Latest / Cultural Fluency Makers: Unlocking Business Growth Through Cultural Intelligence / How Synthetic Audiences Are Changing Consumer Research | A Cultural Fluency Makers Compilation
Transcript
- CFM Synthetic Audiences - Rough Draft === [00:00:00] David Wellisch: I've been doing these podcasts for, I don't know, a year and a half or, or two years, and I have seen, I have seen a massive shift in acceptance [00:01:00] at least to experiment, w- at least to experiment with synthetic. When I began, you know, um, or even a year ago, people were... The reaction was, "No, no, no. Like, we're all about deeply understanding consumers. There's no way that the machine can give us that." What... And that's shifted. Tobias Puehse: Where I see the opportunity, and you see that a little bit in our research industry because we talk about synthetic clones. In many ways, I actually feel like that's a really good chassis to kind of start thinking about like how consumer agents may behave. 'cause if we train these synthetic versions of people, like that becomes multi-dimensional. It's no longer like this agent will just go out there and find the lowest price and make things very easy for you. But they will start looking at all of your different personality traits and objectives. that's where cultural training and complexity comes into, mind. Now the, [00:02:00] democratization of information is super interesting to me because when we look at cognitive load theory for people, we can retain about four to seven pieces of information. And not for very long. And then often we then start abstracting later on, oh yeah, I think I remember. Or I think this is what I've heard before. and the models don't really forget. the models have a much higher cognitive load. Obviously you can give them thousands of pieces of information. Yeah. And again, there's, there's examples of like new applications coming out where you can scan the barcode, it tells you if it's healthy or not, et cetera. Think that you cognitively would not have been able to do previously. Now we're making it accessible, so, so I'm pretty optimistic that like. It will not. It will actually democratize information and it will make humans reflect their values. And then it's taking away activity that maybe was not something they [00:03:00] either could do or wanted to do or had time to do. Right. And so this is. Were in some ways elevates the human right. ~And I mean, that gives them more power in this conversation because~ David Wellisch: you know, it's so interesting. I, I just, um, did a post this week earlier on LinkedIn and I talked about, um, grounded synthetic data. You know, that there's so much about synthetic personas and data that You know is in the marketplace. And at collage we begin with 15 years of data, You know, actual data from actual consumers that become the seed, or you can then model and, You know, you can, you can go from there. And I call it grounded, grounded synthetic data versus, You know, synthetic data that you can model based on, um, not, not much, not much depth. How, how are you thinking about that? Tobias Puehse: like maybe let's say cynic could say like, yeah, but I could just go into one of the tools to one of the large language [00:04:00] models and just have them create personas and journey maps, et cetera. They know so much. But what I basically then retort is to say. They do potentially At the same time though, everyone else has the same access to the same tool. Correct. And they may tell you something that may be telling someone else something else. And so there is, there is this, there is this, uh, level of confidence. Um, question. Yes. And so what I really believe in is what I call a hybrid system, right? And so we do a hybrid. So when I talk back to the synthetic personas, we train them on like 500 to a thousand observed characteristics of behaviors. And when we, the, the technology and the tool we're using, basically we will. 100% reflect those characteristics back when put on pressure, right? Because it kind of refers to that to saying, this is who I am, this is how I would make [00:05:00] choices or trade offs based on observations. But then it would fill in the, the, the other elements that may not be part of the characteristics. So like it may fill in, um, what, how do you onboard a credit card to this particular merchant app? 