Synthetic Populations and the Future of Decision Intelligence

Essentially, it's a swarm of models, AI models that emulate human cognition and emotion and become highly predictive of behavior across populations. So you're creating synthetic populations of people that are then situated in context. Forget Personas, Jill Axline is building synthetic populations that predict real human behavior and that changes everything. Keep watching to learn how. Foreign.

Hello, and welcome to Data Driven, the podcast. We explore the exploding world of artificial intelligence, data science, and of course, none of this would be possible without the underlying data engineering. And with me on this road trip down the information superhighway of the future and Buzzwords is my most favorite data engineer in the world. How's it going, Andy? Hey, Frank. It's going pretty good. How are you? I'm

doing all right. I'm still wearing the hipster glasses because they were recording this about post 3 weeks since my concussion. And as we were telling our guest in the virtual green room that we kind of owe the show's name to a concussion. So true, folks who, longtime listeners, know the lore, so we won't bore them or waste any of our guests precious

time. With us, we have Jill axlein, who is Ph.D. and is the co founder and CEO of Mavera, which is an interesting company and Maverick Era is what I'm told it's short for. So welcome to the show, Jill. Hey, thanks. So happy to be here. Yeah. So you also have three kids and. I have three kids. Andy has three. Three plus two. Yes, that's. I think, I think there's a correlation between number of kids and gray hairs. I know I have kids and five grandchildren, so there you go. But

I'm old. I'm just saying you have an age today, you know. So what does Mavera do and what is brand and business meaning for? What does that mean in high growth. Companies, brand and business. I totally botched that. I'm sorry. I'll blame the concussion because I can do that for another week or so. So what exactly does Mavera do? Sure. So essentially it's a swarm of models, AI models that emulate human cognition and emotion and become highly predictive of behavior

across populations. So, so you're creating synthetic populations of people that are then situated in context. So as opposed to a model that's trained six months ago and then is rapidly trying to iterate, it actually pulls its synthetic database will update on a second to second basis. So you always look at your population in situ. Additionally, I would say it provides a really strong pulse of what that population looks like within the context of your business or your vertical.

Because we support a foundation with deep business context that takes into account not just your business from the time that it was instantiated, but it also is updating

temporally and it creates relational, like relational connections across your business. So for instance, if there's a marketing spend five years ago or about the same time that you launch your flagship product or a secondary product, it's going to show a lot of data on how the context around that might have influenced your outcomes.

So I guess like long and short of it is you have populations situated in context and wrapped around your business, and you can use that pretty expeditiously to make decisions in a much less expensive way than most market research or, you know, strategy research, strategy based research. It's almost like you're taking kind of like the SIMS approach of having these individual entities, I wouldn't call them

agents because it doesn't sound like they're agents. It sounds like they're simulated entities, like you said. Right, exactly. That's interesting. Is there like a. That. That's an interesting approach because that does, it probably doesn't completely insulate you from model drift, but it probably does a good job of, well, we're having a massive windstorm and it's like, you know, negative, whatever. Outside in your

Chicago, it's really cold. It's always sunny and it's always sunny in farmville, as I like to tell Andy. But, but I mean, you can insulate against a certain amount of cold, but you can't really stop it. That's right to think about it. So you can't really stop model drift, but you probably can prolong how, how, how long your models are valid for this by this approach. So that's correct. In addition to that, something that I've pushed on because I've been an

advisor with this team for well over a year. And since I'm a ph dork and I, you know, I'm always looking at evidence. Evidence Ev. I was the original skeptic to synthetic populations. In my last role at Morningstar, I built our market research team. And when I was first introduced to the idea of synthetic populations, I was like, you know, tons of skepticism. I think the big thing here is they've built in a level of AI governance around things like drift, but also to

model the difference between evidence and inference. And so they're looking for confidence scores. They'll gather first party data around your population and then create a synthetic data layer on top of that. And a good example would say asset managers like ice cream. Asset managers like cold things. They like cold, wet things, they like cold, wet, sweet things. And then a

coefficient is that assigned to each of those new synthetic data points. And so while it makes a more robust data set in the billions that allows it to draw inference, it's also accounting for again, what, what's based on evidence and what's based, what is inference of the machine. And then there's also a governor across this swarm of models. So it's going to call on the right model for the right facet of human thinking or feeling that it's trying to construct. And so

