Samuel Cohen & Fairgen on AI Digital Twins in Research

In this episode of the Greenbook Podcast, Leonard Murphy sits down with Samuel Cohen, co-founder and CEO of Fairgen, to explore the rapidly evolving world of synthetic data and AI-powered digital twins. Samuel shares how Fairgen has evolved from synthetic sample augmentation into building category-specific digital twins that help brands test ideas, concepts, ads, and products faster and more efficiently.

The conversation dives into the future of market research, the role of AI-native workflows, and why agility is becoming a critical business advantage. Leonard and Samuel also discuss the changing economics of research, the importance of high-quality individual-level data, and how integrations and AI-driven experiences are reshaping insights teams, product development, and marketing functions. This episode is essential listening for insights professionals, researchers, and innovation leaders navigating the next generation of AI-enabled decision-making.

Key Discussion Points:

• How Fairgen evolved from synthetic sample boosting to AI digital twins
• The difference between synthetic personas and individual-level digital twins
• Why category-specific data improves the accuracy of AI-driven research
• How AI-native workflows are transforming research operations and productivity
• The future of market research integrations, automation, and embedded insights tools

Resources & Links:

• Fairgen
• Fairgen Twins

You can reach out to Samuel Cohen on LinkedIn.

Many thanks to Samuel Cohen for being our guest. Thanks also to our production team and our editor at Big Bad Audio.

Transcript

Lenny: Hello everybody. It’s Lenny Murphy. Welcome to another edition of the Greenbook Podcast. Thank you for taking time out of your busy day to spend it with myself and my guest. And today, my guest is Samuel Cohen, co-founder and CEO of Fairgen. Samuel, welcome.

Samuel: Thank you. Thank you, Lenny.

Lenny: It’s good to have you. I know we spoke probably, gosh, almost a year ago, I think—

Samuel: Yeah.

Lenny: —so it’s good to catch up because you’ve been making some waves. So, I—for full transparency, Fairgen is also sponsoring this. We appreciate that. But I want to talk to you anyway because I cite so much of the content that you’ve put out, from a thought leadership standpoint, around the concept of synthetic. A few weeks ago, you did a wonderful kind of taxonomy on, let’s, get some definition on what we mean by synthetic and there’s different aspects of that.

And you have just launched a new product line around digital twins. So, you’ve moved from the kind of augmentation, you know, aspect, the boost component of Fairgen, now into building real digital twins. So, excited to have this conversation because it’s such a big topic, and you have been a pioneer in this. So sorry, and I just stole your thunder because I’m excited to talk. Why don’t you tell the audience about a little bit of your background and then we’ll get into, you know, all this bigger topic that I just described.

Samuel: Sounds great. And yeah, glad to be back. I think we had a great conversation last year, and a lot of things moved across the year. So yeah, on my end, I was a CEO and founder—co-founder of Fairgen. So we, four years ago, basically brought this concept of synthetic to the industry. I have been doing synthetic things for the last ten years, first from the academic side. So, I studied in the UK, did a master’s and PhD and a lot of research on synthetic data generation during these times, and spent some time at Facebook AI research labs, now called MSL, which is basically where all the recent foundational models of Meta are being built. We created a model called Flow Matching Launch. It was the first version before Flow Matching was called Flow Matching that is now used for video generation and various other things. And yeah, and then decided to [port 00:02:33] this into the market research industry, and had a great time. And it was really well welcomed by this industry. And now we have something really new is arising where, basically, we can do a lot more with the new technologies that are coming, and I’m glad to be able to talk about it today.

Lenny: Yeah, well, let’s talk about it. Let’s dive in. Because I think what was fascinating is that when we last spoke, and when you have emerged, it seemed like you had a unique position in the market of thinking about this concept of boost from a sample standpoint, right? Filling in gaps of information so you could, you know, for hard reach populations, you could still have the right n size. And there wasn’t anybody else doing that. Everybody else seemed to be going, “No, we’re going to build fully synthetic,” not digital twins because often that synthetic was just based off of what was readily available online versus defined data sets, digital twins. Now, from my definition—and I want you to give yours—is you’re using real data of real people to not just impute their—from a modeling standpoint, but actually to create a digital avatar that has a high degree of coherence and fidelity to the person based upon all the available data. Is that—

Samuel: That’s a pretty good high-level view of it, but I’ll basically specify it, like, further.

Lenny: Let’s do it.

Samuel: And yeah, I have this taxonomy world that has, like, I would say, three different variables. The one you saw on LinkedIn is a bit more limited than my internal, I would say, like taxonomy. So, you have basically three axes. Axis number one is stakes. When you need to take a decision, what are the stakes of this decision? Axis number two is going to be quality, right? And that’s inherent to the methodology. That’s going to be actually used, right? And three, basically, is a family of model that you’re going to use. And the quality is controlled by other things than the model itself.

