to your existing system and on the provider side, benefit from this AI system.
Zain: That's amazing. And then what are some of the AI assistant features that come out of the box?
Alex: So the most important is documentation, clinical documentation. The goal is any information that is relevant that is shared either mentioned by the patient, like I am allergic to poem or by the physician I understand you have this illness, so I will prescribe you this. So all this information will capture and normalize in ICD Ten and then put in the right fire. So we use the fire standard. This is basically a foundation use case of documenting every contribution with this.
Zain: Awesome. And is that just something that gets transcribed via a video call as it's happening or is it something that you can do afterwards? I can just talk into something. How does that work?
Alex: It's transcribed in real time and we do that over messaging as well. I really believe in Asynchronous care. I think right now 99% of computations happen over in person or video synchronous video. But I think asynchronous is very important to many, many patients and so we do the same tasks over text. And when you use texting, summarizing is very important. If you interact with the patient through text every few days, you cannot reread all the conversations from start to start, every time you need to summarize what was saved, what is important, which is what we try to do. We also make messaging more efficient on the provider side by trying to suggest the next message we generate based on the pictures, on the context and many things. And they would typically edit it, edit message before sending it, but it's often very accurate, what we call copyat. And the goal here is to make the coaches owners who are doing messaging much more efficient.
Zain: Awesome. And is there a way to set up like a template of like, you know, if you get the same question asked over and over again, is there a way you can just create a template and just be like okay, when we get this message, boom, we just send this.
Alex: You don't even have to do that because the system will create a template automatically. So if you enter twice the same thing, or three times even once, every basis is a template and then if something is relevant in this context, it will pop up or acting like a template that you don't need to manually create a template before.
Zain: And I want to get touch on asynchronous care. So the more and more I think people are getting into digital health and digital health is becoming a thing, all these terms are being thrown out like asynchronous care, all CRMs, all that stuff. But I think you're absolutely right, asynchronous care is the way it's being done right now. The majority of care is asynchronous. You see your provider for 5, 10, 20 minutes, what, a year, and the majority of your care is usually done outside of the walls. And that's another reason why I love AIML these kind of things because it allows us to be everywhere all the time, if that makes sense. And a majority of care is asynchronous patient calls us, they leave a message or they send us a message and then we have to figure it out and we send them a message, call them, usually leave a voice, we usually just pingpoint phone tag or message tag, whatever. And I would say like 95% of care is asynchronous. So it's just that for us it's just normal care. And I think you guys are coming in calling it Asynchronous. It just now has a name to it. So I completely agree with you that asynchronous care is the majority of the way care is delivered at least in the United States, I'm assuming around the world as well.
Alex: Yes, I think in real life you're right asynchronous care is the biggest part. But in in proper consultations are mostly synchronous. So there is a range there. And also most of the players work on a fee for service system where only synchronous care is actually pay for interstate or video competition. But they won't reimburse you because you answered your patient text message.
Zain: Yes. And that's the thing that when you're working in the clinic that's I shouldn't say frustrating, it never goes through our mind, right? We're not like, oh, we're not going to get paid for this call, we're not going to get paid for this message. We're only getting paid for the people we see. We're physically like physically seeing them touching. I mean there's also already like people talking about cutting clawing back on virtual care because that's technically not in person, right? So there has to be legislation changes. But I would say the majority of care is asynchronous and the majority of the things we do in our clinic as providers, we don't get paid for and most of us are staying late, most of us are doing all this stuff after the fact. Maybe we're getting paid for. Maybe I would say maybe 50, 60% of the stuff that we're doing on a day to day basis may not even get paid for, but we're doing it because we have to, not because we're looking. Thankfully we get paid well, I'm not boohoo or anything like that. But I do think that the more asynchronous care gets pushed the more people realize that that is in my eyes, it is a standard of care in most cases because the thing that gets me the most with healthcare in general is the majority of the care. Once the patient leaves our four walls then the follow up is really lacks. There's a lacking of follow up because of multiple things like the amount of people that exist. We just don't have time to interact with every single person and also we don't know what's going on at home. Right? So if we can set up a system that's why I love kind of things that you're building. If we can set up a system where we can create multiple touch points along the patient's journey where we don't have to be the ones physically doing it, we can improve health in general and also it opens the door for the patient to reach out to us and tell them they have a problem. One thing that I've noticed throughout my time is some patients for some reason think that they're bothering us by calling us and telling us hey, I'm having this issue. And that always made me feel bad because I always tell my patients like, hey, we're doing this, I'm here because of you. I'm not here to just chill and whatever. I'm like here to help you. But just because our model is built that way, we've taught our clinicians that way, we've taught our patients that way. You have to see as a person or nothing can get fixed.
