Ram Prayaga - CTO at mPulse Mobile

(Joel Beasley at 00:00:00) Hello, my friends. Today we are talking to Ram, the CTO at mPulse Mobile, and we discuss the innovations they are making in conversational AI, how security is embedded in your company culture, and how they scaled from 3 million to 300 million conversations a year. All of this right here, right now on the Modern CTO podcast. Here we go. This is the Modern CTO podcast.

(Ram at 00:00:35) Hello. Hello.

(Ram at 00:00:37) Hey, Joel.

(Joel Beasley at 00:00:38) I'm loving that background. It's really crisp.

(Ram at 00:00:40) It's nice, ain't it? Yeah. Well, kudos to our marketing team for putting it together.

(Joel Beasley at 00:00:45) I'm really pumped up to have you on the podcast today. I'm a big geek when it comes to conversational UI. I don't know much, but I want to know more. But I want to start out with talking a little bit about, you know, who you are and why do you even like technology?

(Ram at 00:01:04) Oh, wow. Yeah. I mean, are we diving in through this?

(Ram at 00:01:09) Yeah. Yeah, we're good.

(Ram at 00:01:10) Okay. Awesome. Yeah. No, thanks for that question. It's actually a great question. And, you know, it's funny. I look at my role today, but also right from the beginning. I went into engineering because I wanted to be like my two uncles that I grew up with who are engineers. And they wanted to be like their uncle. And so it was kind of like this history of wanting to be like someone else. And so math, things like that, were something that I really enjoyed and really wanted to be good at. Whether I was good at it or not, I was going to be good at it. You know, it's one of those things. Kids convince themselves, and that goes a long way. And so naturally, I don't know, when I was 10, my parents decided to get me a Vic-20. You may not know what that is. It's a computer that has all of four kilobytes on it. And so I was given that and I started to code. And it's just sort of one of those things you just do because you're surrounded by people that seem to think that you should do it, and you convince yourself, therefore, that you should do it. So technology was very much from an early stage something that I enjoyed. And everyone looked to me to go in that direction. And I will say, though, that my dad very much wanted me to be a doctor. He was a doctor. He was hoping that I would. And I'm like, I don't want to get up in the middle of the night and have to wake up for a pager or something like that or skip dinner. Well, little did I know.

(Joel Beasley at 00:02:43) Server's crashing. Yeah.

(Ram at 00:02:44) Servers crash and, yeah, you get problems at any time of the day. So I very much said, hey, technology is something that I really like, and that's what it was. So I know that's a very long and kind of a more historical answer than you'd like. But the flip side of that is, because I think of how I've seen technology being used, it really became one of what is the value of what we're doing and what we're building, rather than necessarily the building of it. And for me, over the course of these years that I've spent being in these kinds of roles, why am I doing this is the guiding principle more than how am I doing it and what am I doing. And I think the why is really related to who you're doing it with, you know, the team, who you're doing it for, the people that get the benefit of, hopefully, your good work. So those things have increasingly become the driving force of why I enjoy technology. Who I'm doing it with and who I'm doing it for is really what I get excited about.

(Joel Beasley at 00:03:58) So tell me about who you're working with today and what you're working on and why.

(Ram at 00:04:03) Yeah. So I joined this company at the very early stages with the goal of trying to make the consumer health care consumer experience a little bit better. We were looking to take some of that uncertainty around, do I have an appointment? When do I have my appointment? I think I have the card that your doctor gave you or something like that. And remind people when they need it and have that contextual awareness. And so this started out with a vision of saying, hey, how can we make the health care experience, health care engagement a little bit easier, a little bit more convenient, a little bit more human oriented? Mobile messaging, as more and more people started not just using their phone to talk, but rather using their phone to do everything but talk. I believe the eleventh most important reason why you have a phone is to talk on it nowadays. That's, you know, based on a recent poll, an interesting fact that the 10 most have nothing to do with talking. And so if we think about how the mobile and devices have taken hold, we were early on in recognizing that saying, hey, things like text messaging or sending notifications to your phone while you're moving about, while you're going from place to place, that's the most convenient and seamless way of integrating with your life in the ultimate goal of trying to take care of your health, get you a little bit closer to do that. We're not going to cure cancer. We're not going to do all these amazing things. But what we can do is get you a little more connected and more aware of how you can take care of your health. Get to that appointment. Get your lab test done. Try and exercise a little bit longer. Eat a little bit healthier. Those are the things that we focused on, and we found that mobile communications are very much a neat way to dovetail into what you're already doing. So our technology is really focused on getting those messages out. And increasingly, you mentioned conversational AI. We've understood more and more that, you know, when you want to do things at scale, you've got to use the best technology that's out there. So we didn't go into AI as, hey, that's the thing. We came into it thinking, how do we do this well and really at scale where we're talking not about hundreds of people, but millions of people. And we've been fortunate enough to touch that many people, and it continues to grow. So, you know, when we're talking about those pagers, now it's not pagers, now it's text messages that I get at three in the morning. Fortunately, I don't have to wake up, and I have a great team that gets at it. But it's that growth as well that we're, as a team, having to manage that scale, incorporating things like, how do we do very smart natural language understanding using the best and brightest in that. Again, putting the value, putting the purpose in front of the guiding principles before using the latest tech because it's the latest tech. And I say that because I think there's some who are using smarter technology than we might be doing. But I think where we've always measured ourselves is, are we delivering that value to our clients and to our members? And, well, I mean, there's someone who told me, one of our investors actually said, hey, if people are buying your stuff in health care, that is the greatest measure of whether you're making a difference or not. You can have the greatest product. If people aren't willing to buy it and use it, it doesn't matter. You know, don't bother. And I've always had that in the back of my mind. Like, let's make sure we will fail, but we will learn from those failures and we will improve. And that's very much also the concept behind AI, you know, the machine learning or whether all these methods are just that. I start out kind of not knowing much, and I grow over time. And I think we approach our jobs the same way. So yeah, it's been a combination of building that scale, incorporating technology that we think is really important and valuable here, and really understanding the domain in which we're working. And that's one of the other areas that we as a company, thanks to all of our amazing clients, have been able to learn what are the things that are impacting. One of the things that we understood about a couple of years ago, maybe we had ideas a bit earlier, but more firmly a couple of years ago, is around the social determinants of health and how that impacts your engagement with health care. It's interesting to see that, and you may have heard about this, but people talk about where you live makes almost a bigger difference than your genetic code, your ZIP code. It has a bigger impact on your health and your outcomes around health than anything else. And so we heard those big statements. We dug in. We built our own index. We looked at all the data that was available, and we see this difference. And it's really interesting to try and make that different. We think about what's going on in the world around us earlier this year with the issues around racial injustice. And we sort of also saw that in some of the engagement that we were having. And when we start putting that lens on it, it was very striking. And so that's where, you know, again, technology, data, our purpose, it converges into saying, okay, we have the ability to kind of, again, I don't want to overstate, but can we make a better future? Can we make a little bit more of a difference? And I think, you know, the proudest thing is yes. The answer is yes. How much? We're going to keep trying. We're going to keep learning. And that's why I love technology because it always introduces new concepts and new things to work with.

