052 - Agent-Based AI & Nation-Scale Digital Twins: Future Decision Making w/ John Cordier, CEO, Epistemix

John Cordier: Those models run, you're able to get the results back, deliver those results back in a more interpretable way for decision makers. And so what that's effectively done. It's reduced the time to value for customers from months to within a week, which is a like, it's a huge deal for, for this agent based simulation industry.

Chris Fanchi: That spark of innovation, it's just waiting for the perfect combination of tools and talent to bring it to life. Welcome to Pitch Build Scale. I'm your host, Chris Fanche, a recovering startup founder, veteran SaaS marketer turned dev tech enthusiast and growth specialist. Each episode we connect with the builders, the visionaries and the problem solvers who are redefining what's possible with today's technology and pioneering the websites, apps and software products of tomorrow. These are the conversations that transform how we build and how we sell and how we scale technology businesses in a world where the tech evolves by the minute, ready to stay ahead of the curve. This is Pitch Build Scale.

Chris Fanchi: Hello and welcome to today's episode of Pitch Build Scale. For those of you just joining the podcast, we are Big North Marketing and specialize in helping dev shops and SaaS companies scale their revenue. So if you want to learn how to take your SaaS or dev business to the next level, make sure to stick around. And if you're interested in a free growth consultation, just drop us the word growth in the DMs. We're really excited to have today's guest, John Cordier, CEO of Epistemics, where he's transforming public health and market decision making through Nation scale, digital twins and agent based simulation. John, welcome to the show.

John Cordier: Yeah, Chris, thanks for having me on.

Chris Fanchi: So to get started, I always like to ask my guests, what is their origin story? How did you first get into tech and then specifically public health and data science?

John Cordier: Sure. So I'll give my story and then the company origin story, uh, and how they both converged. So, uh, in early 2000s, my co founder, Don Burke, he was leading disease research for the US military. And then he left and was running all of the infectious disease programs out of Johns Hopkins. And one of the things that he started when he was there was this mathematical modeling group that's called midus, um, which stands for the modeling Infectious Disease Agent study. And Don, in his time there, uh, identified that of the tools that existed at the time and even today, really public health leaders, government leaders, leaders of health plans, big insurance companies, they all struggle to understand the timing and planning of what to do when there's a new epidemic. Of sorts, when you don't have previous data to rely on, how do you assess what to do? So Don paired up with our, uh, other co founder, John Grefenstedt. Don became dean of the School of Public Health at Pitt. He recruited John to the University of Pittsburgh and they set out to build this digital twin or synthetic population of the entire United States and then coupling that digital twin or synthetic population with a simulation and mathematical technique called agent based modeling. And they did that because in order to understand how epidemics evolve over time, you have to understand the changes in human behavior and how human behavior will change under different conditions. Those conditions could be a change in health policy. It could be a new technology that's getting adopted, could be the spread of information, new products that are getting launched to improve health, or possibly have a negative impact on health. So from about 2004 until 2018, there was a bunch of R and D efforts at the University of Pittsburgh, Carnegie Mellon, and a few other schools that were trying to build this system to make agent based simulation possible for public health practice. When I was figuring out what I wanted to be when I grew up, I committed to reading a book a week for a year in one subject to kind of have ah, that be my litmus test of do I care enough about this? Am I continually getting excited about this topic enough to go back to do a master's degree, PhD and commit a career to it? So after a few years of working after school, uh, like undergrad, I landed on public health, Social determinants of Health, Behavioral economics, and uh, sort of like decision science. That's what I was continually getting excited about. So I was at the University of Pittsburgh for grad school. I was doing a master's of Public health and an MBA of the Normal Health Policy Practice Health Policy program. Went to my department chair and said, look, like dropping out, this isn't my, this isn't the way that I thought the program was going to go. Um, I don't really feel like we're changing the way that public health could be practiced in a bigger way to impact social determinants of health. That conversation led to me being in the Dean's office six or eight, six to eight weeks later with the opportunity to say, like, hey, we're thinking about starting a company around the one technology we learned about in our managerial epidemiology course. And I was like, oh, that's great. Like, this is like the one transformative thing that we've, like, that would be like, able to do public health differently with so uh, we don't know how to start a company. We were just told to get a grad student who is willing to get told no a lot and not give up. And that was me, I guess like eight years ago. And then finished the master's program while basically doing the company full time and then in 2019. So for the last six years it's been a, uh, full on effort in getting epistemics commercialized through the tech transfer process, through getting our early customers through building the team, raising venture money, and that's what I've basically been doing the last six years. So I'm the kind of like grad student, not going to do a PhD going forward type character and kind of found my way into entrepreneurship and company building by given an opportunity to pursue something that I really care about, which is improving the way public health can be practiced. So yeah, I'm um, grateful for the opportunity. And you know, there's still a long way to like see the original vision of myself and the two co founders come to life. Our, our goal is any local public health agency, state public health agency, federal governments, whoever they might be, can run simulations to test the impact of different policies in a way to improve the socioeconomic and health well being of populations. Some people kind of classify us in the category of we're building the real world version of SimCity and then giving that tool so that people can actually make a difference with it. So I um, yeah, what we do is a lot of fun. I enjoy it and excited for what the next six years have entailed for us.