'cause that's something that you can source. Open source. So you cannot kind of open source the journey or like what people talk or where are pain points in the ecosystem. But you overlay that with grounded data, as you called it, um, of the characteristics that You know and feel confident about. Yes. And then it becomes that interplay between open source data and close source data that drives competitive advantage because that becomes unique to who you are to uniquely what You know. Uh, but at the same time, you're not like completely closed off to the world. Right? David Wellisch: We've been collecting, you know, cultural relevance, cultural intelligence data for years, and so for us, AI is really [00:06:00] a very interesting innovation given that we've got the lifeblood of the data and seeing what the machine can or, you know, cannot do. But to this, to this, point, the other emerging, um, uh, space is synthetic audiences. Uh, and I'm, I'm curious, what are your thoughts about its use, its accuracy? Clearly there's a benefit of speed and cost. Do you believe that we can-- that synthetic audiences can emulate the human insight? Ciara Dilley: I, I think we're all on a learning journey here, David. I think it's really important that we have the humility, which is a huge part of consumer understanding, empathy, et cetera, to acknowledge what we are, we, what we know, what we understand, and what is evolving. And I think the role of synthetic audiences is evolving. I, curer goes, I'm fascinated, I wanna find out more [00:07:00] because I'm all about where the data can help us accelerate some of the basic knowledge that can take a lot of just groundwork and time, and help us actually get to more interesting outcomes, more interesting outputs with the marriage of, of the empathy side, et cetera. So I think, yes, they can have a role to play. I think we are learning on the journey, and I think just like everything, I think we need to be about it taking over some of the, um, the foundational methodologies, principles that we have already. But I say that in a very balanced way because we need to be agile. We need to be testing. I'm all about testing, David. David Wellisch: Bye Ciara Dilley: it's my prototyping. You know, I love to prototype. I love to test. I l- So let's test, let's learn, l- but let's not be myopic in jumping in and saying, "This is the be all and end all." 'Cause I, I don't think any one thing is the be all and end all David Wellisch: mean, it seems like that's the thematic here, right? Ex- experimentation, agility, testing, data to prove. I- in this space, you know, [00:08:00] similarly, we like the idea of an augmented synthetic, you know, which is Ciara Dilley: Yes. David Wellisch: we give a Hispanic synthetic audience the values, the passions, points, the occasions, everything that we know because of what we've collected, can, can we leverage for certain use cases, can we leverage speed and cost? And for other use cases, we know that it's gotta be pure. So it's a, it is an interesting space. Reshma: Yeah, and It's interesting because my acceptance level has also changed over the por- over the course of the year, right? Like, think synthetic audiences are a great sandbox, a great, you know, sandbox tool for quick, low-stakes sort of testing, right? If you just wanna see how a message lands or test out a few creative options, I think, you know, synthetic modeling saves a ton of time and it saves a ton of money. Um, but I, at least right now, have not swung the pendulum so far to believe that it can replace the real thing. Um, again, at least not right now. I'm, I'm not saying that I think there's [00:09:00] not a future in which it will, but to me right now, a model just can't replicate those lived experiences that I talked about earlier, those true cultural nuances, um, and those real emotional reactions of a real community, right? So I think if you rely solely or entirely on synthetic audiences, think you risk potentially creating, you know, sterile, potentially clinical campaigns that really could miss that human heart of the brand. And so I think, again, with everything, moderation as they say, right? So it's not that we're... you sh- we shouldn't use it or we, we aren't using it, but I think you can't just solely rely and put all your eggs in that one basket. Natalie: From my side, I think it's helpful in terms of, um, maybe some of the more consistent research that you do all the time. So, um, it, it could be helpful in terms of scaling your brand track, for example. If, if you are able to get to a point where you actually, like, [00:10:00] have a synthetic audience that actually, like, you feel really works. But I personally feel like with research and insights, there's nothing better than being in a focus group and listening to what people say because, um, that's where you get, like, real gold for, um, for campaigns. Like, I remember I was in a recent focus group where we were talking about, like, um, loyalty platforms, and one consumer said to me, like, " You know what? The problem with all loyalty programs is it's just like homework. Like, all of you brands just expect us to do all the homework for you, and we don't wanna do math." And those type of comments are like, you would never get that from a synthetic audience. Well, at least not now. Um, and that's kind of marketing gold because that helps you unlock such an incredible insight and a creative idea that you can jump off. You can be like, "Wow, that's-- Thank you for that." Um, so, so whilst I do think that there's ... I mean, I do think research and insights, there's obviously gonna be a huge amount of AI disruption because there's a lot of [00:11:00] very laborious tasks in research and insights. You know, there's a lot of, like, cleaning of the data and, David: Yeah. Natalie: Where I