(Joel Beasley at 00:10:16) So where's the best place to live for your health?

(Ram at 00:10:19) Oh, I am, you know, a person of privilege. So I will say where I live is actually a pretty good place. But there are many places like that. There are places that even within, we're in Los Angeles, and so parts of LA have a, what our index does actually was go down not just the ZIP code level, but the census tract level. So literally my neighborhood is what we can index and say, hey, your needs are this much versus some other neighborhood. And so I've actually looked at it personally. I'm like, oh, okay. I guess I'm living in a good spot. Not a problem. But you go over maybe a couple miles over, and you see that there's a very different index and also comparably just driving through knowing about it personally. So yeah, that does bear itself out. We are able to, where we did what we did also separate. I was involved in a City of LA COVID-19 hackathon challenge, actually with my family. And because we were all trying to pitch in and see what we could do, and we did an analysis of where COVID cases were and geographically mapped them. And, again, that mapped directly to that social determinants of health that we're talking about. And the higher levels of COVID cases were correlated very much with higher, you know, areas where there was a greater need, where there was a greater impact of these kinds of areas, lower income, less education, less access to health care, primary care physicians, you know, things like that that have an immediate impact on health outcomes.

(Joel Beasley at 00:11:58) See, I was thinking when you were discussing that, where my mind was going, was like weather and climate. When you were talking about this map, I was like, where's the best place to live for my health? I want to go to that region or time zone or whatever. Maybe a place that doesn't switch their time zones. I've heard there's a lot of arguments, people wanting to stop the time changes because of health reasons.

(Ram at 00:12:22) Okay. No. We didn't have all that. But yeah.

(Ram at 00:12:28) No, this index is really getting at the social determinants, right? So the things we as a society have determined, whether by conscious decision or an unconscious decision, that have impacts on your health. So it is things like the average income of your neighbors, right? Are you renting or are you buying the houses or the apartments? Have most of the people completed high school, finished high school? So those are the things that we're looking at. And that is what I'm referring to as the indicators or predictors of how your health will play out.

(Joel Beasley at 00:13:09) So how do you get the health data then? Like, what do you use? Like, okay, I understand what you're discussing. Here's this public information. How do you connect that back to their physical health?

(Ram at 00:13:22) So the way that's done is, in many cases, claims data, right? So you've got the insurance companies who are, you know, quite a few of them are our clients. And so they kind of say, hey, this is where the person lives, and here's how we're looking at the claims, and we can then map it against each other. So not only do we have our index that's based on publicly available data, so we looked at all of the census data that's out there, which is coming from the government websites, and we download them, and we've normalized the data. And our data scientists look through it and try to come up with a zero to 100 score. So 100 means you're really in a very, very tough spot. We then look at the behavior of that person. Are they engaging? For instance, when I text you to say, hey, come refill your medication, are they doing that? So that's direct engagement data that we ourselves are seeing, right? So that's something we can measure. And then there's the claim that, well, they did actually pick up their medication. That's the hard data, the hard outcome that we're also going after. So on both counts, we're seeing that in many cases that there is a difference, right? Where if you have higher, if you're indexed higher on the social determinants of health, then you tend not to engage as much. You tend not to refill as much. And we actually wrote a paper about this with Kaiser Permanente on this because, you know, we had a very successful program, and it was reaching a ton of people, one of the largest programs out there for getting medication refills using text messaging. Again, tremendous success in doing that with Medicare members. But when we dug into the data a little bit more, we saw, hey, you know what? We could do better in certain areas, specifically around those that have the higher needs. And so we're now going in. Kaiser's already innovating in that area and looking to find additional ways of engaging, reaching out that first message. So just as an example of how we did this, we kind of built a predictive model around, could some of these factors play into whether you'll engage with our first message or not? And we found that once you engaged, once you send back a yes, I want to do this, then we level the playing field. Then everyone's equally likely to go get their medication from the pharmacy. But it was at first touch that a little bit of that hesitation, maybe a little bit of skepticism or cynicism around, hey, what do they want from me? Is this spam? Is this not? And I think that is where we're working on now. Like, how can we build up the first touch being that successful touch? So there, you know, these are very subtle challenges, but as you can see, the tech can help us separate these and get us a little bit closer to that better state where we're reaching everyone equally and everyone has the same chance at getting to take care of their health. And that's where I think the data that we have can really drive that a little bit more.