I think in doing that it creates safeguards around confidence. So we, we produce confidence scores, it will give a spread of opinion across a population. So unlike a custom GBT or a Persona and some pre existing platforms that are emulating language, it's actually taking a look at where's their entropy across emotional response and cognitive response in this data set and what does that look like in the spread of opinion for that audience. So it'll tell you the nature of the spread

and where that spread is happening. So now you can account for almost,

you know, sub segmentation within the population. And that might look very different at the top of the funnel when we're looking at thought leadership topics versus the bottom of the funnel in marketing where we're thinking of features, functions, benefits, et cetera. And so that allows at least marketers, but I think others, anyone go to market to really think about what is their message for the right audience at the right time based on, you know, where they are in their

buyer's journey. And so that to me is a little bit different because I would say the last facet of this is the response stability. We're also providing a level of test retest reliability. If you go into ChatGPT recently, someone was flaming me because I've never made caramelized onions. And so, you know, as a joke, he kind of went in and said how many people who are 40 something, you know, like know how to make

caramelized onions? And these percentages swung quite significantly from the first time he queried to the second time to the third time. Whereas we're looking at population response stability and modeling that, projecting it into the future and looking at the trend line from the past on how this population would continuously answer the question. So I kind of guess like when we think about model

drift, I think that's likely inevitable. But if you're situating and updating with minute to minute context and then you're surfacing some of these governance factors around what the Outputs are, we're getting to a closer place where we can actually be collaborate collaborators with the AI and govern it and then build, you know, a greater level of trust is the hope. That's interesting. I'm glad you addressed the skepticism because that was going to be my

next question. Like, how do you know this is real? How do you know that it's accurate? The other question I had, and sorry, Andy, I had a bunch of monster energy drinks today. You could probably run different simulations, like in parallel, right. Assuming you had the compute. So you can see if this happens, if that happens, right. If there's a recession, people are going to do this, go this way. If there's a boom, if it kind of meanders somewhere in the middle, you could probably run

only limited to what compute you have, right? I mean, yeah, I mean, it's a credit based system. So, you know, you buy credits like a tank of gas and it's going to, you know, give you enough gas to, to build whatever it is you want within limits. But I would say, yeah, I don't think you're really, yeah, I don't think you're really restricted in terms of what outputs look like on, on a scenario

analysis. I think obviously if the more data we have, let's call it for a specific company, when I was working at Morningstar, that's 40 plus years of data on the back end in that deep business context, that makes that prediction that much easier. And so I think it also depends on what's coming into the model and what its power is and its ability to be predictive. I

guess I should say that's cool. Because I think this is an interesting, it seems like it's an interesting mix of kind of predictive modeling and LLMs. Right. Because predictive models, I mean, they're not new. Right, but they're not. But they do. I think they're, they're traditionally, they're

very susceptible to drift. Right. But I think also by simulating the individual actors, because a society and economy, a customer base is, consists of, you know, X number of, you know, not sovereign but unique individuals that are going to have certain personality traits. And some of those you kind of can guess from. Like you said, you know, asset managers. Asset managers, everybody likes ice cream, but asset managers probably really like luxury cars. I'm going to go out on a limb. Right,

right. And probably how much the, how many luxury cars they have and which model of luxury car they have is probably going to determine, is probably not, not determine how successful they are. But it's probably a correlation between how successful they are versus like how not. You know, I don't know. I. If you're an asset manager and you're driving around the Hyundai, there's gotta be a good story behind that. That's right.

I agree with you. And I think again, when you can ask the synthetic audience and pull them, you can start to get into further nuance whether those are B2B dimensions of, you know, like firm type, role type, etc. AUM or it can get into that more psychographic or it can get into start, start to break down archetypal differences and you know, all of those then can be mapped into attributes that are built into the channels where we communicate with them.

Go ahead, Andy. I don't want to hog the mic. No, no, it's all good. I'm fascinated and kind of playing off your, your idea of model drift, Frank, and your questions along those lines. I mean, in one sense I would say, you know, a synthetic audience or you know, a synthetic sample or cohort, however you want to classify that. Is model drift happening in that context is probably not unheard of because

there's cultural drift. And if you're looking for ways to effectively simulate that and run marketing campaigns against, you know, the synthetic cohort, it doesn't strike me as out of the realm of possibilities that you may want some of that you may want to even tune for, especially if you're looking at a younger audience. There are emerging trends that come out of those demographics. It's just part of the nature of those demographics. I mean, I'd love to hear your thoughts on. On that.