think research and insights can go is, you know, having agents to help with all that cleaning of the data. And then you have the brilliant minds that analyze the data and bring, um, bring those insights to life for their teams. David: You know, it, it's fascinating. You-- As Kolash, we're at the center of so many brands and in these conversations, and, um, there's such a spectrum. I mean, there are brands and people that are all in, and they've built, you know, hundreds of audience, of synthetic audiences, um, that represent the consumer, and then there are the ones that, um, that are more on the, you know, on the other side of it. And I agree with you that ultimately these go-- You, you called it gold, right? Like these kernels of insights. When you, when you get to one you know, that can be the reason and the impetus for an entire campaign. you know, but this will be a fascinating space to watch, [00:12:00] and we're getting into some experimentation. We're lucky because we- we're sitting on so much proprietary data of years and years that allows us to then experiment. Um, Natalie: And I mean, you guys are at the forefront of this. I should be asking you this question. David: Well, we, we... Again, what's fun is that we can actually Natalie: Exactly David: and co- and that's what we're, we're gonna be doing. Um, and then we can also-- It'll be fascinating because you can create based on all of the data that you have already collected over years, which I would assume make... You know, I think it's a, I think it's an augmented synthetic. It, Natalie: Yeah, I like that. I like that idea. So it's basic-- Because, um, similar to sort of the creative lens, like often you need to build like a very good sort of strong 3D model of something before you can put it in an AI tool and get the best creative. David: Gotcha. Natalie: So I agree. It's probably like a little bit of a mixture of both, where, um, it's super helpful to have like the real rich [00:13:00] data. But I think that, David: Exactly. Natalie: That, that I think a lot of insights teams have is that, you know, by the time they're producing the quarterly report, like the data is kind of outdated. So it's, um, there's obviously a need there to have like more real-time things. Um, and I think that definitely like AI and technology can super support there, um, whether that's through synthetic audiences or inputting the data and analyzing it more quickly or whatever it may be. But, um, it's definitely ripe for, for AI to support in that area in a big way. Yeah. Koley Corte: So we're, we're testing the use of them. I think they can definitely be a useful tool. Like I'm trying to figure out how we can accelerate learning, how can we supplement audiences to scale things. But I am also concerned about this reversion to the mean, right? Which is, um, if you are asking, um, a synthetic audience at some point, how are you just not, um, getting to the [00:14:00] average, right? Um, and you're losing some of the nuances and some of the long tail of what is the art of the possible. Um, but I am trying to figure out how can we speed up, um, insights work, right? How can we better understand customers faster? And I think there are other use cases than just synthetic data. I think we are exploring using AI in the survey process, using it to better screen audiences, using it to create more engaging surveys, David Wellisch: Yeah. Koley Corte: And then using it also to, to synthesize, um, faster. And again, I think similarly though, you have to be careful. I read something, um, week that someone gave, um, a tool, uh, two audience samples of fifteen hundred each, and it said that one set of samples was a US sample, and one set of samples was a UK sample, and it asked to tell the difference, um, share insights between these two samples. [00:15:00] And the tool gave back a very, um, reasonable explanation that the US was more direct, they had more clear feedback, a bunch of other things, and that the UK audience had these different behaviors. Except it turned out that they gave it-- gave the tool the two samples were exactly the same. And so the tool introduced its own bias around what it knew from other things about cultural-- uh, um, David Wellisch: SAML Koley Corte: insights, but it wasn't true. And so I think in each of these steps, whether it's a synthetic audience, it's using AI to better synthesize, it's, um, how you're vetting your audience, I think we always have to go back and verify, right? because just like humans introduce bias, tools can introduce bias too. So, um, it was just something I... But I think there are lots of ways that we can speed up the insights and, and then be able to take action in ways that are, are meaningful. Pankaj Chopra: See, we are all experimenting. Uh, and I talk to my peers, you know, [00:16:00] obviously, and we talk about synthetic. And synthetic, we've been working on synthetic for the last two and a half, two, two and a half, three years. Synthetic is helpful in some use cases like forecasting, where you want to enhance your predictive modeling, where you want to kind