So, let’s go deeper into that. When you have super high stakes decisions, like, things where a 0.3% deviation in something actually matters, right, we still need to have mostly human samples that are collected based on the exact questionnaire design that you’re thinking about, right? And that’s what these companies in the market research space have been doing for a long time. And that’s getting disrupted, but like, pretty slowly.

On the older complete end of the spectrum, you have, like, super exploratory research, where the stakes are much lower because you are in exploration mode, like trying to understand where you want to go and there is no basically hard decision where you’re going to invest, like, $50 million and risks millions of dollars, right? That’s pure exploratory. And then you have basically all of this middle spectrum, right, where you have things, like, you know, testing features, testing pricing, testing ad campaigns, and that’s more in the middle, right? What I really think towards right now, is basically bringing good methodologies for this middle stakes decisions, right? So, I’m going to start by talking about the outcome rather than how to do it because I think that drives more value.

So, let’s think about all these major decisions where, right now, like, we take all of these decisions blindly, right? For example, we’re going to launch an ad or a campaign, and we’re going to hope that it takes up. Like, us as founders, like, how many times were we like, you know, I’m going to launch this ad, I’m going to launch this campaign, I’m going to launch this creative, and I just hope it takes off, right? But maybe there’s better tools than our brain and our intuition to take these decisions. And maybe, you know, traditional research is not a good fit for that because it’s too slow and too expensive, so maybe digital twin is a good answer for it. And now we get to the methodology, and there is, like, so many things behind that. So, maybe I can spend some time explaining our approach. So, I think, Lenny, you’re basically talking about, you know, using data to create, like, you know, avatars of people. The way I define it is basically to take data of an individual, right, like the actual data of a single person—I’m not trying to blend the or blend many people.

I’m blending—I’m using the data of an individual. And there’s many types of data on the individual we may have, right? Survey data, so we’re going to get these surveys to this person, transactional data, you can have demographic data, that’s obviously a lot easier to get to, like, transcripts of interviews, like, all of this data about this person. And you can use that basically, to create a model of this specific person that’s not going to later get duplicated into a set of persons. Just a one-to-one model. And then you can expose new concepts, new ideas, new surveys, to this person based on all the information that you have about this. That’s what I call basically level two of digital twins, knowing that level one is basically just, like, you know, this kind of fully synthetic, or, like, average across data. Level two is what I just said. But actually, what we’re shipping is level three.

So, let’s do that, and after maybe I’ll—[laugh] I’ll pass it to you. You’ll probably have going to have some questions. But level three is category-specific, individual-level digital twins. So, now the main positioning that we have is that there is no way to know everything about a person, right, and to know for this specific person, what this person drinks every day, where this person has a bank account, and is this person satisfied with that bank? And does this person have a pet and what are all the foods this person feeds to her pets? But it’s a lot easier to get on this individual basically all this information for a specific category. You run a long survey about pet food and you ask and you get transactional data about what are they buying for their pet, and so on, and that allows you to create category-specific twins. And then for this [unintelligible 00:08:47], you have so much more information about them. And that’s basically what we’re pushing for.

Lenny: Okay. So, I get that and, not that you need it, but I think that I agree. That’s the—well I do love your framework around stakes, quality, and outcome. That’s fantastic. And that makes a lot more sense trying to boil the ocean on all of, you know, the data to focus on a specific category: pet food, auto shopping, whatever the case may be. Are you building those on a client request? So, somebody comes and says, “We want digital twins based upon—here’s the description of our population. Now, go forth and build the data graph—if you will—of the individuals that are car intenders,” or have you pre-selected and said, “Look, we have enough data on these categories that we’re going to go ahead and launch these digital twins based upon the select category business issues?”

Samuel: Yeah, great question. So, we have two approaches, and I think that’s also something that we’re doing a bit differently from the space. So, we have both a syndicated marketplace approach and a custom approach. What we’re doing is basically we’re partnering with a lot of the large panel companies in order to basically get a lot of category-specific data. Some we buy, some we partner on, but basically we are able to get, through these panel companies, survey data and all these other resources of enriched, unenriched data that allow us to basically in our marketplace, when you know, two minutes from signing up, you can already run your surveys on groups of twins of different audiences, right? And basically, our belief is that we need to democratize, like, the quality of outputs from twins, and that the best way to do it is through syndication at the beginning. Now, all of our audience creation and ingestion framework based on individual level data and other sources of data is also open for basically people to create based on their own data, right? But that’s secondary, right now. Right now, the goal is to get people to have—basically within five minutes’—experience of being able to run surveys on high-quality audiences of their choice. So, we have about 100 audiences across three countries: US, France and Germany, across B2B and B2C. A category, maybe mentioning pet food or auto or, you know, cyber-decision makers in B2B, so that’s, like, super wide, and we’re planning to add more every month, basically.