Yeah, I mean, I don't know that it's a function of. I don't want to make it like a generational distinction, but I do think that anything that's current to context is going to impact on a minute to minute basis in some cases how the population is going to make decisions and what level of like

arousal they have. And I don't mean that in the, you know, cheeky sort of way, but I would say like we're working with an index team in financial services and they asked me on the spot, can you please model a high net worth investor in Denmark? You know, and this was last week just to, just to say, are you thinking about, you know, rebalancing out of blah, blah, blah, US broad index? And you know, the answer was not immediately, but here's my thinking on that

and here's what I would be investing in instead. So now they're trying to think through what's their messaging around outflows in that predominant US broad index? And then how are we surfacing the rest of our family of indexes in its stead? And then he asked, how does this, does the audience, is there a large spread here? And if so,

you know, what is the nature of that? So now we can think about discrete campaigns across this population, which is pretty narrow of, you know, ultra high net worth investors in Denmark. Right. So I think it's

applicable depending on what, what is that trigger, you know, that what is that zero moment of truth for any given population that is going to be influenced by their immediate context. And you know, with that responsibility score, we can then tell them this is something we think will persist over time versus this is ephemeral. And based on what's

happening in the news around tariffs today. So here's something to push out in your channels today versus here's something to build into, you know, a long tail campaign and how to think about product strategy in a different sort of way. That, that's pretty fascinating. So pivoting just a little bit, you, you mentioned quite a few instances of incorporating evidence into this. And I would imagine that I'm an engineer. Okay, that's a warning. So, so is our cto. I'm getting used to it.

I think about open and close loops all the time. It's just, you know, I

don't even have to think about thinking about it. It just happens. But being able to, to become predictive and have that feedback where you, you made some, you know, you made some prediction, some predictive analytic, and then you collect evidence on how accurate you were and not just, you know, percentage wise, it doesn't really apply that much, especially in marketing type and especially in the age of AI where you can collect information and feed it back into the system as training data,

effectively as responses to prompts. So the prompts themselves become part of the data that goes in and then the outcome that was predicted, that's very easy to see. That part happens. But then supplying the evidence you predicted this, the delta between the

predicted and the actual, that's evidence. And so being able to quantify that, being able to feed that back into the engine, I think in early 2026, as we're talking about this, we've not had the ability to, I'd say in, you know, in, in natural language, to provide that sort of information with any sort of confidence that the algorithm that we're supplying that information to, that feedback, closing the loop on the evidence, supplying the evidence, we just hadn't had the confidence that the

machine was going to understand what we meant. And one of the things that sort of slipped into invisibility over the past, gosh, what's it been, three years and a few months since Chat GPT was released? Is that the model mostly understands what you're

saying now. And I mean by, by mostly some number well above 90%, you know, it's going to get what you mean and when it hallucinates, you know, it's going to be because it misunderstands what you said, not because it went off, you know, and started interpolating what you said and made something completely different out of it. It's the way it was stated, wasn't quite clear. And nowadays I hang out mostly in Claude and Claude code. So when I'm going back and forth with, you know, with the engine,

it's, especially in Claude code, it very often will pause the conversation and stop and say, hey, I have this question, you know, and here's the options. I think you're, you know, based on what you said, I give you 1, 2, 3. And then number four is you just type and tell me if I completely missed it. And I rarely find myself on that bottom option. Most of the time I'm picking the, the

top option, which the one it thinks is most likely. And so having having that sort of evidence based feedback, number one, be so much easier than it is before. And so I can see that limiting model drift. I can also see it kind of making your predictions align with the timescale that you mentioned. So not just the population being so, so small, which is infinitely harder than dealing with big data, right? Dealing with a small set of data. How do you predict in all of that? And before I

ramble anymore, I'll just stop and let you respond. How about that? Well, it's interesting and I don't want to get over my skis because this is really where our CTO shines. He has the ability to create some audiences out of what he would say he would call dark matter. The best way for me to think that through is when I look at a tree and I see its various branches. I'm looking at the tree to define the tree, but there's so much more sky and negative space around that tree that also defines it.

And so he's starting to look at data and how it affects other data and then putting that in context and using that kind of negative space to then define the audience that's so small. So that is, you know, in the case of when I was at Morningstar, Acid owners, really small group of people, but one that Morningstar really wanted to understand a lot better. And so that institutional audience, they're regulated. It's hard to, you know, get permissions because they're so small.