of, uh, refine your scenarios based on robust statistical data. So that's where synthetic becomes helpful because it gives you the granularity, it helps you to kind of peel the onion a little bit uh, helps you to make the tool a little more robust Synthetic becomes a bigger challenge when you have a lot of more interpretation that is needed, uh, for the data. So for example, if you say, "Ankit, can you start using synthetic data for consumer trends?" I would start getting very worried because you know how it works. Like synthetic data is eventually generated. It's synthetic for a reason. That's what it is, David Wellisch: That, that's right Pankaj Chopra: I [00:17:00] would not personally at this stage, and I haven't seen any successful use case where synthetic data has worked in a little more ambiguous environment. So... And even when I talk about volume forecasting, you need synthetic data and parallel path your current predictive models so that your quality of synthetic data improves as you train that David Wellisch: Yeah Pankaj Chopra: data generation. then once you feel that, hey, your deltas and your, your errors are minimized or zero or almost zero, then you can go on synthetic data. But challenge, David, is that, you know, it's nice to talk about synthetic data, right? But if you make a business blunder because your synthetic data is not giving you the right signal, it could cost you millions of dollars. So as a responsible AI leader, I would be very careful on synthetic data. Experimentation is great. I love it. It's a lot of fun. But would [00:18:00] only rely only on synthetic data when I have proof for a fairly long time the data gives me the right David Wellisch: Did the right Pankaj Chopra: and is guiding me in the right direction. So we are in a journey David Wellisch: So, you know, it, it's interesting in the, um, research technology, the ResTech arena, smart money is pouring millions and millions of dollars in synthetic players. Are you, are you a... Wi- with one of those crazy valuations that, you know... Are you contra- are you a contrarian in this? Like, do you think that those pan out over the long run? Pankaj Chopra: See, synthetic data eventually will David Wellisch: Yeah Pankaj Chopra: Uh, there's no doubt. It's just that... that it'll pan out for a few reasons. As I said, you know, if it is initially use case specific, there are certain domains where synthetic data is more relevant. It'll pan out for companies who are able to use [00:19:00] proprietary synthetic David Wellisch: Yes Pankaj Chopra: and integrate it in the system. But I'm not sure that in the next four s- so long term, probably after 50 years it'll all pan out, right? if you're talking three David Wellisch: Yeah. Pankaj Chopra: I'm not sure of the synthetic data providers will be able to drive value. Because as practitioners, still haven't seen the proof in the pudding to the extent where we can say, "Hey guys, we're all in." We are not all in yet. We are still working on it. So if you have a three-year horizon, and if you're invested heavily in these stocks, David, not bet my life David Wellisch: Yeah. Pankaj Chopra: If you have a 50-year horizon, I'll go with it. david-wellisch_39_05-05-2026_161824: Do you, uh, do you have an opinion on, um, the validity of synthetic audiences as a way to understand consumers faster and more cost-efficient? Daryl Butler: Uh, it gets a little, it gets a little david-wellisch_39_05-05-2026_161824: Thaisey? Daryl Butler: interesting, dicey because, [00:20:00] um, a synthesized audience still is manufactured. david-wellisch_39_05-05-2026_161824: Yeah. Daryl Butler: Now, it may be manufactured based on a, on, on a wealth of input, I mean, I think we can-- we know the difference between something that we have in our lives that's manufactured and it's something that's real. I mean, it's, it's, it's just different, and you, and you know that. And that there may be a place for it to a degree, but not where it becomes, um, the predominant source, right? There's nothing like I, I, I, and I keep saying this, Skyline High School is not, you know, it's a stone's throw away. and I'll go up there, you know, during a basketball game or during football season, and I just sit and watch, absorbing. I'm seeing how kids are interacting. you know, I just did a, um, a career day at Bishop O'Dowd High School about a week ago, and, you know, had an au- a auditorium full of high school [00:21:00] kids, right? That interaction, one to many, and even one-on-one, and in some of the breakout groups is invaluable, right? Now, sure, it's microcosmic relative to a, a, a broad audience size, but it still is a great reference point, right? Because what I heard at Skyline what I heard at Bishop O'Dowd is the same thing that I hear from my kids from when they were teenagers. So there are themes that you can pick up if you're paying attention, if you're listening. can you do that synthetically? Yeah, maybe. david-wellisch_39_05-05-2026_161824: Yeah. Daryl Butler: is it going to-- Are you gonna feel the pulse? Are you gonna feel the heartbeat? Are you