Lenny: Okay. All right, very cool. So, I’m going to try not to mention any of your competitors. I’m just thinking through my question, but there’s one I don’t think this is competitive [unintelligible 00:12:03], Steve Phillips from Zappi recently launched a new platform called CEOFriend. I don’t know if you saw that if you saw that or not, but the—and it’s kind of a similar approach, where it’s like, okay, we’re going to create a digital persona of a CEO, of a CMO, of a CFO, that can answer very specific questions, so around—for advice, right? The goal was to create a digital persona that early entrepreneurs could tap into, to have a kind of a virtual advisory board. So, my point is that’s kind of a no-brainer exercise in thinking, all right, I just, I want to have a more personalized, specialized, digital persona that’s trained, especially around the specific needs. And it sounds like that you’ve taken that and put it on steroids. So, you’ve expanded that to say, you know, sure, we’ve built this. So, there’s a digital Samuel and the digital Lenny, from the perspective of, you know, these specific business issues, and to deploy that from a survey standpoint. So, I talked myself into a circle. Forgive me [laugh]. But why are we in a survey? Why do that? Because, you know, I would think that a conversation, something looks more akin to an IDI or a focus group—why use a digital twin for a survey?

Samuel: That’s a great question, and actually, we have both surveys, quant-qual. Next, we actually have also one-on-one interviews with these twins. We actually also have focus groups, basically, all of the same. So, all of these twins we’ve built can be run through a survey alongside other people from this audience, can be run through a focus group, and can also be run through one-to-one interview where they can also be exposed to concepts, pictures, videos, like, prototypes, like, Sigma prototypes, for example. But basically, all of these things are already there.

Lenny: Okay. All right, thank you. So, that means that Fairgen is now in the business of basically being a virtual, full-service research company, from the standpoint of—

Samuel: Correct.

Lenny: —here’s the business issue, here’s the audience, here’s the tools, and I assume, you know, here’s the outputs as well. That’s a pretty—that’s a pretty ballsy evolution there, Samuel, from where you were. Why did you decide to shift it? Obviously, I get the opportunity. But what was that journey, like, to decide, yeah, you know, we’re just kind of filling in gaps of information and almost a utilitarian perspective into something that’s, like, no, we think that we can actually now replace spend from a primary research standpoint.

Samuel: Yeah, that’s a great question. And by the way, like, one thing I think that’s super important to say is that this approach is also run in partnership with a lot of the market research companies we’ve been working with that are also going to serve this service to their customers. So, it’s not like we’re going at the war with the whole market research industry [laugh].

Lenny: [laugh].

Samuel: I think my current positioning is that basically building end-to-end experiences is necessary in today’s world. And last year’s world was another world. The world completely changed over a year. And so, basically it’s super important to have everything integrated in one place right now, regardless of, basically, you know, the way—of who is going to be using that thing. But basically, you know, cutting a lot of the links is something that really helps basically drive value to people. And people don’t have the same attention span that they had a year ago. So, that basic experience, like, I read a lot about user experience, that’s something that I’m super passionate about, and I’m really trying to optimize for user experience and what customers expect, and what they expect now is end-to-end experiences. Another thing is, we absolutely have to partner up with market research agencies and panels across the world to get data, like, high-quality data, and also around a lot of things, around questionnaire design, so leadership and a lot of these things. But the pure product experience is yes, has to be an end to end, basically.

Lenny: I love that, Samuel. I truly do. So, if you think about the value chain, so you’ve created a new monetization engine for the panel companies without necessarily incurring additional expense with a new study from an incentive standpoint—so you’re allowed to unlock value from this data asset they’ve had sitting there—same thing on the full-service supplier side, allowing them to create new value while giving them a new tool from a pricing standpoint, to compete with the, you know, the DIY platforms would have been kind of the way to think about that, but a fast and inexpensive solution, which I suspect their play, they’re thinking, is, okay, sure, do this with a lower-stake issue; now you’re going to identify where are the gaps of information, and that then triggers primary research to say, “Okay, now we need to go out and actually do real research.”

Samuel: ...

Lenny: So, control in the value chain and the upsell possibility from the clients. You’ve just made yourself invaluable, Samuel. You’ve made Fairgen, the connective tissue that unlocks value and revenue streams at the two primary sides of the industry that are necessary: data and output. Very, very smart. Very smart.