Their time is worth a lot. So it's an expensive panel to construct. And here he was able to build from again, like that negative space to then recreate the audience. And, and he is surfacing that confidence variable. And if there is a hallucination, hallucination risk, it's tagged and it will prompt you for what sort of data it then needs. Or it's going to say, actually have to refractor the audience a little bit differently. There's too much entropy for me to continue and

it will go and run it again. So. And again, I don't want to get over my skis because I'm the social scientist in the mix, but that's how it's been described to me that I can, I can best understand it. That makes a lot of sense actually. And like you can kind of, I think there's a lot of inference here in terms of what you can infer. Right. So my, my kid, my middle kids, my two younger kids are really into and really the three

year old just likes hanging out with his big brother. They watch Dragon Ball Z, they watch the Jujutsu Kaizen, like all the crazy anime that's very popular now. I bet one of the things you could do, I, I've actually gotten into it. I was never much of an anime fan, but like, you'd say, like say Dragon's Ball Z. Right. Dragon Ball Z has been around

that I'm aware of, maybe 20, 30 years. Right. But. So you can probably, you could probably make a solid assumption that there might be some Gen X folks that are Dragon Ball Z fans, probably a lot of millennials, a lot of Gen Z, Gen Alpha, whatever they're calling them now. But there's probably not a lot of people in retirement homes, boomers and up there are big fans of it. Is it because they wouldn't like it?

I don't know. Maybe. But it's just, it tends that since that demographic skew is kind of small, you're probably not going to find a lot of them that are going to be into that in the retirement. I don't know that that's just me just firing an analogy. I mean, my parents liked K Pop Demon Hunter when my kids made them watch it, but I have girls, so I don't know,

they're just really cute though. That's really cute. It's a very well done kind of cross genres, but yeah, yeah. And K pop is very, very, very addictive. Yeah. You know, so like it just sticks in your head. I don't know how they did it, but who, who are the industries? What are the industries that are really interested in this? You obvious, you mentioned Morningstar, obviously, I would imagine financial services. And Morningstar is asset management. Right. Is that what it is? Or a hedge

fund or it's, I'm. Not exactly sure, data and research. So I mean, I think primarily they're known for their research and data and how they've rated funds over the years and they've expanded from there by way of acquisition. So PitchBook is a part of it. DVRS is an index business. So they, they have seven different pianos that really like traverse

financial services. At this point I think financial services has been interested partially because I'm in financial services and I'm literate and being able to discuss it and showcase its benefits. Right, right. I would say this is more like functionally, like accurate for any

place that needs human intelligence. Right. So I've worked with private equity teams that are helping to arm their portfolio companies with a marketing tool that doesn't have them, then looking to boutique agencies to do this level of market research and understand their ICP and find product market fit or message market fit. So there for them, it's very easy to kind of get in there, even the technical founders, and try to augment maybe a gap in

their marketing acumen. I would say marketing agencies, creative performance, et cetera, they have taken to it really easily because they're already wizards who wield, you know, traditional wands on doing this kind of work to understand a market, to understand the message that's going to fit with that market and then to make sense of what the real results were when the market either engaged or didn't. Right. So and

building the creative around that. So the ability to pre test all of that with the audience gets them to the starting line before they put money behind it or have their client put money behind it with the best possible set of options. So I think agency has been pretty prolific there too. And then the last. And again, I'm kind of biased because I came out of enterprise.

Enterprise marketers who are finding gaps in the kind of the traditional products that are, have easy distribution within the enterprise are looking to a tool like Movera to try to get more what decision intelligence that's human based in what they're doing and so that's, that's where we're seeing a good amount of traction would be like in that mid market and enterprise level marketing team, whether that be product marketing or demand gen or market

intelligence. And I came out of brand strategy so I found great utility for it there in corporate comms. So again I think it's really that go to market team where human intelligence becomes so important to decisions and current like traditional research methods are really slow and they're quite expensive and not everyone can do them, you know, or they think to just grab the information from within the four walls of the firm and

anecdotes of talking to customers. Right. So this is a good way to augment an expensive way to augment some of that decision support. So you can like throw together like a, what's the, like a test market simulations and you can probably, there's probably knobs and dials you could do. So you can kind of like get