(Joel Beasley at 00:16:38) How did you meet, like, what's the story of you meeting the executive team at mPulse Mobile and making the decision to join?

(Ram at 00:16:47) It's an interesting story. The guy that had this idea, he and I were talking for like two years, and the company that he was working on at the time was more around mobile marketing, slightly different. And so I was like, I really want to get into healthcare, and that's really where I see my passion and what I'd like to focus on. And he's like, yeah, that's the vision. We're going to be doing that. And so I had talked to him, and we talked a little bit about some of their technical challenges and my experience in that and tried to come in and share some ideas. So we had a little bit of a relationship, and it turns out that he was able to get one of our clients—it's still our client today—as sort of the beginnings of the healthcare market that we were going after.

(Ram at 00:17:41) So we worked with this company called Humana, which is more Midwest. It's a large, very, very large healthcare payer. And the person that he was working with came on as well saying, hey, this is a cool company. He was our client. And so Chris Nicholson, who's our CEO, was heading up a very large $250 million unit over in Humana and was working with the company that started thinking about doing this healthcare thing in a big way. You know, I got a client here doing some use cases that were interesting, but not quite deep into healthcare. And so he was able to get a CFO, Brian Chadley, from WellPoint. Had a great experience, but also had Activision on his resume. So kudos to Jared, Jared Wrightson, who kind of had this idea and brought together Brian and Chris, and I was part of that as well. And so we basically said, hey, this is an amazing opportunity to take some of these ideas, make them real. We've got some traction. We've got some clients that are maybe on outdated technology. We can build some of the new ways of doing it. And we basically bonded on the fact that we wanted to make a difference in healthcare.

(Ram at 00:19:05) We want to do something interesting and exciting. We were tired of sort of just accepting that this is just how healthcare is, and we're not going to make any change. I mean, Chris, who had led a lot of innovation on his side, knew it was possible. Brian on that finance side had that strategy and finance background to say, hey, here's what people are looking for. So we didn't go in there naively thinking we just have to get in front of the consumers, we'll disrupt everything. We didn't go out there with a bunch of buzzwords, but real, very tactically focused, operationally focused—like, how can we move the needle in this? And I think that expertise that Brian and Chris brought to the table was just, I mean, to this day, they're very much part of the team, invaluable to our success. And having, you know, come with that depth of the financial strategy and the innovation side from Chris, I think was great. And so my job was like, okay, let's build some cool tech. You know? Let's make it happen. And so, yeah, I didn't need much convincing. Let's put it that way. I was on.

(Joel Beasley at 00:20:08) So you guys refer to it as conversational UI, right? That's the category that you play in, conversational UI and healthcare. So what's been like the progression of this conversational AI? Because like, at first, I was thinking, you know, I think my first introduction to it would be like an Intercom-like type chat or like a Drift, like something in the sales process. And then you get into the Alexas and start talking about the interfaces with the conversations there. And now we're in healthcare, and it looks like—correct me if I'm wrong—it looks like you guys build like a suite of tools, a suite of infrastructure, and then sort of customize the solution for your customers. And then you may have some products people can buy off the shelf too. So I'm just, I want to know, and we can talk—I've got more questions too about mPulse Mobile specifically—but right now I'm really curious, and just for me better understanding conversational AI, like what's the progression of it? Where is it at today?

(Ram at 00:21:06) Yeah, that's a great question. It's a really exciting question. I think we've been thinking about it. So let me kind of go back to a little bit of your earlier question. Like, how did we sort of get into this? And I love the story because, even though from a pure, if I kind of put my CTO ego on, it doesn't sound as, wow, you know, you had this vision, and everything sort of fell from there. Like, going back to what I was saying about value first. You know, if there's a problem that I need to solve, solve that problem. Don't make problems. You got enough out there. So the problem that we saw is, when we first started out, our initial goal as a platform was to send out messages. I need to remind you that you have an appointment tomorrow. And that seemed like a pretty reasonable thing to expect the system to do, and we were pretty good at it.