Samuel: Correct. I’ve always tried to bridge basically different sides. And that’s something, like, we’ve always been at the middle of the, you know, panel, like, supply and demand sides. We’ve always been right in the middle since basically the inception of Fairgen, and basically even more now. And the key to the success is to map the areas and use cases where it makes sense to do that and to be able to redirect basically to full collection, like, real data collection. So basically, we’ve spent a lot of time with customers, like, end customers and brand-side clients, to basically map that like pretty well. I think we have a pretty good idea of the areas where it does make sense over doing, like, a full survey, and that’s basically everything that’s directional and exploratory, right? So, there is so many types of ideas that you like to test, but you don’t because you don’t want—you can’t afford a full survey. You can’t have—yeah, so that’s all this places where we need that. And so product, like, we’re not only targeting, like, research and insights teams, that’s going to be something that will be extremely useful also for product teams, marketing teams, and so on. And what I—

Lenny: ...

Samuel: —and one of my missions for this year is basically to expand to this, other areas, like, pretty significantly and bring research to basically, areas where research until now was not practical.

Lenny: Yeah. So, that would include, I assume, the SMB market, the—

Samuel: Correct, also.

Lenny: Yeah. So, as you’re looking at this… so I’ve seen other companies that have emerged over the last few years in this space that because they were, kind of, AI first, AI native, they’ve done some things I think are incredibly smart where they have moved away from the idea—they have gone where the buyers are. They’ve embedded, via MCPS, into the workflow, so into Notion, or Slack, or, you know, whatever the case may be. I’ve often thought, why is nobody embedding solutions like this into e-commerce platforms, you know, et cetera, et cetera. So, is that what you’re thinking is that there’s, now, you’ve built this marketplace, for all intents and purposes, and to tap into these underserved populations, either because revenue or they sit outside the insights organization, but still they want to be able to test stuff, on making that accessible by embedding into other platforms?

Samuel: Yeah. That makes total sense. So, what I think is going to be the next generational version of what we’re building is going to be a lot around integrations. So integrations, basically, today, is distribution. You get distributions for integrations. That’s essentially how things work. And basically what integrations bring you is two things. One, it brings you data, and obviously that’s key. And two, it brings you basically a way to connect and communicate directly with the customer, right? So, on one end, you basically can communicate with, like, Claude or ChatGPT or Whatever tool the user is using, and on another end, it also opens an opportunity to get these people’s data, basically to feed these models, right? Obviously, opt in, right? Like, I’ll give you an example. Someone, like, manages Meta ads’ budget for this, like, scale 00:22:31, right? And this person wants to use this tool to test ad campaigns, right, to expose these ad campaigns to their target, digital twins of their target buyers. Obviously, if you also basically get this Meta ads data from this person basically to feed their private twins, you’re going to get much more accurate optimization of these ads campaigns, right? And that’s why integrations is key to driving more value. So, right now, all of our focus of the last, like, I would say, like, you know, half-year around twins has been basically on building amazing experience, right? Like, streamlining the creation of audiences of twins from individual level data, streamlining, like, focus groups of twins, questionnaire design, analysis and text generation, all these things that you need to have, like, basically an end-to-end experience, we’ve been focused fully on the experience. The second phase of all this will be basically focused on integration.

Lenny: Okay. So… for our audience that we’re focused late in the day, and we had some challenges because I had—I was in an all day workshop with a supplier client, and this topic, general topic came up a lot, so it’s very top-of-mind. What was interesting through that as we’re talking with them through the applications of all the versions of synthetic was also the integration with specific frameworks. So, you know, somebody has a specific behavioral framework that they applied, let’s say Maslow’s hierarchy, whatever. Are you enabling that as well?

Samuel: Yeah.

Lenny: So, somebody could load up and say, hey, we use, you know, Porter’s Five Forces, whatever, and that’s a frame that we’re going to apply here across a digital twin—at least from the output standpoint—so that we’re restructuring it that way from kind of a guardrail and governance perspective?

Samuel: So, we’re launching on Monday, fully custom questionnaire design. So, there’s chat assistant that allows you to build, like, a questionnaire with all the question types you can think of, like, from, you know, simple single selects to full blown grids of matrix questions and so on. And so, basically you can tell this chat, you know, you select your audience, I want to talk to cyber-decision makers with director plus of cyber at companies of 20 million-plus in revenue. So, you specify that audience, and then basically you specify to the chat, saying, “I want to build a questionnaire based on this basically way of testing concepts”—

Lenny: Got it.