multiple answers and I, I get it. So you can kind of, you can hit your, whatever your campaign is going to be with the running start as opposed to it's a little bit more guided than just throwing stuff at the. Wall and seeing what sticks. That's right. You know what to throw. You have better idea what to throw and where to throw it. That's right. And I mean we had, even when I was still at Morningstar, pre

tested like the first time ever they built commercials. You know, they didn't, they don't really do brand level, you know, television commercial. They were deploying in Chicago, New York and London. And it was shown that in London it wasn't, whatever it was, the voiceover, the creative itself wasn't going to resonate with that audience as well. And so that gave us the foresight to take a look at what the voiceover is, what channels we might use, how much money we would put behind it before

we deployed in that market. And so that, that kind of helped with channel strategy, it helped with content strategy. It certainly helped to evaluate that creative before any money changed hands. And so I think that was a super helpful thing. And now it's an award winning campaign. I'd love to feel like Movera had something to do with it along with all the brilliant minds that worked on it. That's cool. So you can get down to the macro, not macro, micro level of

like the voiceover may not work in this market and things like that. That's cool. Yeah. In fact there's a. So we're in multiple modalities.

We had used, I helped to co author the CEO's speeches for multiple years. And so we made him pract again and again and again, and we would. We would record them. And so the video analysis tool would look at the substance of what he was saying, the creative that was behind him on the deck, and then also his performance. So as it evaluated him, it said, you know, you're not taking time to pause for emotional

resonance. And it gave all the timestamps across his speech where he should pause and why, and potentially even for how long. So it was looking at audience engagement and emotional connection. Then it started to take a look at, well, your message isn't that highly differentiated. And because we have this deep business context, we know that X, Y and

Z are also talking about the convergence of public and private markets. This is what they're saying, here's what you should say so that it sounds uniquely Morningstar. So it now is helping to differentiate the message. And then when we got down to the creative, it's saying, you should do things that are a little bit more dynamic. You should back up what you're saying here with, you know, more data, graphs, charts, et cetera, less imagery. And so it was giving us guidance on three

dimensions of that speech. And as we did it over time and recorded him, we saw his scores go up and up and up. And then it ended up being a really successful speech at the flagship conference that spring. So, you know, I had even said to him, like, maybe we should use this before earnings calls. You know, you never know. I could see the. I could see the applications and, you know, in fintech, I could also see applications of this in political campaigns.

Yes. I was just thinking that. I'm like, you know, yeah, they would eat this up. Yeah. So we have been in some conversations, and I obviously can't talk about it with someone in the House of Representatives because we also have a news digest that will metabolize the news and give you the perspective of specific audiences. So he wanted to look at the two counties, you

know, that he. That are part of his constituency. But then he was also looking at the committees, you know, so he's on two different committees and how are they responding to the news and what is it that they're doing? So it was doing this kind of social listening and moderate, you know, modeling of the audience. And then he could say, well, this is what my response would be to it and get them to vet it before he, you know, would push send on a communication. So, yeah, that was. That was

something that. It's so timely Particularly with that news digest. Yeah, sure. And you know, particularly in it's, you know, the sentiment analysis angle on that's huge. And being able to do that in near real time, I think has, you know, applications across not just those two markets, but a bunch of different verticals as well. Because you almost. The perception is if you don't respond or react, that's a response or reaction, you know, so. Yeah, that's right. So I, I'd say between access

to news content and then also connection with APIs. So we have Bloomberg flowing through the platform Pitchbook. We've got it for marketers, Ahrefs and Semrush data. If you're looking at SEO and you have thoughts towards what does it mean to show up in answer engines, all of this data flows and could be called through the platform so that you're looking at real data again, we leave a receipt of like this is where we drew this data from. You can see it. And here's where we

inferred. So now you can use your own best thought and strategic thinking on. Okay, do I need to get that

inference score down or do I feel good about this and I can build it into my argument in a really defensible way? So just curious. That's cool. Yeah, I'm, I'm down with it. I'm just curious how, in your experience, how have the, how's the opportunities presented themselves for someone to kind of step out and be creative is probably a nice way to say it. Or, and, or controversial. You know,

there's, there's value in that some of the time. I mean, from a. If you're talking about marketing a product or service, you definitely want the differentiation. You mentioned that earlier. If you're talking about a campaign, whether it's a marketing campaign or a political issues type campaign, the opportunity to either be portrayed as a maverick or see what I did there or to, or to be, you know, just portrayed as somebody kind of breaking the mold, stepping outside

the talking points. You know, how's, you know, how's your, how's your product and service addressing that. But also too, there might be some. I'm sorry, I didn't mean to cut you off. No, that's trying to cut off Andy. And then I cut you off by mistake. But also to the inverse of that. Like maybe there's some things you people, you don't want Mavericks, you don't. We want stability. Financial services kind of comes to mind.