(Ram at 00:22:00) And our goal was, you know, can we crank up the pace? Can we get more messages per minute, per second? How fast can we go? But when you start doing that at scale, someone is going to respond back to you. They're going to say, oh, I can't make it. My, you know, I'm too busy at work. Or why are you sending this message? Or things like that that you just don't, you just want to tell them that they have an appointment tomorrow. It's as simple as that, right? The problem to use case is a one-way use case. It's a notification. Life doesn't work that way. Humans don't work that way. They would like to respond. In text messaging, the more and more we think about it, you know, when we started out, it's a two-way channel. You send a message to your friend. Invariably, she or he is going to respond to you. Right? And you'd be a little upset. Like, what? They didn't respond to my awesome, you know, message that I just wrote. Why didn't I get a thumbs up? And so it's just observing human nature to want to respond even to a one-way message led us to invest in this.

(Ram at 00:23:01) What are they saying? Why are they not making it to their appointment? Wait, we don't have to read their minds. They're telling us. It's in the text. It's literally that they don't think it's important, or they can't afford it, or they think that their car's broken. And you just look at it as, hey, what are people saying? And you try and understand them a little bit better. Before you can respond to them, you have to listen to them. I have to learn. I have to understand. And so our ability to say, hey, it's important for you to take care of your health should only come when I understand it, when I think that you don't think it's important. Right? If I sort of lord it on you and say, it's important, like, dude, I get it. I'm going to make the appointment. Just chill out. Like, you don't need to say it. So that subtle sort of understanding comes from not just what you say, but also all the other things we know about you. Have you not made appointments before? So when we think about our natural language understanding, it's not just that I can understand the words that you're saying. It's also putting those words in the context of who you are that's saying them. Right? As much as we can. Again, we don't want to overstate that. But if I know that you tend to have a propensity to cancel appointments, like you've just canceled and rescheduled, well, then maybe I should kind of nudge you a little bit more. Right?

(Ram at 00:24:32) Versus if it's just your forgetfulness and you make your appointments, fine. If you think that eating rice, white rice isn't a problem, even though you know you're diabetic, I need to educate you that what rice may have and needs to happen in your diabetic control. So it's kind of, again, understanding where your levels are from a pure medical perspective, but also the kind of thing that you're telling me that I can then incorporate into my, quote unquote, intelligence that I'm running behind the scenes to make that decision. So we built our platform sort of thinking about this in two levels. One, at the immediate, what is Joel saying to me? What do I respond in that sense? We built this, you know, dialogue management system that can have conversations sort of in real time. After a conversation is completed, what did I learn? What did Joel say? How did he respond? Was the sentiment good, negative? Try and look at that a little bit sort of at a higher level. Then we have this engine on top of that that does the tailoring. So the next time I talk to Joel, I might have a different conversation with him based on what he said in the previous, but also all the other conversations I've had with him. So, again, we're trying to bring all of that smarts from the immediate conversation of what I need, simplify that problem, solve it easily, but also bring to bear the next time I have a conversation with that historical or that learnings that I have or understanding that I've had built up.

(Ram at 00:26:21) So our profile, for instance, our AI is very much built on that dynamic profile that I build about you on a regular basis. Every time I send out a message, I'm continuously updating that profile and keeping track. So then that conversational AI is really around not just that individual messaging back and forth, but what conversations do I have so that we have increasingly more meaningful, compelling, engaging, relevant conversations, not just that are fun to have, but get you closer to your health and get you a little bit more, you know, engaged with the healthcare. So that's how we interpret conversational AI. That's how we've developed it. Like I said, it was from my need of just saying, hey, people are talking back to us. Is there an opportunity here? And then being able to leverage that with the technology. Broadly, I think, you know, when you look at beyond what we're doing and what others are doing, I think a lot of it has been focused on task completion. Right? You build something, you want to reschedule your flight, or you want to find out what your credit card balance is. So those are great opportunities for building these kind of chatbots that, you know, are very transactional in nature. You ask them a question. They'll say, okay, did you mean this? Did you mean that? I think that evolution has also been very impressive to follow. And as we see, you know, the investments that other companies have made, larger companies, like Googles, Amazons of the world, Microsoft, we're able to leverage some of those learnings where we're incorporating some of those findings and some of the AI that we use. So by all, I mean, there's no way we would have, we would be able to solve some of these problems if it wasn't for the work that these great companies have done.

(Joel Beasley at 00:28:10) Okay. So I'm following you on this, and I want to ask you some deeper questions. But how do you refer to like a specific style of conversation? Do you call it like a use case? Do you call it a domain? Like if we're talking about appointment reminders and you send the notifications and people respond and they're interacting back and forth and then it comes to some sort of conclusion. What do you call that?

(Ram at 00:28:34) We like to call it a conversation.

(Joel Beasley at 00:28:36) Okay. So it's like a type of conversation. Okay. So my next question is, how do you structure your teams or how do you structure your technology organization, like to have expertise in the different styles of conversation?