Chris Fanchi: Yeah. So you, you use the term digital twin. Can you talk a little bit more about that? What does that mean and how is that impacting what your business does?

John Cordier: So uh, the concept of a digital twin is really the digital representation of something in the real world. Common use cases of a digital twin, like before a house gets built, you have a 3D model of it. You could say that's somewhat of a digital twin of a house. People build digital twins of airports and malls and other things to understand how people are going to move and flow through it. Um, in our case, this digital twin of a population is this population of every person in the US we have places where people go, so schools, households, workplaces, all modeled out for the entire United States. And we're doing this for other geographies too. And the cool thing about a digital twin is it's supposed to represent the behavior of the actual system in reality. So in our case, our digital twin of a population, we have daily patterns of life. We have this concept of ages and stages. So populations evolve over time. So people age as people age. Some, um, people might age out of school and start working. People might age and have kids. People might age and pass away, People might age and get divorced. All of those things are uh, parts of what a population of people, like what happens in the real world. So our digital twin is our best representation of these social processes that go on all around us. People, where people go, how people spend their time, the products that people adopt, the diseases that people end up getting over time, those are things that people use our software to model. And by getting a good model of those things, you can then start to test using this digital twin concept, how does that population behave under different conditions. The same way that when somebody's like building a model of a car, you want to understand how that car is going to perform in rain, how that car is going to perform, perform in snow, in winds that are blowing across the car when you're driving on a highway. Um, in our case from a population digital twin, you're trying to understand what happens if this policy gets put into play. What happens if um, there's a new, like in the case of COVID like we have a novel disease, like what's going to happen with the population? What's the sentiment to adopt a vaccine or the sentiment to adopt a new technology? What's that going to be? And so using this digital twin, companies, uh, that we work with mainly in pharma and in health insurance, health benefits today, they're testing product adoption, they're testing utilization, um, they're testing other strategies to improve the health of populations to show a longer term ROI of investing in preventative health measures. So yeah, those are a number of the things that we're doing today. In the future, the same principles of epidemiology, which is how something spreads in a population. We'll get into consumer product adoption. That's a vertical we're really excited about. And then of course there's the government use cases using the SimCity for the real world reference.

Chris Fanchi: So currently are the bulk of your clients in government or are there other verticals that are adopting this technology?

John Cordier: No, pharma, uh, is the biggest vertical for us right now. Pharma and health insurers, their challenge is anytime that there's an existing product, new products are getting launched all the time. How do you manage a portfolio in the immunology space? New product launches for pharma is like a really great use case for us. But Even for existing products in the market, trying to understand impacts of sentiment, change to revenue, overtime is a big area for us to forecast. And then when it comes to health plans, health plans are trying to understand how to improve the health of their population that they're managing, while also trying to forecast here are the needs of this population, what is the utilization of different products and services going to be? And so doing utilization forecasting and then helping, uh, the health plans and also the employers that the health plans are working with, how do they put in programs to improve the health of the people that they're aiming to serve and show the ROI of both improving health, understanding, utilization rates, and the combination of those two things being economically viable for both parties. So that's some of what we're getting into today, for the most part.

Chris Fanchi: Okay, so how large is the company today, what's your team size and how many different products do you have on the market?