So sorry, I'll shut up. Yeah. So I mean, you can construct your own Brand identity that's going to say, you know, typically, here's our brand standards and here's our brand expression, which can come across creatively or tone or what have you. So that can be constructed and put on the back end so that everything is then scored against that and can tell you how far away from that you're drifting. Then you can put it in front of the audience.

Typically, anyone who's working with is going to have their own framework for understanding. You know, how do I evaluate whether this message, message can go to market under my brand and how much risk am I willing to take? You can ask it even to assess the risk given the audience response.

And as it splits that audience where people are having a difference of opinion, you can isolate that and say, is this my most likely buyer or is this the part of the audience that maybe there's a huge population that would like this more provocative message, but it's a, it's an audience, as it's described, that would churn. So, like, it allows you to make a little bit like, more strategic business

decisions based on like, what. What are the attributes of that audience that are going to resonate with that more provocative message. The other thing I would say is just, oh, no, it's okay. This is built on a gan. So it's an adversarial network. And I would say, as opposed to being sycophantic, like so many models that are like, oh, yeah, I agree with you. And then you're like, no, don't agree with me. Be like adversarial. You know, push back. It's built

to push back. In fact, we have a Persona specifically meant to poke holes and ask you questions and get you to question your assumptions. And I always start there. It's called Osprey. And I, like, that's my number one first stop on the bus is here's how I'm thinking about this competitive analysis. Let's like sort through what.

What is wrong with that or how I can improve it. Same thing with a market sizing exercise. It feels like that should be wrote, but as you lend more specificity to it, I might be market sizing against not just a product, but a specific use case that I want to build up, campaign around. And now it becomes like a much more nuanced way of modeling an audience. So I always, again, start with that adversarial model to get me to think better, you know, like, really improve

my strategic critical thinking. Kind of like the 10th man in world War Zone. Okay, I don't know what that is, but should I watch it? I'm sorry, Andy. Andy, I cut you off. Yes, it's an interesting concept. I don't want to spoil it for you, but, like. And it's based on a real, real army unit where they basically become their contrarian. If nine people agree on something, then it's. They randomly will. If 9 out of 10 people agree on something or something like that, or 10

out of 10, they will randomly pick one to. You have to poke holes in it. Oh, sorry. Encountered. That's okay. I first encountered that in World War Z. So. Yeah, that. That was where I saw that. The. It sounds what I was thinking as you were describing that. I guess the phrase that popped into my. My mind was, you know, there's no such thing as bad publicity.

And if you are peaking interest, whether it's positive or negative interest, if you're provoking some sort of reaction in that, and I think a lot of the social media algorithms are tuned around being able to do that very thing, you know, to. To get a reaction, either an agreement or a disagreement, then that can lead to engagement. And if that's the goal, that makes perfect sense. That's right. I. In fact, I have a book right here called Filter World.

I think that's what it's called. Yeah, Filter World. And it's really all about how algorithms can. Can do that, feed you back things that are more sensationalized, kind of like yellow journalism going back to Hunter S. Thompson. Right. That are meant to create some sort of response, whether good, bad, or ugly. So, yeah, I think that's right. But at least you could test. Yeah, at least you can test some assumptions first prior to taking it to market and getting slammed for it and

having unintended consequence, potentially. Yeah, Well, I mean, if you think about it, I'm just basing this on my experience, because I have the most experience with my experience. I love a comeback. Right. I just. I love it. And often the way that that comeback begins, the. The arc starts with me first noticing something and having a negative reaction to it.

