Big Data, Machine Learning, and Health Care with Julia Hu, CEO, Lark Technologies

Michael Krigsman:

(00:02) Welcome to episode number 151 of CXOTalk. I’m Michael Krigsman and we are speaking with Julia Hu, who is the CEO and founder of Lark Technologies. Lark Technologies is aggregating a large amount of data that it puts into the service of an app to improve health care and so we’re going to be talking about data and health care today. Julia how are you.

Julia Hu:

(00:36) Great thanks so much Dave, great to be here.

Michael Krigsman:

(00:40) You know a lot of people call me Dave even though my name’s Michael but that’s okay. I’ve been getting that through my whole life, but anyways tell us about Lark Technologies.

Julia Hu:

(00:54) I’m sorry I’ve been sleep deprived Michael.

Michael Krigsman:

(01:01) I know you’ve been travelling a lot and you’re in an office some place across the country.

Julia Hu:

901:07) That’s right, I am in our partners boardroom right now but thank you so much Michael. Great yeah great to be here.

Michael Krigsman:

(01:19) So tell us about Lark Technologies, tell us about what you do.

Julia Hu:

(01:23) Sure, so we really try to help you know help people connect with someone who understands them and can really coach them towards better health and happiness. So Lark really started when I was a little kid. I mean I didn’t start Lark when I was a little kid, but the spirit of Lark really started when I was a little kid.

(01:52) I grew up with these horrible horrible stomach attacks and it went every few weeks and it went undiagnosed for about 20 years. And so I think I just survived my childhood and lived a great childhood by working with my pediatrician and my father and they were like my coach team. They worked with me 24/7 practically you know giving me advice but also cheering me on.

(02:28) And so that’s who Lark is. We try to create a 24/7 you know personal health coach that’s there by your side, anytime and anywhere.

Michael Krigsman:

(02:41) So when you say a personal health coach, what exactly do you mean?

Julia Hu:

(02:46) Well it’s actually a text buddy and it is a chat buddy, but you can text anytime. Lark also text’s with you, your Lark coach and it looks at all of the information, health data about you, you know from your steps to what you ate, your sleep and it looks at that and then it coaches you at the right moments with tiny little you know conversations.

Michael Krigsman:

(03:19) Well then I guess the obvious question then becomes how does it know what to say to you and how to coach you, and how do you go, how do you make it not just something that’s superficial but that’s actually meaningful.

Julia Hu:

(03:36) Yeah absolutely, so the whole thing around having a coach in your pocket right, using your mobile phone it’s really opened up because of a couple of macro trends, and the first obvious trend is the mobile phone and it has so many sensors now, health sensors, you know motion sensors. We can actually get information about you know that person without Joe telling you what he was doing. We can see not only if he took 10,000 steps but really that 10,000 was a 30 minute bike ride and then it was you know pacing around the office.

(04:23) And so we really can get so much passive data from this so called Internet of Things, you know from your Apple watch, from your wireless scales and all of these things are coming into Lark, and Lark is basically looking at that and saying, okay, you know what should the coach say now.

(04:47) And so we have live coaches and we have some of the best Harvard and Stanford faculty and health and behavior change on our coaching team. But really what’s extending that personal 24/7 experience to everyone is the advance of artificial intelligence technology.

Michael Krigsman:

(05:12) So your tracking or the user is tracking through the sensors in their phone, but at the same time you are collecting data from experts as well and mixing the two types of data, basically the sensor data and the expert data is that correct?

Julia Hu:

905:34) Yeah exactly, so experts are the ones who are interpreting the data and coaching you at the right times, whereas the sensor data is really just coming in so that we can build a better understand of who you are, what your goals are, and how you’re doing at this exact moment.

Michael Krigsman:

(05:57) So tell us how the expert data works, and I know that you employ data scientist and machine learning and artificial intelligence. So maybe break that down for us, describe how that pieces together.