John Cordier: Um, so we're 25 people right now. We have two of our, like our two flagship products. One is called Populous, the other is called Polaris. Populus is the digital twin of the population that enables people to add data into it, connect different data sets so you get a really rich picture of the population. So that's kind of the standard a lot of data science teams use that you don't even have to do agent based simulation with it. You can use it for other types of data science techniques. Product is Polaris. Polaris is the scenario planning and modeling offering that a lot of times they're coupled together, but Populous is usually what, what people buy first and then they're like, oh, this is a really valuable data set. Now we want to start running scenarios. Polaris is the next product that people end up utilizing. Apart from those two, we have specific modules for different types of industries. One module is called Epidemic X. So the ability to forecast any new disease in a population. And you can do that by changing different parameters and variable that help you better understand this new disease. How might it transmit across the population under different scenarios?

Chris Fanchi: Okay, do you have any interesting anecdotes about your company and what you found or what your, your clients maybe found during the COVID era and any impacts that your technology was able to have?

John Cordier: Yeah, I think the biggest impact we had was getting the global events industry live months ahead of time. If you look at the ability to get people together at the super bowl, the Consumer Electronics show and whatever your big industry convention is like, those were not happening in uh, 2020, 2021 and even parts of 2022. Uh, so I think the biggest impact that we had was helping that industry find ways by running these scenarios and by understanding the demographics of who's showing up at what events, the geographies of where different events were taking place. Like it's a pretty big complex system and we were able to model all of those components out and then say, all right, if this event is happening in uh, Orlando, Florida at this time of the year, with this demo, these demographics showing up, here's the impact this event has on the local population and here's what you would expect of like gathering these people together. How many people are going to get sick? What can you do to prevent people from getting sick? All those sorts of things. All in all that created about a couple of, I uh, think it was estimated at 4 billion that we were able to return to that industry months ahead of time. So that's a pretty big deal for a lot of those companies. We worked with schools during that time, we worked with colleges and universities, uh, we worked with some companies on their kind of get back, like return to office type planning. But yeah, I think the biggest economic impact we had was with the events industry. And the cool thing from a public health perspective is we got so good at doing those models that we actually beat a lot of the CDC forecasts months out. We were able to pick up on when one strain would kind of, kind of die out and another strain would kind of become the dominant strain. There's another set of things that we got into where companies or and colleges, universities, like they would try to beat the benchmark that we gave them. So as an example, uh, there's one university in the Midwest that they worked with us, um, and they wanted to say like, okay, we want to get all of our students back onto campus, but we also don't want to like mess up our college town with thousands of people getting back together and that leading to like some super spreading event that was negatively impactful all over the place. So we gave them a forecast. They were able to like say like, okay, if we get everybody back under this set of conditions or uh, under an alternate set of conditions, what are we going to do? They were like, what would the outcomes be? And for the first couple of months they're like, nah, we're not going to listen to this epistemics team, like the heck with that. Then they kind of saw themselves running down one of the bad paths, um, of how the outcomes could have played out, made some changes, got back within like a normal um, amount of cases that were popping up with the given precautions that we're, we suggested, uh, they put into place. And the rest of the semester they're like, you know, we'd get a call every week, hey, you know, we were one less case than what we thought we were going to be. Uh, hey, we were, you know, two cases more. Hey, we were four cases less. And it just became this, like, we gotta beat this epistemics projection. But in effect, it kind of, it created a way for decision makers to be able to interact with data in a new way that they weren't used to. Um, and how that impacted decision making, which ultimately impacted the lives of people not only on that college campus, but also in their town, was kind of a cool example of actually using data to help drive decision making.

Chris Fanchi: So how much do you go back and look at what actually happened and compare it to what your models predicted to see and tweak your models, make sure everything is working the way it's supposed to be?