And then as I get more information, I go, well, yeah, I could kind of see where they're coming from and, you know, begin to identify with it and empathize and. And then every now and then it's rare, but when it happens, it happens huge. And I think part of the reason is because I started so negative with it, my support skyrockets, you know, a little. It's not a line, it's an exponent, you know, very exponential curve of

support that Grows out of that. And like I said, I think it's powered by stretching that rubber band in the opposite direction to start with. Yep. Although I would say some people are built that way because my dissertation looked at processes of empathy and processes of perspective taking and how counter. Counterargumentation happens. Right. What are the various factors, either in an environment or in a message that are going to create that? But there are also some things just in

you that might have that approach to say. I would say my 7 year old, my little guy has like, he comes from a space of no. We always start with no. He's also like in the 99th percentile for math. I think he has like an engineering mind. Like, I just, I was gonna say. He sounds like an engineer before you even

mention math. Yeah, yeah, yeah. Likes to take things apart and put it back together. So that's it. No is a good spot. Yeah. Yes. That's funny. It reminds me of. Here's a story from way back when. Once upon a time, I worked for a fintech startup. We'd call it. It wasn't called fintech then, but it was basically in early 2000s. And it was a banking portal, but it was meant to be kind of banking for people who weren't comfortable with finance. Right. But the,

the rationale was they wanted to make the site really friendly. And one of the things they did was they put little cute cartoon characters on every page, which people. And this was in Germany. So like it was a, you know, banking culture in the US is very conservative. Even Germany is even more so. And that's being kind of. Turns out people didn't want to put their money into a website. Which again, early 2000s. Right. Still, that was already a stretch

with cute little cartoon characters. They wanted serious, they wanted stable, they wanted boring, they wanted, they wanted the suits, they wanted that. And it was kind of like when I saw the website, the design rolled out, I was like, I don't think this is gonna work. I better have my plane ticket home just in case. And you know, it turns out I was right. You know, trust me, I, you know, I didn't want to be right because I would have, you know, had dot com dreams and, you know, all that. But.

But I mean, you're right. Like sometimes it would have been helpful if they were to test out, if they had the capacity to test out. Hey, what if we went for a cutesy K pop kind of demon hunter thing for a banking portal. It might fly today maybe, but probably not. Just depends on the audience. Again, yes, Exactly. Know your audience. Right. That seems like a tough sell. It, you know, in Germany in the late 90s, early 2000s. I don't know. Right. It definitely was.

I think after half a billion euros were spent, I think they acquired 120 new customers. So, yeah, it was br. It was bad right there. It was bad. And I might be rounding up ratio right there. I can do that. Yeah. So, I mean, again, I think audience, you can't really replace, like human response to something. You have to get something out into market and see if trust is established and people engage

and ultimately make a decision to purchase. But I think getting to the starting line with the best set of options, with defensible reasons behind why he went with those options, is kind of a better spot than we were a year ago or two years

ago. Right. And so I think, I mean, we can only go up from here, but I think, you know, I'm, I'm optimistic that if people were to start integrating this, it doesn't have to take them out of the job force. It just can help them do their job a lot better, you know. No, absolutely. Yeah. How did you get into this? How did you get into this? Because your background is in. Your PhD is in communications. You're getting used to dealing with engineers. Yes.

How did you. How did you end up at a company that is largely driven by engineers? That seems. Yeah, this is a great question. So again, I was kind of that skeptic who was running a market research team and always being pressed on my budget. So the budget was, you know, in the high six figures. And it's like that's the

first place everyone wants to cut when everyone's looking at margins. But it's also such an important place to make sure that product strategy, message strategy, all these things are kind of coming together in the right sort of way instead of wasting money downstream. And so I was trying to, you know, A, look for a way to cut cost, but B, I also really wanted to understand what was coming with this whole, like, generative AI thing, you

know. So when I heard about let's scan LinkedIn, LinkedIn profiles and create synthetic Personas, I really started to pound the pavement to try to understand who's approaching this in the right sort of way aligned to how I think about modeling human populations, which is what I was studying. So when the strategist I was working with kind of heard me thinking out loud about it, he introduced me to the co founders at Marvera and,

you know, I think I asked some hard questions. They could see that I was nerdy and skeptical and willing to try. And so they gave me access to it for almost a full year. I took it through the compliance process, which was helpful for them, and it was good to see how Morningstar was thinking about this progressively and then just hammered it and, you know, brought it into the C suite and

brought it across the firm in my presentations. And I think through that, it really helped me to understand what the true value of it was. And after seven years at an enterprise, I, you know, I was definitely someone that liked to make decisions quickly, thoughtfully, but quickly. And I was kind of looking for, you know, maybe there's another opportunity to take my expertise and apply it in a different way. So I had a sabbatical. It was like

a, you know, six weeks every four years. Thank you, Morningstar. And during that time, I just spent some time with them to really understand the technology, really understand the go to market motion and look at their capital raise and try to get involved in that process. And then six months later, they asked me to join them. Oh, that's cool. Yeah, that's cool. It was cool. I have to say, I'm drinking from the fire hose because working with the AI engineer, Full Stack