Julia Hu:

(06:13) Yeah sure so I think you know, one of the best examples I think is actually explaining how we first started. We met this woman Cheri Mah, and Doctor Mah is a great researcher at Stanford and she also happens to coach MBA stars and Olympians and NFL stars. And she would take these $5,000 suit monitor, medical grade monitors and she would put them on each of her athletes, and then at the end of the week she would collect all the data you know by hand, and analyze it and then she would start texting them and coaching them. And we thought that wow, if everyone could have a Cheri Mah in their pocket they could be breaking world records like these amazing athletes were. And so she helped them sleep better and then perform better.

 (07:24) So we wanted to replicate that experience, and so really if you think about that it’s in two categories. The first category is of course gathering that personal data and you know the second category is really into coaching.

(07:47) And so the way we’ve kind of cloned Cheri so to speak is by understanding what she would say to all of her athletes in all different scenarios, and map that to you know basically a system we can then pull off every single Cheri conversation at any given time depending what the user, you know what the client or what the user is doing, and so it was a lot of mapping of you know, let’s get the data in somehow.

(08:25) So we actually built one of the first wearables in the wearable industry that was wireless. So we built the Bluetooth wearable and we sold it in Apple stores and it was helping you monitor sleep. And then we took that data, we went a step further; we didn’t just show the data to the user. We actually took that and ran that through our generation one artificial intelligence engine that really then said, okay this is what Cheri would say at this moment you know, and since you’re travelling this is what Cheri would say to get rid of jet lag. So it was really building on that, and now we’re on our fourth generation AI engine.

Michael Krigsman:

(09:12) So we have a question already from Twitter from Alan Bergson and he’s asking about the privacy and the liability aspects of this so maybe jump in and lets address some of his concerns about hose two issues and then let’s come back and talk further about how do you ge this expert data, how do you model it. when you talk about artificial intelligence what exactly does that mean.

Julia Hu:

909:46) Yeah sure, so let’s talk about privacy first. You know privacy is super important to us and what we really believe is that every single person is controlling how much they share with us to the most granular level.

(10:04) So basically we’re providing a service based on how much information you’re willing to give us, and the way that we do that is you know we basically ask you permission in the very beginning for everything. And what’s great is if you’re familiar with Apples Health Kit, Apples Health Kit is this wonderful I think wonderful example of where big data is going. The idea of an open data ecosystem, you know much like Google Fit, Samsung Health.

(10:47) This open data ecosystem where different devices you know, say your Apple watch information or your wireless glucometer information or you know any wearable information right, all of this information or app information, you know data from Strava all of this data can actually come into this one, you know holding ground, this warehouse. And then Apple created Health Kit which allows and permissions, you know allows you to permission other apps and other services to take data from this pool.

(11:34) And so a user can actually say, I want to give Lark my you know my exercise data but I don’t want to give Lark my sleep data and I don’t want to give Lark my height and weight data. So it’s really interesting, you can permission to the nth degree and that allows for a true sense of privacy that I own what I own and I give it, and of course Lark doesn’t share it with third parties, and so we don’t share personal you know level data to any third party.

(12:14) So that’s how we see privacy, and I think it becomes increasingly important as more and more of this IOT health data becomes available.

Michael Krigsman:

(12:27) So the reality of control that the user has over their data creates the sense of privacy and therefore, I’m assuming by extension also creates a sense of trust with the app and Lark Technologies, your company that creates the app.

Julia Hu:

(12:49) Yeah absolutely and in terms of liability you know we help people focus on a few areas, so we help people eat heathier, exercise in a more affective and hopefully more ways so we’re encouraging them. You now sleep better, stress, we work on stress and overall we do it in a very positive way through some of the gold standards of coaching and therapies, so CBT (Cognitive Behavioral therapy) has been very effective.

913:28) So we do all these things that are really around lifestyle management. We coach on different chronic diseases, so if you are pre-diabetic or you are diabetic we coach you in different ways. So we have a pretty wide chronic disease platform that focusses on different disease management areas and also wellness management areas.