John Cordier: Yeah. Uh, so every time, um, that, like, if we're any good at our jobs, like, we, we obviously should be doing that. We've also been able to build systems where if we're connected to the data set that we're getting measured against. So in like the consumer space or in the pharma space, a lot of times it's product adoption. Like usage could be utilization of services, number of units purchased, those types of things. We can calibrate the models to those data sets. And so you can see in real time, like, here's the data that we have coming in. So that's like an observability thing that a whole bunch of other tools are good at. But those things can inform what trajectory you are from a forecast perspective. And so, yeah, we reevaluate, remeasure. There's plenty of times when some things have happened outside of what was anticipated, and that's like really messed up a forecast that happens a lot more in like, health rather than in actual product adoption. However, using like this approach to do more behavioral forecasting, you know, we've already shown that like, it can be an improvement on traditional actuarial methods. And it's just like another, another tool to have. And it's complimentary. It's not trying to like force the way that actuaries have been doing this stuff for hundreds of years out. It's just like when you have a change in behavior or you have a change in the environment that people are operating in, you just need a new way to look at your forecast and your Scenario planning, digital twin of the population and synthetic or uh, and digital turn of the population plus the ability to do agent based simulation on top of that is a good alternative or good addition.

Chris Fanchi: So you were obviously around well before the explosion in the LLMs and in the generative AI era. How has that technology impacted your business? Is that a part of what you guys do now? Did you work generative AI or LLMs into your technology? And how has that impacted adoption of your technology?

John Cordier: So LLMs and like this current AI craze, it's impacted every business. It's just like how thoughtful do you want to be about it? For us, uh, from a product standpoint, LLMs are making our type of simulation more accessible to people, therefore increasing our total addressable market. So it's a big positive for us. The way that we've done that is we've been able to train an agent to build agent based models. And so you can go from a question that people ask into a set of models that get generated that map to that question those models run, you're able to get the results back, deliver those results back in a um, like a more interpretable way for decision makers. And so what that's effectively done, it's reduced the time to value for customers from months to within a week which is a, like that's a huge deal for, for this agent based simulation industry. The accessibility part is really important, but it also enables decision makers to interact with data in a new way. And so let's say you're just using ChatGPT. You type in your question, you get some text back, you don't really have an underlying diagram representation of like where all these different words, how do they show up on the page? With agent based simulation you have a data model and that's like a totally transparent and interpretable data asset that gives people the ability to understand like how this system or how this uh, how um, this process is put together, what are the inner workings of it. So the ability to help people get to that point quicker is a really big deal if we go on the sales side. So I think LLMs have been in some ways like more detrimental to most sales organizations. In some ways speeds up a hell of a lot of things in others. But there's just like a hell of a lot more noise today than there ever has been. And a lot of like the, you know, people are infinitely creative but sometimes people are asking LLMs to write the same stuff over and over and over. And so I think there's just like more, more medium or average quality content that's getting produced and as a result it's just causing a whole bunch of noise in the sales process for not only Epistemics, but a bunch of other companies too. So I think um, a thoughtful use of AI can help speed up some mundane tasks. But at the end of the day, like we all can pretty much tell when an AI has written an inbound email that we're getting from a stranger or uh, something else. And it's just like, okay, there is still an important element of like understanding. Like there's a person, that person's going to make a decision, not that person's avatar in the context of like LinkedIn or something. Like so yeah, I think there's, there's been a lot more noise generated and I think it puts a higher value on just like the human to human interaction that you have in enterprise sales.

Chris Fanchi: How are you seeing policymakers reacting to your technology? And has that changed since the AI revolution, so to speak?

John Cordier: Uh, I would say it's definitely changed over time. Like whenever a new technology comes out, there's going to be a set of people who are very skeptical about it. But then as the use cases become more humanized and more relevant to their work, then they're like, oh, like I can adopt this, I can accept it. They might not be doing the same amount of diligence in fact checking or uh, those types of things over time. But for the most part I think as AI has become just more common in different types of businesses, like you have to refine what the use case is going to be, of course. But I do believe policymakers are more open to the idea of having an AI or data assisted way of making decisions. And from a uh, policymaker perspective, a lot of times the interaction we get like, and a lot of startups do this, you're always pitching, here's the great virtue of our technology and what we've done. Well, in the back of that person's mind, like it's like, okay, cool, like uh, I'm hearing all the good stuff. Well, what's the bad stuff? What's the downside? Risk. What happens if this doesn't go well? And it's interesting that one of our use cases of our software is running through scenarios to understand the unexpected outcomes. And so when like you're running these simulations, you might run one scenario 100 times, you're going to get a hundred slightly different results, but you can at least try to understand what made these results go better or go way worse than what's like the average or most common outcome. So when policymakers, and this is just me presenting back what, what one of them said, he's like, look, John, you know, I'm an accountant, all right, by training now I'm a state representative. And um, you know, we're trying to do our best here, but like a lot of times we're aiming for outcomes A, B and C. And like A never happens, like it was too lofty of a goal, B happens somewhat, C happens a little bit. But then you end up with this like D, E and F kind of outcomes too that like we weren't able to anticipate. And so the ability to like have a whole model to like run these scenarios out and uh, we can see what these probable outcomes are within timetable, like timescales that are reasonable. It's like, I'm in office for four years. What are the things that I can make happen in this four year timeframe? What are the things that I can put in motion that might take 20 years to improve? And how do we avoid me making some policy decision today, messing up something that is going to like be somebody else's mess 20 years from now? And that type of thinking, I think unlocking that for some policymakers has been really powerful.