Samuel: “And here are my concepts, and here are pictures of the concept, and here are description of each of these concepts, and also the framework for concept testing that follow is A, B, C, D types of questions,” and it basically builds and program the whole questionnaire, right? And then you can launch this questionnaire to all these twins respond, and then get the results, basically. So yes, that’s basically allowed.

Lenny: Okay. So, you’re doing it more from the data collection analytical standpoint, the conversation I was having earlier today was, can you embed those structurally into the persona overall? So, there’s another company that I know of that’s been—they’re building digital—more synthetic, not digital, not personas, but they incorporated some system one, system two stuff, kind of, foundationally into the system, and that’s kind of their differentiator, was they that they built that in. You’re saying, let’s not worry about that. The here’s the data, and we’ve modeled it out. If you want to apply some type of specific framework within your interaction with the twins, that’s the efficient way to do it. Which makes perfect sense.

Samuel: Exactly. We don’t want, basically, people to, you know, have to bother with thinking about how the twins are answering, right? Because basically, that’s something that users shouldn’t have to worry about from our experience, and we should be able to be the ones controlling how this twins are being modeled. And our assumption is that the right, like, correct and accurate and calibrated reasoning of the twins should be driven more by deep data that’s getting fed to these individual twins rather than smarter ways of modeling. And the reason why a lot of people in the space are talking about these issues with modeling and LLMs and so on and so forth is because they’re give too much power to the LLMs, right? And that’s when you have to start from saying to an LLM, like, “You are an Asian, 24-year-old living with Wisconsin, working in manufacturing,” and so on and so forth and so on and so forth, then there’s so much, basically, degrees of freedom for this LLM to decide what to answer for, and then you have to constrain it through some well-crafted, like, other system one-system two, and it gets too complicated because basically, then you can’t control what’s going to happen. Whilst when you have—and the problem is very expensive—but when you have super high quality category-level data, basically there’s a lot less degrees of freedom that you give to these LLMs to think about how to answer and how to answer right, and that’s more of the approach that we were taking.

Lenny: Have you gone to the, kind of, third-party data resellers?

Samuel: Yeah.

Lenny: Because they sit on a butt-load of data. A lot of people, you know, purchase it, you know, et cetera, et cetera. And I haven’t heard anybody going to so I—[unintelligible 00:28:50] name I’m familiar with Data Axle, a company that collects lots of third-party data for marketing purposes—but I haven’t heard anybody going to Data Axle and saying, “Hey, can we use your data to help inform our digital twins or personas?” Is that—have you guys looked at that at all as well on just [crosstalk 00:29:10]—

Samuel: that’s because people—I think a lot of people are obviously not wanting to share their secrets, but people are—

Lenny: Sure, sure. [laugh].

Samuel: —yes, are doing that. And, yeah, and everyone has a different way to do that, honestly. Like, I think every single company trying to compete in that space has their own, like, secret sauce and approach. And obviously, we have a lot of secret sauce internally, but yes, like, basically, people are doing that.

Lenny: Okay.

Samuel: And you have—unless you have providers going, like, super vanilla, just, like, literally, like having, like, a prompt baked in for personas that are [unintelligible 00:29:54], I think we are probably the most, like, granular and data hungry, I would say, company in the in that space, in the in that quadrant. And yeah, as I was saying, like, I think you really need to have different types of data, and need to have individual level data. A lot of providers are—I think there’s another big debate in the twins community around personas versus individual twins, right, and what makes more sense. Also, a lot of providers are, like, turning one person’s data into many different people, or turning many different people into one blended, like, persona. So, you have basically all of these different approaches, and our internal positioning is that all of these other things basically shouldn’t be done, and only one-to-one individual level twins are right. And then you can aggregate at a second level or layer, right, through asking the same questions to many of these twins from a segment, or through, you know, running a full-blown survey on a full load of one-to-one twins. But you shouldn’t aggregate at the modeling or model level, basically.

Lenny: Yeah, I think you’re right, although I’m sure… all right, it’s like one of those things, you’re like, I don’t want to answer that, Lenny. Earlier this year—we saw some very we saw some significant valuations and raises for companies playing in this space—

Samuel: Yeah.

Lenny: —and… you know, I know you, I know other folks that are building and everybody has the question, “Why did they get that?” Because, as I—I don’t know. They knew the right people [laugh]? I don’t know—because, from what I could tell, I did not see why their secret sauce was better than someone else’s secret sauce, from what I could see. I didn’t see that the data sources that were that—at least they had publicly acknowledged—were better than the data sources that you and others were using. But yet, my takeaway was, hey, look, the rising tide floats all boats. So, obviously Silicon Valley thinks that this category is worth a billion dollars, so that’s a good thing. How do you feel, right? When you’re watching that happen, it’s like, what [laugh]—I’d like $100 million please. Thank you [laugh]. What’s that been like for you to see this growth in the market, but in maybe a different way than we would have anticipated.