(13:56) And in terms of liability, hat we don’t do we’re not a medical device, so we’ll never for example you know titrate your medicines, give you different dosages of medicines. What we do do though is with our health care partners we do escalate. When we see potentially you know a worse trend we can escalate to real nurses who actually can help you.

Michael Krigsman:

(14:27) That’s really interesting so the app can go off the phone so to speak and connect with the users real world health activities.

Julia Hu:

(14:40) Yeah that’s right and so we are a B2B2C company, and so on one hand we have a B2C product and the B2C product is out there, it’s free and you know it’s basically the Lark coach if you want to get a little fitter, stay healthy and it chats with you about certain things.

(15:07) We also use the same AI technology and coaching technology to provide deeper coaching services for people, you know who are like me who are basically chronic disease patients but we’re also regular people. And so it’s just a casual conversation to kind of keep us on track so to speak.

(15:30) And those healthcare programs they’re much more with our health care partner, so we work with health insurance companies, we work with pharma companies and basically we work with and help prevent chronic diseases. And in those different programs then the escalation to a health care provider or a nurse or a coach is definitely an offering.

Michael Krigsman:

(16:11) Okay, let’s go back to this notion of artificial intelligence and machine learning and the data science behind your application, so would you elaborate on that.

Julia Hu:

(16:24) Yeah sure, so it’s really interesting because in the past the way that we have really approached health care is by understanding general demographic trends. So you might have heard people talk about population health, you know what is population health, it’s this idea that each population you know for example Asian American women who have a background of this, you know each population has different characteristics, and so when you want to coach them to get better or you know you want to look at how to influence them it’s done by these demographics.

(17:20) You know but with IOT and with personalized data, you know it’s kind of opening up a new area it’s almost population health for the population of what right. it’s your data and so for example we found that research shows that if you don’t get a lot of sleep that you are much more likely to be stressed.

(17:54) So we can actually see that if you haven’t slept very well for two or three nights in a row. You know at 2PM that’s when you’re most likely to be stressed the next day and we’ll say, hey how are you feeling, you know, you might be a little stressed because you haven’t slept well in the last few days. How are you feeling? And you can say yeah, I do feel stressed, and we can say well that probably means that you’re probably craving a lot more fatty foods right now, so think about getting some nuts instead, for example if this person wants to lose weight for example.

(18:38) So it’s examples like that, that just kind of show the power of understanding a person more and more. You know if you just have the number of steps that you took, your coaching can’t get much more complex than, ‘hey, you hit 9,000 steps yesterday; today try to hit 10,000 steps.

(19:03) but we think that you know big data is so much more. I’m kind of an extremist in the way that I view data. I’m not a technology geek, so I’m the non-technical cofounder of a tech company and the way I see it, you know the more data there is, the more continuous data there is in health care and in many other things the better we can get at predicting issues and predicting wants and desires. And for us what we want is as much real-time data as possible, as many streams of it, but when we talk to you we don’t talk about the nitty-gritty data because you as a person don’t have time to deal with it.

(20:02) And so we really believe in ultra-simplification of all of this data and turn it into a conversation, and so really that’s the way we think about it. We do tons of machine learning in the background of what activity was that motion, and based on the way you use your phone how many minutes did you sleep at night. We use all of that machine learning to figure out the patterns that you have and then we simply just have a conversation with you, so you don’t have to type in, so I started sleeping here, you know I started my run here, you know I ate three grams of white rice today. It’s basically done in a text message way and mainly behind the scenes.

Michael Krigsman:

(21:04) how do you model the expert advice that you get. How do you model the experts?

Julia Hu:

(21:12) Yeah, it you know it really came in phases. So we’ve been working on this for five years now. At first we modelled what I consider the brain, so we basically looked at these incredible research papers and we looked at how are our health experts intervened and coached their patients, their customers and their members.

(21:48) And we first looked at just transferring like methodologies, so it was very dry, and I would say we would say okay, if a person hasn’t eaten for several hours, this is probably what they’re craving, this is what they usually eat, lets’ address these things.