(Ram at 00:28:51) Yeah. No, that's a great question, and actually we've invested quite heavily in that. And that's what we call our behavioral data science team. They do the research of what makes for good conversations. So it's, I want to kind of pause there first thing. Behavioral data science. All right? With a little bit of emphasis. There are a lot of teams out there that do behavioral science. As you know, there are a lot of teams that do data science. We felt that the right combination to answer your specific questions like, how do you have these conversations? What are the kinds of things that people say? And how would you pick them up? Right? Our NLU or our natural language understanding isn't just, you know, very general. Like I was saying, that there's some domain-specific problems that we're trying to solve that, you know, when we're talking about the sentiment analysis, we just couldn't solve it with what's out there. We had to build our own. And that's a very much, you know, that department that we've started about two, three years ago now, has been a huge part of our strategy and part of why we think we're successful. That investment of not just, you know, conceptually that there is such a thing—we feel very proud of having kind of taken a charge in doing that—but also the people that we've been able to bring on board and the kinds of ideas and the learnings that we've been able to work with our clients on and kind of co-develop, that's what makes those conversations more relevant. Right? It's that understanding that gets us there. So if, you know, if you want to get into the specifics, I'm more than happy to. But there's a fair, you know, it could be a longer conversation, but happy to dive in.

(Joel Beasley at 00:30:37) Well, what were you doing before that, and how did you identify that this needs to be extracted into its own team?

(Ram at 00:30:44) We were doing it. We just weren't calling it that. You know, there's an aspect of what we were doing. Well, okay. So evolutionarily, right? So our biggest problem was we're in a healthcare environment, which is highly regulated. People are very sensitive to if I send you a text message and you don't want to receive that text message, I should immediately opt you out. Right? And you indicate that to me. Similar to our, you know, do not call list and things like that. So fortunately, for the text messaging channel, for SMS, it's well protected and it's been around for a little while. So we started off just saying, hey, people are telling us instead of using the word stop, which is the traditional and, you know, from a requirements from the act that's out there, CTIA, which governs some of this. They say, hey, if you text stop, you must adhere to that. But people don't remember to text stop. They may say something like, why are you bothering me? I don't want to receive these messages. You don't see the word stop in there. You don't see anything that, you know, is compliant. And yet, it would be a very poor user experience if I just told you, I'm sorry, I don't understand that. Try again. Right? And we wanted to kind of get to that as quickly as possible. And so our first two things, actually. Our first thing was, let's make sure we are very proactive and accurate about any intent to unsubscribe. Because when it comes to my standpoint, we want to be always on the side of the user experience. Make sure that they feel protected. So that was one. The other was we also saw this. I'd send you a reminder, as I keep going back to the example. We were doing more than that, but let's, that's just a nice simple example.

(Ram at 00:32:34) I send you a reminder, and it's just a notification. You have an appointment 10:30 tomorrow with Dr. Smith. Thank you. I didn't expect you to say thank you. And I'd say, sorry, I didn't understand that. This was us, you know, many years ago. And Chris Nicholson and I, our CEO, had this kind of debate. I'm like, why are we saying sorry when someone's saying thank you? Like, can't you just, you know, fix that? And so, well, okay, I could. I mean, yes, we could hard code the response to thank you. That's easy enough. But I think we should solve the problem broadly, and he and I chatted about that, about do we want to start understanding beyond? Thank you. Thanks. Thumbs up. That was great.

(Ram at 00:33:22) Thanks for reminding me. All of these things. And then as we started doing that, we started understanding not just what people were saying, but again, the intent, why they were saying it, the behavior behind it. And so then the concept of behavior science came in. Are people saying things that mean that they will do it? Are people saying things that mean that they aren't ready yet, they need more education?

(Ram at 00:33:45) So then if you think about behavior change models and some of that theory that comes from that, we're able to read from that literature and that research and say, "Hey, wait a second. This is what's going on here. I have people who are not ready to change. I have people that think that flu shots are, you know, might cause them to get the flu."

(Ram at 00:34:04) So their health beliefs. So we started looking—when I say "we," the team, I didn't personally, but the team started looking at that. And we're able to draw very quickly from this body of literature on behavioral science, which exists and has been around for a while. But in addition, we said, "Well, we're cranking through millions and millions of messages. So can we pair that back with what we're seeing from a data science perspective?"

(Ram at 00:34:30) And so we were starting to do that naturally. Things like sentiment analysis, it's a very data-driven classifier that we have. So at some point, we said, "Let's call it behavioral data science because that's what we're doing." That sort of formal statement probably came out about three years or so ago, if memory serves.

(Ram at 00:34:50) But it happened organically. It happened, I think, very much with, what are people telling us about learning, that listening, that sense that we need to pay attention before we make brand statements. And then one of the things that I think is very helpful for us as a company is we're always looking to learn from others, our clients, other research. So we quickly went to the literature and looked at what's out there and tried to learn from that. And in some cases, we'd come up empty, and in other cases, like, "Wait, there's actually some good stuff here that we can learn from." So that's kind of how our thoughts sort of formalize and crystallize into this behavioral data science approach.

(Joel Beasley at 00:35:40) So do you guys do things like if I want to book an appointment, I can book an appointment or it's like a conversational interface on the health care provider's website? Do you do that yet? Or discuss or share with me the range of products and solutions.

(Ram at 00:35:58) Yeah. So you mentioned something about off the shelf and all that. I mean, we are working with very, very large health care clients, enterprises. You know, I mentioned Kaiser, Humana. There are other large companies that we work with, and so oftentimes, they have very complex problems. We're working in tandem with other engagement solutions they have.