Chris Fanchi: Yeah, it'd be nice if there was more of that, honestly.

John Cordier: Yeah. I mean another example during, and this is very much health oriented in 2022 when in a number of states Roe v. Wade was looking at getting overturned. One of the simulation use cases that we had is what happens if you restrict birth control and abortions in a given state? There are people who are still likely going to go seek those two things out. Where are they going to go? What does that do to demand for those things in other locations? Are there going to be the appropriate amount of providers? What's like the safety issues that are going to come up? And so we're able to run all these different simulations and we were able to say, okay, state of Indiana, let's game plan this out. If Illinois goes, you know, says yes and um, Kentucky says no and Missouri says yes and Iowa says no, like what does that do for like our region? And they're able to like kind of turn on, turn off different policies. So they're able to see like uh, what happens if our neighbors do this, what happens if we do the opposite? And in some ways it created a more empathetic way of doing policymaking because you're able to see if we make this decision, it has a cascading impact on the people around us, and I think the power of this digital twin of the US Population and the other populations that we're building, it enables policymakers to see, yes, you have an impact in the population of people that voted you in or that you're representing. However, those actions also impact things beyond just your political boundary or geographic border. Um, and so that way of making more empathetic decisions as a policymaker, I think was a. Like, that's. It's a great use case. I would love to talk more about that some other time, but, um, I think that creates a template for how other policy decisions could be made in the future, which we hope to enable.

Chris Fanchi: Yeah, that's great. That's great. So, going back to talk a little bit more about the business itself, longer term, where do you see epistemics going and what are your ultimate goals for the company?

John Cordier: Um, ideally we get mentioned in the same set of players as Snowflake, Databricks, Palantir and others. Like, we become a household brand that data science teams are using all over the place. Um, a lot of the people that we kind of, uh, can work with or compete against are a lot of the major consulting firms that they might throw a whole bunch of really smart people at trying to understand these processes and systems, ride recommendations back. We can enable them to do that work, but we also can enable them to do it quicker by using this type of technology. So over time, I view our health vertical as like a really big and important one. There'll be a consumer vertical. Um, in Malcolm Gladwell's book the Tipping Point, he talks about these emergent phenomena and how do you understand if it's going to be a fad or a true trend? How do you understand the population? Like all of those examples are terrific for our software to solve. So health vertical, the consumer vertical, insurance, government, and there'll likely be others as well. So, um, I mentioned we've been at it for six years, um, kind of hit an inflection point earlier this year in the pharma markets. And I think over the next six years we'll see this type of technology get adopted in others. So longer term, uh, if we go down the path of becoming big enough and we have a solid enough revenue base and customer base to go public, I'm sure our investors would love to see something like that. There's also a scenario where we stay private and are able to make money through those different verticals, but also provide our software to NGOs, nonprofits, and possibly even state and local governments at no cost, but still enable them to have a very powerful tool to run through these. So I'd love to see something like that in the future. I do believe that agent based simulation as a more modern form of data science is going to become commonplace for all these different use cases going forward. It'll change demand planning, it'll change actuarial sciences for insurance and healthcare. Definitely the way that economics is done. One of our collaborators, Doane Farmer, his team at Macrocosm, they're applying agent based simulation to understand energy markets and economies. Like there's a whole bunch of use cases that people haven't been able to think through the downstream implications of before. Agent based modeling enables you to see that ahead of time. I think if you're any, if you're like a Fortune 2000 company, your business is impacted by like changes elsewhere. Like you gotta be using tools like this to understand risk.