developer and. And looking at operations and looking corporate taxes and all these things. No, that was not really. I carried my. You didn't wake up and you were like, I didn't want to do that. Like, that's interesting. The first thing that comes to mind, and I totally lost my train of thought. So if, Andy, this is an opening for you while I kind of reboot my blue brain blue screen. So give me a second. Oh, now I remember. You're welcome anytime, man. Having

you mentioned regulations, this is what kind of. True. I was very skeptical of synthetic data, just in general, just because, you know, you're basically feeding machines into machines. And I'm old enough to remember when you took it like a tape cassette and you copied it and you did that enough generations, whether it was VCR or audio cassette,

you had an issue. Right? You would get some kind of degradation. However, in reality, I've seen synthetic data do amazing things in the AI space, in the data space, more than it has any right to, basically. So that's why I was not skeptical when you mentioned synthetic crowds, because it's one of those things where it's worked better. But one of the upshots of synthetic data is that the reg, particularly around generating synthetic health data and things like that, you don't quite have the same

regulatory constraints. Right? There is no PII to speak of. And you mentioned that there were regulatory hurdles for, for this. Like what, what were the regulatory hurdles in this case? I'm curious. Well, how, how could you use the outputs? Where would they be applied? If you're reconstructing the brand voice, what are you basing that off of? Is that, you know, is that considered for them proprietary information that would then feed the system for other, you know, competitors or

just writ large? I think that was something that they were looking at. They were of course looking at data privacy. So you know, I was uploading not just our creative, but I was looking at our business strategy across the P Ls and trying to get it to incorporate when things are launching and where is their convergence.

So that if I'm trying to create an umbrella level message at the brand level, it can render really strategically down to the different business units and create continuity and coherence in the message. So but that's, you know, that's their strategy. So they're really worried about like, you know, at what point can we feel like this is safe? And so, you know, in earnest, the team approached the ISO

42001. They had a SoC2, the ISO37000 something because I'm never great at remembering all the numbers but like they really did like get after it in terms of, of ensuring that enterprises specifically would feel, you know, really like safe in this environment and everything. It was abundance of caution. What's that? It was, I'm sorry. Okay, that's fair. Yeah. Because one of the big selling points I've seen is it's not real. Right. So I'm sorry to cut you off

again, but. Yeah, no, because it's not. But it's a synthetic data layer that sits on top of proprietary data and data gathered from like first party sources externally. So I think once you have the mix of multiple things, they just have to ensure that whatever's put in there is proprietary, is protected. That makes a lot of sense. Cool. Yeah. This is the world we live in, you know, that's cool. So any other questions Eddie or. I didn't mean to hog them up. I'm just fascinated by,

is fascinated by the discussion. It's, it's one of those other. Well there's other discussions and topics where we see

the kind of the real world interacting with the artificial and I don't say artificial in any kind of negative way, you know, in the sense of synthetic and, and to me it feels a lot more like art imitating life, you know, and as we, we find more and, and better ways to have technology enrich our ability to do our jobs well. I just. I find it fascinating. So. And it's. It's cool. I can tell

that you found a real. A real fit for your education and your skills and it sounds like your personality and, you know, and kind of likes and that. That's always good, you know, when. When you can do what you are. Yeah, it's so true. I love nerding out every day on this stuff. Plus, like, I'm. I don't. I'm just. I can't naturally sell anything. I have no selling ability. But I can talk about it from the perspective of, like, a practitioner, you know, and a skeptic one at that. So

that's really where I'm coming from in any conversation is like. Like, tell me why you don't buy it. Because I, like, I'm gonna get in your bandwagon and not buy it with you until we can figure out how it actually, like, works and fits into this process, you know? So. That'S a good way to look at it. That's. That. That's really not just selling with empathy, but selling with, like, sympathy, I guess. Right? Like, yeah, yeah, that's cool. That's cool. Where can folks find out more about

you and Mavera? Yeah. I love talking to everyone, so please connect with me on LinkedIn. My name is Jill Axlide. Again, Mavera IO is where you can go and check it out. We liked people to kick the tires, so there's a free trial for everyone for 14 days and you can connect with anyone on our team to walk through how to use it. Cool. Awesome. And we'll let our AI finish the show. Awesome.