(Ram at 00:36:22) So while there's a good chunk of work that we can, kind of based on our overall platform capabilities, we can bring to bear. And what we've done is these solutions that we have, whether they're getting you to refill your medication, whether it's taking care of your diabetes because you got a recent lab test that says, "Hey, you're actually controlling it, keeping you controlled," which is a very different type of problem that we're going after, or it's like you were saying, scheduling appointments or rescheduling appointments, getting you to—we did, we're doing one where we know that the opioid crisis has been very devastating, had a devastating impact even for a while now. And one of our clients is working on sending out these pouches to people who have been prescribed opioids. But once they're done with that treatment, they can put the medications into this pouch that basically dissolves them and makes them ineffective and not potent anymore.

(Ram at 00:37:34) And the reason to do that is to sort of reduce the potential of these opioids sitting in the house that someone may use. Right? So we can, again, try and give tools. In this case, it's a pouch that we have nothing really to do with. And we were brought in saying, "Hey, can you remind people to use the pouch? And we'll be giving incentives." So that's another yet another problem that we're going after. When COVID-19 hit, we immediately kicked into gear and started putting together ways, things that you could do, and we actually worked with an illustrator to come up with a photo—what's called a photo novella, or it's like a mini graphic novel of things that you can do to prevent the spread. Right?

(Ram at 00:38:18) And so we went to our clients and said, "Hey, look, let's get this out there." So you have a little message that comes to you and says, "Hey, we built a little story about keeping yourself, keeping safe." They link over, and they look at this graphic novel, which is just three or four scenes about washing your hands, keeping the distance, you know, trying to stay away from crowded places. And so that's yet another thing that we do. So underlying all of these are these capabilities that we can assemble very quickly towards a specific problem because we have our messaging engines, the dialogue management engines, the conversational tailoring engines that I was talking about. These are sort of layered in, very much a platform model, a SaaS platform where you can configure them to your needs. So when we talk about building custom, it's really configuring for your specific use case.

(Ram at 00:39:13) And in certain very common things like medication adherence or chronic disease management or getting them to screenings that they need, what's called gaps in care within the health care industry. We've built sort of almost ready-made solutions, and you can say, "Hey, I wanna use that immediately." Like, here you go. Here's what the content looks like. Here's all the smarts that go into the natural language understanding, but the main specific things, the tailoring. And you can customize a little bit. You know, you go in there and spend some time with our account teams and boom, it's launched. So there are those versus—and like I was saying, we didn't expect COVID-19 to hit us. And this was back in March where we were thinking, "Okay, it's gonna hopefully get over soon." You know, we can try and, if we do everything the right way, and here we are in November still struggling and it seems worse now. So, you know, some solutions we think are maybe just at a point in time. Others, we know that that's just part of health care. Everyone, you know, flu season, for instance.

(Ram at 00:40:25) We changed our flu solution this year, partly because of COVID-19. We think, you know, you may have heard the problem where you've got the COVID-19 crisis on top of a flu, and maybe a flu virus as well. So getting that flu shot in is at least, you know, until we wait for the COVID-19 vaccine to come in, we can at least try and nip one of those things in the bud as much as possible. So we're working with our clients to get the word out so you can get the flu shot done. That's a start. And so our platform allows us to engage with our clients to do all of these types of things.

(Ram at 00:41:08) It could be, like you said, very simply, "Hey, I wanna reschedule my appointment with so and so." We do that too. It's also—we don't integrate with smaller, like, a dermatologist on the street. We're integrated with large enterprises that have systems like Epic, which is the biggest EHR, electronic health record system out there.

(Ram at 00:41:36) And so we can get data from there and push data back directly into their scheduling systems and things like that. Yeah. So it does run the gamut, but our—I look at my team's job is to make these capabilities easy to use, and not just from a configuration standpoint, but also from a reporting analytics standpoint. Did it work? Did it not work? You know? So that's very much part of what we focus on and make sure we get that back to the clients.

(Joel Beasley at 00:42:10) So going from like 3,000,000 conversations to 300,000,000 conversations a year, what did you learn from that as far as leadership perspective and organizing the people and the technology? What's your big takeaway from the current scale?

(Ram at 00:42:26) Wow. It's a great question. And I think, you know, architecture matters. Making sure that you do not solve—you don't build an architecture around solving today's problems. You build an architecture around solving future problems. And I think the quote that I really resonate with when you think about the term "architecture," in our context as a CTO, it's the decisions that you make that are very difficult to undo later on. Right?

(Ram at 00:43:01) So today, I make a decision that seems really good and right, but always understand that if I have to undo it later on, then you're making an architectural decision. And so one of the things that we've also thought about, we said, "Look, we may be a small company today. We may only have 3,000,000 interactions or 3,000,000 members. But we think we're gonna be successful."

(Ram at 00:43:29) Right? I mean, this is why we're here. We're not here to work with a hundred people. We're here to work with as many people we can get to because we believe in what we're doing. And there's a little bit of kind of a sense of, like, we have a responsibility to grow this. Right? That's kind of also where we're coming from, that purpose. And everyone on the team that we've hired is very passionate about health care. That was the other part that I think is important. But do you think of engineers just wanting to write more code?