Samuel: I would say 200% positively. I think you probably don’t expect, not expect that, and probably—and maybe some people, like, listening to this would think I’m crazy, but basically it only brought good so far because basically it helped educate the market on the category, right? And another thing is, I don’t think you need a 100 million dollar of budget for two years to get something out of the world that actually provides value. And that’s, I think, what we’re basically demonstrating right now. So, that’s the first thing. And the second thing is, having less resources, while having resources, like we’ve raised, like, you know, a bit shy of $15 million, and that basically, I think, provided us, like, you know, more than enough budget to build the right thing with the right focus and mindsets. Now, it’s going to be very interesting to see what happens, again, in the coming year. And our internal approach basically has been to partner with a space. So, we partner with a big share of the great players of the space. And that basically gives you, like, sort of extra, this extra fuel that you need to get things done. So, our personal answer to basically this massive race—races, actually—is basically to work hard, partner with the right people and be laser-focused on everything we’ve learned, like, the last four years, and get it into application at scale. That’s basically our internal positioning.

Lenny: Yeah, I think that’s right and respect it. Yeah, because there’s certainly—I can’t help but think, looking at history, that just because somebody raises a lot of money doesn’t mean that they get me around for a long time, which, if you’re listening, those other companies, and you’re thinking, look, he’s talking about me, I’m not specifically. It’s just looking at the challenges in the marketplace, right? You can throw a lot of money at somebody and they do amazing things with it. You can throw a lot of money at somebody and they fail. Throw a little bit of money as somebody, like the $50 million you’ve raised, compared to, you know, the $100 million, and you do the hard work to build the business from a bootstrap standpoint and, you know, we’ll see what happens with all of those things. But I certainly, because I mostly bootstrap my own businesses, I tend to think, just get out there and doing it. Doing the hard work is the, you know, is the right way to get there and build success. So, that’s very cool. Samuel, what—obviously, we could go on and talking about all kinds of geeky things here—but there is one other question that I am particularly interested in from your standpoint, that in a world where it seems like literally, every day there is a new change in the technology, in one form or fashion, right, whether it’s about computer, a new model, or, you know, whatever, every day there’s something new, and those technologies are the, kind of, foundation of what you’ve built. How do you deal with that? When you’re trying to build a stable product, but yet the underlying technology is anything but stable, from the standpoint of, like, all right, we’ve hit a plateau, and it’s not going to change for three years, no, it’s going to change tomorrow, how do you deal with that?

Samuel: So, I was raised in the cyber… school, actually. So, my first couple of experiences, like, even years ago, were in cyber. And the first thing I was taught when I arrived in Check Point, which is, like, one of the historical, like, big cyber companies, is, you know, as good as you can get, there’s going to be someone on the other side is going to try to attack you; it’s going to be better than it. And the only thing you can do is basically keep adapting and being agile with basically what’s happening on the other side. And I think it is these lessons, like, are very useful today, where basically, things change all the time, and if you’re not agile, you’re dead. So, the only way, I think, to survive what’s currently happening is to build a framework from as a company—as a founder of a company—I have to make sure that my company is basically AI native. And AI native is a big world—you have a big word. You have all sorts of AI-native levels of companies. In our case, so every single function in our company is AI augmented in a deep way, from programming, product, design, marketing. Like, we have so many things that we’ve built in the last six months that just allow us to keep up with the changes that are happening and being agile. I can tell you, it was extremely hard. Like, eight months ago or nine months ago, when I started to realize that there was this exponential acceleration, we basically had a lot of team members that were super senior and that were in their respective, like, areas for the last 15 years. And basically at some point, I had to stop and pause everything and say, now we learn to change and adapt to this new world. And if we don’t do that, we’ll be dead in six months. And so, every single, I would say, employee of Fairgen is now probably three to 4x the person this person was eight or nine months ago. And everyone has to be ultra AI native in everything that this person does. And the beautiful thing is that this is, like, very easy to observe because basically, when you measure productivity from the outcome, like, perspective, you can clearly see that the outcome that we’re getting as a company are basically, you know, much more significant in terms of rates and growth rates than they were eight to nine months ago. So, that’s basically how—so just to summarize everything you just said, one is basically agility, being agile in all of these things are that are happening, and the other one is you have to structure everything around that agility, and not only engineering or not just whatever product or marketing.