Alex: Yeah, and I think some patients won't hesitate to text you every too many times and others will never text you because they are afraid to disturb you. And so it's a bit random and it shouldn't be that way. It should be processed and properly organized. And of course there is a lot we can have with machine learning kind of problem.
Zain: Yeah, exactly. They say the squeaky wheel gets the grease, right? We have some patients that are constantly calling us and then we have some that should be calling us that don't call us. And that's why I think it's important for automated reach out like hey, if they started on something or drug or something like that, that in a week or two we know that they're going to experience side effects within the first week. So why don't we just reach out to them first week. And I think if everyone had I think almost everyone, if we talked to any healthcare provider, if they had the staff or the time to do it, they would 100% do it. It doesn't matter if they have paid or not, but they will 100% do it. It's just that the staff doesn't exist, the time doesn't exist. So that's why tools like you guys are creating and other people are creating is what gets me excited. That's the stuff that gets me really excited right now. Because for me it's the boring stuff. That really is what makes the biggest impact.
Alex: And so in asynchronous care, when for instance you want to mess with your patients, actually messaging the patient is very fast. What takes a lot of time is to get the context for the nurse, for instance, who is doing this? They would spend two or 3 minutes remembering who is that, what should they say? What is the state, what is the context? And then it will take 10 seconds to write this message. And this is where I think, really think machine learning has a very short term impact. Just summarizing like if you had a perfect assistant, okay, this is X, you should say this because of that and then here is a message I drafted, just click send and we can enable people to handle maybe ten X, 100 X more patients with very simple things like that that are obvious in other industries. But I've not reached escape yet.
Zain: Yeah, I mean exactly. The first thing when we ask provider or a nurse is asking talking to somebody or even talking to me and I'm like who is this person? And then they're like oh, it's that person that has this medication or they have this or they have family that comes in. Okay, yeah, you're absolutely right. If I can just get a summary saying this is the patient. These are the quick hitters. And they're like, oh yeah, that's right. Boom. And then the other problem with the way we do it now is a patient call leaves us a message and then sometimes we need the provider or the doctor to answer the question, right? So they're running around in clinic and we're trying to get a hold of them and then they answer that question. So we call the patient back. Either we get a hold of them or they leave a voicemail. But hey, the doctor said this. And then patients like, oh, you know what, but what about this, this and this? You're spending these interactions can take almost half your day, right? Just one person. I'm not talking about multiple people. And if we can just get all that information all at once and kind of like a quick rolling thing, that would save so much time. I mean, I can't tell you how many times one patient has taken up. Not at a fault of their own, right? It's just the way our system is built is not built for good communication and it just takes up so much time. To your point, like, if we can get some of these, we can do it, we can automate certain things or we can even just show suggestions of how to answer certain questions. That in itself saves so much time. And I don't think people realize how much time it actually saves.
Alex: Just summarizing the context rather than context about the case. Our patients would be a huge time saver. When we mentioned AI in healthcare, people think about good robots acting as a doctor. And actually the real impact today is in the kind of thing you just mentioned.
Zain: Everything is going perfectly. We have all the data. You're able to get all the data in the world about healthcare. Where do you see AI ML eventually being like where do you see your vision of AI ML in healthcare in our field?