(22:18) So we would do things that were very by the book so we automated the American dietary Guidelines for example, and so our experts would help us automate all of these guideline innovations that were all research based.

(22:40) What we found was that when we just told them information and educated them about these rules so to speak, you know it didn’t actually change behaviors that much. So as I was reaching into my childhood and understanding the parallels, basically what we did we translated what the doctor was telling us right, you need to eat less, sugar, you need to eat more carbs, you need to sleep at least eight hours. So we were translating the education and that was important, but it doesn’t help with engagement as much, it doesn’t help with actual behavioral change that much.

(23:37) So then we started thinking about okay, my father was my cheerleader. He was my stick and carrot and my pediatrician was the one that taught me to be more mindful and test things about myself. So how do you systematically use artificial intelligence to mimic those very soft things, the idea of compassion, of love, of no judgement, how do you actually translate that?

(24:22) So we started looking for people who were experts in systematically understanding that. so it goes beyond just psychology you know and positive psychology and that’s where we look into the most effective types of therapy is CBT, Cognitive Behavioral Therapy. It’s about reframing your mind to think about the situation in a different way, in a more positive way. You know that’s a very simplified version of it.

(24:57) You know we looked at Cognitive, and we have a professor who is an neuroeconomist and studies how people make decisions. And we hired a person who was at the Stanford research lab in compassion and altruism and basically we then used their research to start creating a personality that felt very personal to you. So you know, Lark figured out how much of a sense of humor you have and you know changed its tone a little bit to you, but basically we systematically kind of brought in the science of behavior change of CBT into the tone and the warmth of the voice.

(25:54) I think that’s why when people talk to Lark you know they say, they don’t call it a tool and they don’t call it a monitor, they say things like she really understands me and she’s my buddy, and she gives me little nudges. And that idea I think is really exciting. The idea that we can start to bring the best of coaches and nurses and and doctors and extend it in a personal way to people.

Michael Krigsman:

(26:33) So you spoke…

Julia Hu:

I can’t hear.

Michael Krigsman:

(26:37) You can’t hear now, how about now?

Julia Hu:

Still not.

Michael Krigsman:

(26:45) Okay, well don’t you guys just love.. No dice.

Julia Hu:

(27:02) We’re having technical difficulties can you hear me?

Michael Krigsman:

(27:04) I can hear you.

Julia Hu:

I can’t hear you.

Michael Krigsman:

(27:08) Okay, I have an idea hold on. You know the way this works is I have a list of questions.

Julia Hu:

You want to call me?

Michael Krigsman:

Hold on.

Julia Hu:

Oh I can’t see that either, oh okay, okay, what is the connection between data and health care?

(27:48) So what’s really exciting is that health care has always been a very data driven industry. It’s just that the data has always been frankly very very you know sporadic. Right, so you go to your doctor once a year to get your annual check-up and you get a couple of markers there. You have all of these tests, but it’s very sporadic. But health has always been a data centric industry.

(28:31) What I think is so exciting is that now you are able to fill-in perhaps much less granular data, but data no of the less about a person, you’re able to fill-in those gaps and what’s exciting about that, is that then you know health care can then move from a very focused on a Band-Aid solution to address those few points in your life. It can be a preventative, it can be continuous improvement.

(29:07) So you know health care doesn’t have to be as drastic you know, you are preventing diabetes by reducing their BMI, their weight and that can extend your life quality and your life by years. So what we’re seeing is this shift towards continuous health care at a more manageable daily wellness cycle. So I think that is an amazing transformation, and that’s not possible if you don’t have all of that data and your benchmark.

(29:50) You know we look at everyone as a profile of themselves and everyone has a kind of a digital signature of your life. You know we are creatures of habit; we have a signature of how our life is and sometimes it’s hard and sometimes it’s lower and hopefully the graph should be a sustained improvement. So I think health care’s really improving that way.

Michael Krigsman:

(30:19) Can you hear me now?