Lenny: Yeah. I could not agree more. And even as a… for myself, which I’m not running a full company, right? Just as a consultant—it started with augmentation; now it’s unlocked value creation. So, that’s the productivity thing. Productivity is there, right? I can do four or 5x the volume of work because the time has decreased. But I would also argue the quality is increased and now also that all those things seem to be exponentially connected to think—for me—oh, but what about this? What about that? Oh, that’d be cool. Let’s try that. And the barrier to entry to experiment from an agility standpoint is decreasing as well. Where something that used to take, you know, six months and whatever, $100,000, like, well, give me six minutes. I’m sitting on the toilet, on [laugh], you know, on my phone with Claude of, like, all right, well, let’s—can you build this? Okay, there we go.

Samuel: I can give you a crazy example. I actually posted about it this morning. True story. We decided, we’ll say two weeks ago, to have 50 more audiences for our launch. Now, one of the things you need for each of these audiences is basically primary survey data collection so, like, hundreds or thousands, sometimes for some of these audiences, of people going through ten minutes LOI survey, and also this order enrichment and so on, but that’s on the side. Like, let’s focus just on primary data. We want to do about 50 more to cover basically more countries and more categories that are our early users were asking us for. So, in order to do that, you need to write questionnaires, program questionnaires, find the right partners and field partners, scientist partners, run the field, do the quality checks, and then integrate this data into our product and our audiences and so on. Now, I was talking about 50 surveys, right, so we’re talking about what, 50k respondents across B2C and B2B, and each of these surveys basically a different audience and a different area and different criterias and so on. Like, in what world, like, could we think, even six months ago, that you could do that, all of this within eight days, as a company that’s as a startup, right, with 20 employees that are already all at 100% or 150% of their capacities. And we made that happen. And we had amazing partners, like, we work with various partners, including Opinion and [unintelligible 00:42:38] on the B2C and B2B fronts that were absolutely amazing. But, you know, on our personal end, besides our incredible work making all the fields happen, also all those advisory on [unintelligible 00:42:49] and so on, we still wrote 50 questionnaires, but all of them program tested, we run the quality check, integrated into the product, all of this in eight days, right? I think that’s the best proof that’s basically these days, you can do things you would never have been able to do, you know, even three months ago.

Lenny: Yeah. Well, and tomorrow it’ll be even more.

Samuel: Yes.

Lenny: Yeah, one of the things I keep telling listeners of—I don’t—I’m not sure that there’s a winner in the LLM arena by any stretch of the imagination. So I, personally, I like Perplexity because it’s an orchestrator of. So, it’s like, I don’t need to worry about who’s best; I’ll let Perplexity worry about who’s best based upon the task. Although the flip side—and this gets into the war—is that I’m burning through tokens when I’m trying to build something, and that gets expensive. So, Grok now, there’s no token burn. It’s a flat fee, and Grok can do everything that Claude can do right now. Now, tomorrow, that’ll change. My point is for everybody to get to this example you use, you don’t know what is possible until you try. But my advice would be, don’t get stuck on ChatGPT or Claude or Grok or anything else because they continue to leapfrog off of each other from a capability standpoint. Stay flexible and agile and think about the business, what is the challenge you’re trying to address and find the right solution that can help address that challenge because they have different strengths. Would you agree or disagree with that?

Samuel: I would agree, and I’ll tell you more about how we personally do it. So, I personally have subscriptions to all this, basically, products, like Grok, Perplexity, Claude, [unintelligible 00:44:48], like, you know, ChatGPT, yeah, every single one of them.

Lenny: Yeah.

Samuel: Also for—

Lenny: It gets expensive, though, when you’re just trying to experiment. So—

Samuel: Yeah, but let me show you how the ROI is, like, it’s so incredible, right? So, I use all these products and I keep updating my internal [unintelligible 00:45:09] on what’s best for what. And I spend a lot of time internally in my company across our different departments, basically trying to digest what are the right things to use for the right kinds of use cases. And we always—like, you know, every I think that changes on a, I wouldn’t say weekly basis, but definitely monthly basis, in terms of what gets better than what. And basically, I think the only way is to have someone that’s—or a few people—that are always experimenting with these tools. And by the way, like, this assistant that I’ll just—one side of the story because you also have everything around marketing, like, you know, generating images, picture, videos, creatives, and so on, posts. You have all the coding stuff. Like, and I personally have, like, subscriptions for most. But if you think of it at a level of a 20 people company, if you have one person using all this thing and driving the right, basically, approaches for the rest of the company, actually it’s very dollar efficient because people are delivering higher quality work by using the right tool and they can know what the right tools are through people that are using all these tools. So, that’s kind of how things have been happening, like, over the last 9, 10 months of Fairgen.

Lenny: Okay. All right, so you are swimming in the middle of the AI ocean right now, right? You’re practicing what you preach. The—

Samuel: Exactly.