(Ram at 00:43:57) Well, a lot of them wanna write code that makes a difference, and I frequently get some engineers coming up to me saying, "Hey, Ram, I'm not seeing how my code made a difference. Like, explain that to me a little bit more." Well, if you don't tie it back, and I think you will lose some of that momentum and steam that some of the team needs. But going back to the question, we've tried to work our architecture in such a way that I can scale without investing everything today because we didn't have the money today to fire up an unlimited supply of servers even though we could, and that's an investment of cost. Right? But most importantly, it's an investment in a certain architecture that allows you to scale, and that's, I think, where we've been fairly successful. Again, did we have to refactor? Did we incur some technical debt? You're not gonna hear me say no. We were perfect. Of course, we did. Of course, we had to go back and redo stuff.

(Ram at 00:44:47) And it was painful. That's for sure. But I think it's having that mindset that, "Hey, don't assume that this is always gonna just have this level of capacity." So one of the things that, as an example, like, when you think about a response time and you put in place, like, I expect a certain response time to be less than ten seconds or less than a sub-second response, depending on what the particular thing is. You can say, "Look, given this quantity of or this volume of requests that I'm gonna get, this is fair."

(Ram at 00:45:26) Our immediate question is, what happens if I double that? What happens if I 10x that? And if the answer is I don't know how to get there, go back to the drawing board. Figure it out because if you can't, then it's not ready. If you can convince me that this is a proof of concept that we will throw away and we're willing to invest in redoing it, fine. We will do that. We will invest in, you know, pilot projects that do not scale, but then we cordon them off. We say, "Hey, that's not our platform. That's a proof of concept."

(Ram at 00:45:56) So I think you have to have some discipline around that, and then you say, "Hey, those systems that we have, we've sold to our clients and that can scale, we have to answer these questions every time before we can do it." So load test and performance testing even beyond the capacity in which you expect, I think it's absolutely necessary. And then there's a willingness and the humility to just say, "Yeah, we didn't do it."

(Ram at 00:46:20) So you roll up your sleeves and get that 3:00 AM call where you're like, "Oops, there's smoke here," and doing what we can. I think the other part of architectural decision right from the get-go, we're on the cloud. We're an Amazon partner. We invested in that. We knew that that was gonna give us much faster scale than if we had to build our own data center. And I was involved in doing that in the prior. It just doesn't seem like that's the smart thing to do when we have so much growth in front of us. By having an infrastructure where all that can scale with us, having software that effectively is written to scale by virtue of making it as much service-oriented as possible. So individual components can scale it based on their capacity needs.

(Ram at 00:47:09) I think that's the other thing. And then testing that out as much as possible with as much rigor as we can with as much future planning as we can, is the other part of just, you know, getting to that 3 to 300. But I will tell you, it was painful in the very beginning because when you're strapped for cash and you're strapped for resources, we don't have a ton of engineers. You're gonna cut some corners, and we did. And, you know, we got lucky. I think that's fair to say.

(Ram at 00:47:40) We got lucky, but we also had an amazing account management team that was able to work with our clients and straight up tell them, "Hey, look, you know, we're working on that capacity." And five years ago, we had very tough conversations, and I had none of those kinds of conversations now. But, you know, as we look through the—as we continue to scale, there are things that we're always—now, it's really about monitoring. Am I noticing things that, you know, before things would crash? Now things kind of do things in very different ways where there may be a creeping, you know, memory leak or something on a system that's not used very much. Now we're sort of like, "Okay, hey, gotta get it though."

(Ram at 00:48:24) So our ability to monitor and build in the alerting systems, our own dashboards, is where our investment is today to just know when we have to add capacity before it starts to fail. So understanding those patterns earlier on is where we're shifting some of our infrastructure effort. And so our engineering team is really looking at, you know, what are the monitors that we have—you know, we throw this word around. What are the known knowns and known unknowns and unknown unknowns. Right? So we're trying to reduce the unknown unknowns by increasing the known knowns. You know, it's just like a numbers game at this point. So, yeah, that's how I think we continue to scale.

(Joel Beasley at 00:49:11) I like that. I was talking a little bit with this company called Gremlin about how they're approaching site reliability. Have you heard of them?

(Ram at 00:49:18) No. I haven't. No.

(Joel Beasley at 00:49:20) Yeah. They're pretty cool company. It wasn't necessarily exactly what you were describing, but the guy had been over at Amazon on their site reliability, and they had this guy that would run through the server rooms and just rip out the cords and then see—

(Ram at 00:49:36) Chaos engineering?

(Joel Beasley at 00:49:37) Chaos engineering. Yeah.

(Ram at 00:49:39) Yeah.

(Joel Beasley at 00:49:39) Yeah. And they're like one of the cooler emerging players in the chaos engineering. But what got me is, you know, as an engineer, the way that I had seen a couple of these, but the way that they described it where you actually have the specific scenario and then you test against it and then it goes into your CircleCI essentially, like that type of build-deploy concept where it's always testing these different scenarios. And it was less—I think it was less about the technology and more about the methodology of how they approached it.

(Ram at 00:50:10) Yeah.

(Joel Beasley at 00:50:10) They kinda went hand in hand, you know, like a pivotal tracker or something.