Julia Hu:

(30:20) Yup.

Michael Krigsman:

(30:21) Oh good, you know these things happen on the internet. It’s a wild west out there on the internet. So we have a question from Esteban Kolsky who is an industry analyst and it’s really relating to again what we were talking about how do machines learn from experts, so can you again elaborate on how that happens.

Julia Hu:

(30:47) So how do machines learn from experts, and right now I really would not say we are an  AI company. We really are kind of this nice combination of coaching and using AI to extend the power of coaching. So I think it’s actually not just how can data, machines learn from experts, it’s actually a feedback loop. So what I think IOT and all that is we have incredibly sped up the feedback loop in which we learn, so I’ll give you an example.

(31:38) You know, in the beginning we started building you know basically a set of systems that allowed us to very quick figure out you know which interventions actually work, which conversations do not actually encourage someone to do something.

(32:03) So you know there’s a lot of research that says you know if you give someone like a little bit of positive feedback right after they do something then they’ll internalize it is a great thing. We have hundreds of different types of conversations that tell it to you in different ways and different tones. Some of it is a little bit more sarcastic, some of it is a little bit more warm. So the idea of having these thousands of A B tests because all of us are different and seeing how the user reacts to those, and seeing what they actually do after we tell them what to do. These systems of feedback can then be presented to our coaches, and then our coaches can then help us make broad coaching changes.

(33:17) So our conversations are not written by machines, they are written and they are actual real coaching conversations that are created by real human beings. But it’s just that the feedback loop is so fast because you know, we ask a person of what they ate for lunch. They tell us that they had a lot of cabs and a little bit of vegetables and we’ll tell them, will challenge you to eat more vegetables at dinner and in a fun way, and we can see that if they did that at night. So it’s really amazing because of instead of clinical studies of a few hundred people in a very very protected environment, in a very structured environment we have real-world studies of how people are acting, and we are making all of these tweaks to our engine constantly based on all of our conversations.

Michael Krigsman:

(34:24) Arsalan Khan asks on Twitter asks, do you have your own home-grown tone developed, AI engine and how long did it take you to develop?

Julia Hu:

(34:35) Yeah, we do, we home-grew it and our baby is five years old now so it’s taken a long time. I laugh and you know I talk to my co-founder and my team and I laugh at how in the first four years our AI, our coaching felt like a four-year-old at best. You know it was just so clunky, it didn’t really understand things and I think at the fourth year we had a turning point and the conversations felt more natural. Of course, we have so much more to go but you know we use all of the different data pieces, so we have 70 devices that we hook into now and most of the smart phones out there, so that’s a couple of thousand products that we also hook into and we just coach on top of that. And now we are going to be pre-installed in all of the new Samsung phones as a chat buddy so it’s pretty exciting.

Michael Krigsman:

(36:02) So, let’s talk about the business and the entrepreneurial aspects, as you mentioned you have a deal with Samsung to be on their new phones and I know you have a relationship with Apple, so how does a company such as yours develop these kinds of relationships. I think every entrepreneur wants to have these kinds of relationships with larger companies or such significant channel, how do you do it?

Julia Hu:

(36:34) Well I think you know, what’s really exciting is you just put a stake out there and take some risks and you just try to innovate. I think that’s the best way, because innovative companies are looking for innovative companies to support and frankly that’s developing relationships with them. You know we had our products sold in Apple stores globally and that’s how we started our relationship you know but really I would just say it’s great time for start-ups because you know as long as you are pushing the envelope there are so many corporate development folks in innovation that work on finding industry leaders and thought leadership.

(37:43) So press obviously helps a lot and Forrester research a Wall Street analyst named as the most innovative digital health product of the year and published a large paper on us and things like that certainly help, you know establish credibility and get your name out there. But I think it’s a great time because they are are so many companies focused on innovation now.

Michael Krigsman:

(38:17) What are some of the challenges that you have faced in building this business?