Lenny: —and that is unique skill set in and of itself. So, that actually—so last question, right—or topic—you know, we recently launched an investment fund connected to the Insight Innovation Competition—maybe you saw that or did not see it—anyway, point is… that’s had me thinking a lot more as an investor. The criteria that we used to evaluate companies a year ago are not the same criteria to evaluate a company today.

Samuel: Yeah.

Lenny: In the world of the potential zero-human company, so to speak, when AI is the force multiplier, but also a teammate, and the need to stay abreast of these changes, understanding the team and the founder and their capability to adapt and to ingest information and to constantly be innovating is a bigger driver of success, I think, than the traditional way that we would think about those things even a year or so ago. Yeah, do you agree? What’s your take on the skill set necessary for success today versus literally a year ago?

Samuel: Yeah. I can tell you that what I see—so we started Fairgen in 2022 right? So, I was really… able as a founder to see basically how decision-making and the pace evolved between then and now. And I would say that I have to take critical decisions now on a daily or weekly, but often daily basis, right? And these critical decisions are, like, you know, I would say far reaching, right? If you consistently have to make calls. You have to make these calls fast. And if you’re as you’re saying, like, not agile, you basically are completely at risk. So one, I would say, good example of that was going towards twins. And we could have stayed comfortable with our boosting business only, which was growing, like, very well and we were quite happy. And we were working with maybe—and we’re still working with, I would say, probably, like, you know, 50% of the top, like, 30 players of this space, right? And the usage kept, like, kept growing and so on, so we could have said comfortably there. But basically what I saw is that, you know, market research is changing and about to change even more drastically, in the sense that, basically, people need to make decisions faster, and the decisions can be directional. Like, people are now accepting not having absolute, like, perfect decision every single time. And that’s why I decided to basically put some chips in the directional business by going to twins. But you know, if I hadn’t been agile, then probably, like, you know, Fairgen would have been at risk. Because today, in the venture world, if you don’t have, like, explosive revenue growth and everything growth, basically you’re doomed. So, what used to be good at the time isn’t good anymore. So, if we hadn’t been agile on this decision and others and basically we would have been at extreme risk. So yes, founders have to be agile. We have to make sometimes crazy calls. Like, sometimes our partners, right, think we’re crazy, right? And I’m very happy that they think we’re crazy because if we weren’t a bit crazy, then we’d be doomed. So, sometimes a bit of, you know, bluntness and fast decision-making is something that can change everything.

Lenny: Yeah, that’s—there’s the great words of wisdom. You got to be a little crazy, right? I hear that old song in my head. Samuel, this has been great. Is there anything that you wanted to touch on that we have not touched on?

Samuel: I think that’s a lot to digest. What I can say is probably the thing that I said a year ago at the end of our podcast, and I said, many, many times is, you know, all of these tools are easy to try and the best thing is to make an opinion for ourselves on what tools are a good fit for our use cases. So, one trait, like, personally trait that we haven’t talked too much about, but I think it’s super important today, is curiousness. That’s something that we talked a lot about, agility, flexibility, being crazy, right, but, like, being curious, I think it’s super important. And the beautiful thing is, these tools are easy to use and easy to try, so I really recommend playing with them, and always happy to hop on a call and talk more about our positioning. But yeah, be curious, and always happy to have a chat about all these things. And yeah, thanks a lot, Lenny, for this great interview again. And I’m very excited for next year’s one, and seeing what [laugh] what the hell we’re going to talk about next year.

Lenny: [laugh]. Yeah. You know, we’ll see. Hopefully you’ll be—that conversation will be you sitting on your private yacht in, you know, the Mediterranean or something, at that point. So, Samuel, where can people find you?

Samuel: Pretty much everywhere, but LinkedIn is probably the easiest place to find us. And you guys can also sign up for twins directly on our website. So, just fairgen.ai/twins, or main website, sign up and you’ll get an invite, basically.

Lenny: Great. Great. All right. Well, thank you so much. Thank you to Brigette and Emma, our producers. Thank you to Big Bad Audio who makes this look and sound good. And thank you to you also for being a sponsor. We appreciate that, and we’ll definitely be talking again soon. So, congratulations on all the success, Samuel, and thanks for making—listeners, if you’re not following Samuel on LinkedIn, you really should. You’re putting out lots of provocative, interesting content on a regular basis as well. So, really appreciate the thought leadership that you’re demonstrating, too.

Samuel: Thank you, Lenny. Thank you so much.

Lenny: Welcome. All right, that’s it for this edition of the Greenbook Podcast. We’ll be back soon with another one. Take care. Bye-bye.