Chris Fanchi: Very interesting. So in terms of short term, what, what would winning 12 months from now look like?

John Cordier: Within the next 12 months we'll close another round of funding and we're teetering around profitability. So there's the question of do we raise more money or do we kind of just keep things going with our ability to close these bigger enterprise deals and just grow the team more organically. So those are the questions that we're going to be grappling with over the next 12 months. Like having an impact in more of the pharma companies, having an impact in more like of the public health use cases that we have. Like, thrilled to have that happen. Also really excited about opening up the consumer good vertical and seeing more adoption within the insurance companies that we're already working with. So I'd say those are the kind of the three areas that I'm most excited about in the next 12 months. Um, and with all of those things coming to fruition, it sets us up to have a lot more government work and gives us the ability to invest in making our product easier and easier to use. We're not going to get here in the next 12 months. But like one of my favorite classes in high school was AP Human Geography. And this type of tool of like, you have the population, you're trying to understand, like different public policies and different technology changes, how it impacted populations going forward. So what that whole course is about trying to understand like the development of cities over time, the rise and fall of different empires, a whole bunch of different stuff. Uh, I'm thrilled to one day be able to like provide our software for that course that enables AP Human Geography to combine with computer science, combined with math, combine with other data science, things are going to be popping up and I think that'll ideally inspire a whole number of other people to think about the world differently. Like what could be possible under this set of conditions or an alternate set of conditions and getting to that point, probably not in the next 12 months, but ideally as soon as we can, um, I think that's going to be really, really impactful for people.

Chris Fanchi: Okay, so given that these enterprise deals typically take time, what is the biggest bottleneck that's, that's slowing down your adoption today? Is it the sales cycle? Is it compute costs?

John Cordier: What's the biggest struggle? Honestly, um, we kind of view what we're doing as creating a new category. So I think there's education of utility. Um, we educate most of our customers through what people we've worked with and people who are continuing to work with are still using the software for. So, um, I'd say this early stage in like creating this sort of like new category. Um, yeah, education. And then I'm probably the bottleneck for most things in the company at this point still. So, uh, I'm so far enough to be able to admit that at least.

Chris Fanchi: Yeah, you need more than just a digital twin, you need an actual twin that can be thrown that off. Right.

John Cordier: Maybe that's one of those, uh, robots that can take on some of the work, who knows? But no, we have a great team and um, I'm glad that we're growing as a team. We're building a lot more process in place so that things are repeatable. Yeah, I think the education of what this technology can do and you could liken it to 10 years ago, like if you dropped LLMs, people are like, what the hell are you talking about? Like, I can't use this. Like, this isn't really useful for me. But until you make it so accessible that like people find their own use cases for it, it's like, yeah, you're going to be fighting bigger enterprise. Longer sales contracts like Populous, we've been able to get data scientists adopting that product. Polaris. This is like the make agent based simulation, as easy as working with LLMs kind of initiative. So I think, uh, over the next year we'll see how well our team and, or like how far along we've gotten in making this type of technology more and more accessible for people. But education and accessibility, two biggest challenges.

Chris Fanchi: Well, John, this has been really fascinating. Thanks so much for coming on the show. For our listeners that want to explore more about Epistemics connect with you directly. Where should they go next? Where can they find you?

John Cordier: Hit me up on LinkedIn. Um, um, you can email me john cordierpistemics.com. um, I'm pretty diligent on following up with folks, so, yeah, reach out to me, email LinkedIn. Um, you can also go to the epistemics.com website. If you're a data scientist and want to explore this synthetic population or ability to build your own digital tournament of the population, populous is the product you'd want to start with. And as if you're a decision maker on the call and you're trying to understand trade off between different strategies or different policy implications, Polaris is, uh, the tool for you. So, um, yeah, hopefully there's a few people to reach out.

Chris Fanchi: Fantastic. Well, to everyone listening, thank you so much. Please be sure to subscribe, leave us a review, and as always, keep pitching, keep building, and keep scaling.

Chris Fanchi: Thanks for listening to this episode of Pitch Build Scale. Uh, be sure to subscribe on your favorite podcast provider so you never miss an episode and we'll see you next time.