gonna have the eye-to-eye contact and the true appreciation of what the excitement is when a kid talks about Playboi Carti, right? I mean, yeah, you might get that, um, again, synthetically through systems, but it's nothing like being in the room full of young men, 16, 17 year old, talking about this person is better [00:22:00] than Jay-Z, and this person is better than... you know. And obviously I'm a, I'm a, you know, I come from a different era. So there's a debate that happens right there, time, the excitement and enthusiasm can't be replicated, right? That's the difference between human interaction and, using a system. david-wellisch_39_05-05-2026_161824: I love your energy about the interaction. I mean, I can feel it. You know, I can feel the energy that you derive from being in the room, Daryl Butler: Being in the david-wellisch_39_05-05-2026_161824: right? Like being in the room. And I'm sure that the nuance, I don't know how we replace the nuance. Maybe the information and the insight might even be the same, but, you know, life and energy from that nuanced interaction, it's hard to replace. kajoli-tankha_1_09-08-2026_103319: You know, I haven't spent a lot of time with synthetic communities, but I do have experience with one tool which, uh, in-- that, uh, my peer has built, and it is mind-blowingly helpful. So, uh, my, [00:23:00] my experience has been extremely positive what I have seen. Yeah i'm also a researcher with twenty-five years of experience. So if there's something like brand... Because the, the digital twin is only as good as all the research that have fed into it, yes. Right? So I need to have the brains to know that I'm looking for something that hasn't been researched at all. Yes and so, so for me, my digital twin work, we know-- which is called the, you know, anyone can look it up. I think it's in, um, I think our CMO has talked about it on LinkedIn. It's called the AI Messaging Assistant. Yeah is kind of my starting point for anything anyways. So you kind of go into it, you have a blank workbook. If you're even thinking about how to, um, you know, like how... What is a general way in which I should talk about something to customers? You don't even have concepts. You don't even have messaging. Based on everything we know, what is a general way in which I should be talking about, [00:24:00] uh, this idea to customers? You can just talk to AMA and get like a back and forth going and get some input. And because all of our audiences are loaded in, I can totally see, okay, this idea is really gonna work with iWorkers, but it's not gonna work with consumers. You know? Like, I get some initial hypothesis, but depending upon like what the s- what the thing is, I can either, um, decide that I'm gonna do some, uh, new research with new customers or this is enough. So that judgment I have to make, and that judgment is very important, but honestly, I couldn't live without our synthetic tools right now, 'cause now I'm used to them david-wellisch_47_09-08-2026_133316: Yeah. You know, it's, um, it, it's very interesting. We, for 15 years, obviously in the cultural intelligence world, yes cultural nuance, people, I mean, it's, it's, it's so, it's so critical, and we have begun to experiment with synthetic digital twins, uh, because to your point, we've [00:25:00] got so much data. And so we can really embed all of that data in terms of values and passion points. Um, and when you have that, um, and, and we're just in the process of validating. And so it's a, but it's a fascinating, yet another fascinating, uh, chapter in our ever, ever-changing, ever-changing world, um- kajoli-tankha_1_09-08-2026_103319: I think you've, you've got something there that you didn't-- I'll make explicit what you were implicit about, is that I think approaching all of these tools with some curiosity, yes. Yes that's the point. I think, y- you know, you don't have to... Like, people, uh, you know, when- whenever you do research in the market, like AI sentiment tends to break out with pessimists, skeptics, and optimists. Yes kind of typically how things break out. And, you know, just like I always feel that we shouldn't let our own biases hold us back from exploring new things. I think what you said, like you didn't first-- at first you didn't think that, but now [00:26:00] you have this huge library. And, and David, I might even make a parallel, like people like you and me with lots of experience who have done research a lot, like we ourselves are LLMs, right? You know, we are, we're... Yeah. Right. Of course. Uh, you know, so I mean, how-- imagine how many times you've kind of, you know, your, your client gives you a brief and you look at the brief and, uh, you kind of know somewhat of the answer already yes. Yes sometimes you don't. So, you know, in the same way I would approach like synthetic data, uh, which is basically an LLM version of the average of your customers, and sometimes it's really helpful and a few times it's not. And you know, you gotta decide when it's helpful and when it's david-wellisch_47_09-08-2026_133316: Yes. And, and I, [00:27:00]