Julia Hu:

(38:25) Oh my gosh so many! So I would say that you know market timing is always so critical, and I think we are taking a huge leap. You know, we walked away from a hardware/software business and just went straight to software, and you know we walked away from a couple of million dollars in revenue and the hardest part was really finding product market fit. We knew that we wanted to really focus on personal coaching and creating a personal health team. But you know, AI takes a long time to build and so we really had to you know sell the long dream and we had to get VC funding.

(39:35) And now that our product is finally there, you know after years and years of R&D, finally good enough to be an extension of a nurse or a coach, our chronic disease platform is finally being used by health insurance companies to basically augment and in many cases even you know drastically reduce the amount of expense in having a very large telephonic coaching center, that doesn’t know when to reach out to you, it’s troublesome. So if you have all these coaches and nurses not knowing when to reach out to you, and not knowing to be able to personally coach each and every person all the time, we are basically being an extension of that huge industry.

(40:49) You know that’s taken a long time, so I would absolutely say you know, as a start-up balancing R&D and cash flow, and to finding and sticking with you know your ultimate dream of where you think your company can go. And we felt personally that we could not be the best company we could be if we focused on both our hardware device and our software coaching services. So we took a big gamble and basically you know ripping part of our company out, so that’s been a big challenge yeah and I hope that we see that things are paying off

Michael Krigsman:

(41:39) We have just a few minutes left, so maybe you can share your advice for other entrepreneurs what did you learn during this product market fit, developing your value proposition and making the difficult decision to abandon a multimillion dollar segment of your business in favor of what you’re doing now. So just in our last few minutes maybe share your advice from your experience going through that process.

Julia Hu:

(42:09) Yeah, well you know I think what I would say is really dream big. No matter how big or small your vision is you know it’s going to take a lot of years. It’s going to take a lot of sweat and so you might as well dream big because you know if you don’t hit the stars maybe you’ll hit he moon right. It’s great to have something that drive you but also drives people to follow you.

(42:54) So I really learned that in the beginning I was much more timid and my dream was actually to be in sky mall I wanted to be in sky mall and I wanted my device to be there and I didn’t realise that I could dream so much bigger. You know and I think taking a few steps in that direction I just remember one of my mentors saying, hey you know if you don’t want to go big then just go home. Then I started being able to believe that my idealism and my hope that everyone in the world could have a 24/7 personal help team – someone who could be there that cared for them, that understood and didn’t judge for them, that helped them get better and helped themselves with the tools to empower themselves.

(44:06) I think that dream was you know always part of me growing up and it was just to allow myself to believe that we could change the world if we focused on that vision. I think entrepreneurship is about dreams and being able to believe in that and not get to jaded is really important.

(44:38) I think another thing is you know, you’re obviously not going to hit that dream on the first day and that’s okay, and that failing really quickly and learning from these failures is totally part of the process. And you know I think being forgiving of yourself when you do fail is really really critical because you’re going to fail every day, I failed all the time you know and I feel that I’ve become more resilient to the failure and I don’t let it hold me back and I sometimes use it to fuel me.

Michael Krigsman:

(45:35) So having the confidence to push forward and at the same time being resilient when things are not going your way fundamentally is the bottom line.

Julia Hu:

(45:47) Yeah.

Michael Krigsman:

(45:50 Okay, well we have been talking with Julia Hu, who is the cofounder and CEO of Lark Technologies and this has been episode number 151 of CXOTalk, Julia thank you so much for taking the time today.

Julia Hu:

(46:07) Thank you so much Michael or should I call you Dave.

Michael Krigsman:

(46:10) Yeah call me Dave! And everybody thank you so much for watching and we will see you next time. Bye bye.

Companies mentioned on today’s show            

Forrester Research     www.forrester.com

Google Fit:                   https://fit.google.com

Lark Technologies:      www.web.lark.com

Samsung Health:         https://shealth.samsung.com

Samsung                      www.samsung.com

Strava:                         www.strava.com

 

Julia Hu:

LinkedIn:          www.linkedin.com/in/juliahuceo