Alex: Because there are so many different potential in what we do at Nabla, we really want to build perfect AI assistance for every escape professional. And so imagine you have everything on your side from morning to evening, knowing what you do, listening to the consultations, looking at what you are working on, on the EHR. And every time you can do something on your behalf, we'll do it. We'll just do it. We've square.
Zain: Is that with Scarlett Johansson?
Alex: Yes. You don't see her? It's an audio assistant. Give an ambient AI. AI is aware of what every person does all day and tries to be helpful every time she can with a woman in the movie. And I really think this is what we are trying to do long term. And not just us at Nabla, but many people try to do. And this is necessary because there is a huge shortage of healthcare professionals. There will be 20 million, I think, Health professionals missing by 2040. So we need to do something. And on the other hand, they do so many things they shouldn't do. There is a huge opportunity to have a positive impact here.
Zain: No, I 100% agree. I think that it's great to hear. Sometimes I talk to technologists or people in technology, and they are trying to I don't always agree with what they're trying to do, but I'm excited about what they're building. But what you're mentioning and what you're saying, I 100% agree with. Like I said, healthcare is an interesting field where constantly we need to help people, but we don't help ourselves, if that makes sense, right? We're constantly thinking about everyone else, and then we're just taking everything on our plate, whether we need to or not. And eventually we have to get over ourselves. We need to be like, hey, we need the help, and we need to I don't know if taking a chance is a good way of saying it to healthcare professionals. It sounds like we're taking a chance, but I think that we need to open up and be open to other ideas. We need to get over the fact that, okay, the EHR did not work for us, right? We need to get over it. Things change, life happens. We need to technology is always constantly moving. We can't be stuck in the past. You brought in, they were working on paper charts and they were really happy, right? They were like, oh, this is amazing. This is so easy. I was having a conversation with my provider at my hospital, and everyone was reminiscing of the days of the paper chart. And they're like, oh, man, I would love to go back. I would love to have a system shut down so we can just go back. And I laughed at it because I'm like, man, I could never deal with paper. I'd rather deal with this. But they're only thinking that way because a good solution hasn't come or they haven't faced a good solution yet. So I'm hoping that people like yourself, your team, where you guys are building a Napoleon, I love what you guys are doing. And the more we can get of that and people with the right intentions not using health care as a box of money, right? It's not a free paycheck, because that's the other thing too, right? That kind of happened here in Covet is people just kind of jumped into health care with not the greatest of intentions. And I think that hurt. It did two things. It sped up the move into technology because it needed to. But then also, now that the dust is cleared, it's also made people really weary. Before, when you had a good solution, you might be able to get a conversation with, like, the CTO or CIO or whatever. Now it's like, well, we'll see. Do we really need to do this? Do we really need to do that, so we'll see where it goes. But I'm excited for the future with what you guys are doing and what other people are doing for me, it's exciting.
Alex: Yeah, we have already sent it too.
Zain: Like you said. How do people find you personally, your team? How do they reach out to you guys?
Alex: If they have any questions, on Nabel.com, our website and my email is firstname.lastname@example.org, whether you're a physician or if you run a digital clinic and things like that, you are interested. We are very happy to talk and see your use case and work together. We are at the stage where we are very flexible and willing to talk to everyone. Better understand the problem we are trying to solve. So any contacts is welcome.
Zain: I can speak from experience. They have an amazing team, really awesome. They're very great people to talk to. So even if you're curious, just reach out to them because they're a great group of people. Is there any social media you wanted to plug or anything like that for people to follow you or follow the journey of Naval?
Alex: Yeah. For Twitter. My Twitter is lxburn. And get everything out there. Okay.
Zain: Yeah, I'll have all that stuff linked in the show notes below. But Alex, I want to thank you for your time. Very generous with your time. Thank you so much. Great day.
Alex: Thank you so much, Zain. And thanks for your invitation.
Zain: No problem.
Alex: Thank you.
Thank you so much for your support! Please do not hesitate to reach out if you have any questions or leave a comment.
If you enjoyed what you heard please consider subscribing and sharing!
I hope you have a great day!