(Ram at 00:50:14) Yeah. Yeah. But it—

(Joel Beasley at 00:50:15) It was quite fascinating, and so I was talking with a couple people about it and I brought it up to Jason over at GitHub and he's like, "Yeah, they're on our radar." And I was like, "Oh, okay, cool." So sometimes when my spidey senses start to tingle about something that's really cool or interesting, I start checking it with other people and it's—

(Ram at 00:50:33) No, I think we've looked at chaos engineering and actually, you know, on the security side, we invest quite a bit. And that's one of the things with health care, as you can imagine it. And I think every industry nowadays, I mean, we don't need to go, "Oh, this is health care." I don't think you'd want to know your Facebook data to be out there anymore than you'd want your health care data.

(Ram at 00:50:54) You know, they're equally important to you. But because it's a much more regulated environment, we have to sort of prove and go through these certifications and third party attestations and things like that. But one thing that was really interesting, which I went through an exercise, is what's called a tabletop exercise. And what we do is you hire an outside company that sits down with you, and it's like this horror movie that gets played out in front of you. Like, okay.

(Ram at 00:51:21) You're getting, you get this email that's about a ransomware attack. Like, okay, how would you respond? And so we had our CEO on there, our director of compliance and our engineering. You know, we had all of the leaders in the organization, and we had to kind of play out like, what will we do?

(Ram at 00:51:38) You know? And you build that sort of sense like, oh, wow, we don't really have a good answer there. We need to get better about that. Just a simple thing like communication. Like, hey, wait, do we tell our clients? We gotta do that, and who's doing that? In some cases, it's very trivial process. In other cases, it was like, we need a system to help us pick that up faster or learn that. So what they do is they just inject new events into this three-hour-long process.

(Ram at 00:52:10) And at the end of it, you're exhausted. You know? You're like, "Oh God, what did I just go through?" And yet you learn a ton, and the whole team just gets a huge amount of value.

(Ram at 00:52:23) And so those are the kinds of things also, you know, just because they are definitely theoretical, right? And we're not actually writing code or no one's doing a pen test on us or anything like that. But those kinds of analytical exercises are also hugely valuable. Like I said, it's about sometimes the methodology than necessarily the technology that you're employing.

(Ram at 00:52:45) The other thing, and I'm very proud of this compliment that I got from someone aside—that's why I want to share it here, I'm gonna brag—but one of our clients, their chief information security officer said, "You know, Ram, the thing I like about what you guys are doing is you've incorporated security as a culture." And I'm like, wow.

(Ram at 00:53:07) That's it. That to me is the highest level of—because, you know, when you talked about how do you scale, right? There are certain things, and I think culture is part of your architecture. And so when you think about capacity planning, when you think about how you want to scale, if that's not part of your culture and trying to do it after the fact, man, it's painful. It's rough. And I've been in environments where you just literally say, "I'm just gonna rip this out."

(Ram at 00:53:33) Right? There's no other way because you're retrofitting culture or you're retrofitting something that isn't fundamentally there. And so security, scale, growth—if that's not part of your culture, if that's not sort of inherently how you think about problems, you know, methodologies will flow. You think less about those problems, or your reflexive nature is to reject the findings that, you know, you're looking for confirmation bias.

(Ram at 00:54:08) I will be able to—you know, I don't have a capacity problem because I'll only use the solution for this much. If you think in those terms, that's a reflection of, I think, potentially a culture that's not built around scale, around security, around growth. And so that's, you know, it's getting to what you're talking about. The Gremlin CTO seems to be like they're trying to get companies to think about it with chaos engineering, and great idea. Like, you know, hey, pardon the term, but shit's gonna happen.

(Ram at 00:54:40) Right? It just is. And if you think that it's not gonna happen and you somehow can write amazing code, you're not thinking about it right. You're not fundamentally—that culture of humility is missing, and you're gonna have to pay that price.

(Joel Beasley at 00:54:54) Yeah. And you have low experience because it's—

(Ram at 00:54:57) Or, yeah. Things break since day one. Yes, exactly. Right. So absolutely.

(Joel Beasley at 00:55:04) We did it, my friend. We made a podcast. How do you feel?

(Ram at 00:55:06) Oh, we did. Okay, yeah. Awesome. This is fun. This is really nice. Thank you.

(Joel Beasley at 00:55:12) Was there anything that we didn't get out there that you want to get out?

(Ram at 00:55:16) You have another ten hours? But no, no. I think I really enjoyed this. I mean, this is obviously something that we live and breathe, and still being able to talk to someone out there. The one thing that I always do is, if there are other CTOs that listen to this or hear this—I'm sure they are because you obviously have a great podcast—and appreciate you having me on, but I'm always looking to learn. I always look to learn from my peers. And so, hey, reach out to me and, you know, get in touch with me and tell me what I'm doing wrong or thinking about things differently that I should. And we as a company, I'll say our biggest asset is our willingness to learn. And I think, you know, that's how I approach my job. It's like, "Hey, what do I not know? All right, so keep looking." So that's my final ask, if you will, is for those that are listening. Hey, come talk to me and tell me what I should do differently or do better.

(Joel Beasley at 00:56:12) Excellent. Yeah, we'll put your information in the show notes. It was a fantastic conversation. I look forward to talking with you in the future, Ram.

(Ram at 00:56:18) Yeah, thank you so much. Thanks to everyone on your team as well for making it happen.

(Joel Beasley at 00:56:22) Thanks. Have a great day, buddy. Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email: [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.