Unpacking Synthetic Data and Digital Twins with Koen Pauwels

[00:00:00] Koen Pauwels: My own research, and this is, you know, Europe, uh, uk if I can still call it Europe, and the US actually shows that the more senior your position in a company, the more you’re likely to over rely on ai. And so, and so, it’s the more powerful people that are actually overusing AI more than the more junior people.

[00:00:18] Koen Pauwels: Research just out in general of marketing, not by me, has showed that the least you know about gen ai, the more you’re going to use it. So, so there’s a danger that the. People who are least aware of the potential biases and caveats actually use it the most 

[00:00:33] MRII Announcer: Welcome to MRII’s Insights and Innovators podcast, where we talk to top market research professionals to get their inside stories about innovative and enduring best practices.

[00:00:43] MRII Announcer: Now here’s your host for today’s episode. 

[00:00:47] Niels Schillewaert: Hi everyone. Uh, welcome to today’s episode on unpacking Synthetic Data and Digital Twins. I’m Niels Schillewaert, your host, and I’m a member of the MRI Board, as well as the head of Research and [00:01:00] Methodologies at Conveo. Today we’ll be talking about AI and how AI and synthetic data, digital twins.

[00:01:07] Niels Schillewaert: Changing the way marketing researchers understand their consumers and ultimately make, uh, decisions. I’m very pleased to have our guests today, Koen Pauwels, uh, who I’ve known, uh, when we were doing our PhD together, uh, back in Europe. Uh, so good to see you again, uh, Koen after all these years. Great seeing you.

[00:01:27] Niels Schillewaert: For people who don’t know, Koen, Koen has, uh, had a tremendous career, academic career. He’s currently the distinguished professor at Northeastern University. He’s also the editor. Chief of one of the most leading, uh, scientific publications in the field. Uh, the International Journal of Research and Marketing.

[00:01:45] Niels Schillewaert: And he’s also one of the world’s leading voices, uh, when it comes to applying, uh, AI and marketing and market research. The podcast will be centered, uh, or is triggered by the fact that Koen recently convened and helped [00:02:00] lead a two day conference at Columbia University about synthetic data and digital twins.

[00:02:05] Niels Schillewaert: The nice thing about the conference, it brought academics and practitioners together, um, to share what’s working, what’s promising, but also what the pitfalls are about digital twins and ai. So in this episode, uh, Koen is going to share. Five practical lessons for marketing, uh, and insights professionals. Um, we’ll learn a little bit how to embrace, uh, this new transformational technology, what to be cautious about, but also how to make these new tools fit in through the way we work.

[00:02:34] Niels Schillewaert: So Koen, welcome to the Insights and Innovators podcast. 

[00:02:39] Koen Pauwels: Great to be here. Nielsen, great to see 

[00:02:41] Niels Schillewaert: you again. Yeah, absolutely. Koen. Let’s start simple. Um. It’s a buzzword, right? Synthetic data and digital twins. Could you maybe explain what it actually means for marketing and marketing research and why we would be, why would we care about it?

[00:02:56] Koen Pauwels: Yeah. Well, so synthetic data in general, uh, [00:03:00] is, is basically, uh, you know, generated the data by the machine, by AI that has the same properties as, uh, the actual data would have. Right. So, I think originally really started from a view of privacy to make sure that we get all the learnings from you. For instance, you know, I worked at Amazon ads.

[00:03:18] Koen Pauwels: And we want to get all the insights we get about what people like with which combinations to give recommendations. But we don’t want your personal data to get out. So the origins was really kind of to have a nice, um. Kind of summary of, of what people are doing, uh, but then in a more synthesized way. Uh, right now of course we have digital twins and digital personas distinction that I only learned during the conference.

[00:03:43] Koen Pauwels: So digital twins are really, um, um, kind of, uh, a statistical algorithms to, uh, to resemble you personally. So your digital trend as Neils, uh, is supposed to react like you similarly to. In particular marketing stimuli. And then a digital [00:04:00] persona is, is again, synthesized data, but it’s, uh, trying to represent the whole segment.

[00:04:04] Koen Pauwels: So if you think about personas in the traditional market research here, now we have some, uh, you know, some algorithm that can react as the representative person in the segment. 

[00:04:16] Niels Schillewaert: Got it. Interesting. I like the fact that you referred to synthesize data and not synthetic data. 

[00:04:21] Koen Pauwels: We just said that in the preamble.

[00:04:23] Koen Pauwels: Yeah. I really like, it’s kind of like a synthesizer in the music when we grew up in the eighties, right? Yeah. That it’s, yeah, it’s uh. Yeah, I like the analogy too because you know, we used to go from pianos and then to synthesizers and the sound is a bit different, it’s a bit more artificial and we’ll talk about some biases, but it is kind of supposed to make it easier accessible for people to play it and in market research to do something with the, with the insights.

[00:04:47] Niels Schillewaert: Yeah, absolutely. And synthesizers have made good music. We Exactly. Just listening to deep 

[00:04:54] Koen Pauwels: mode on the way here. Exactly. 

[00:04:57] Niels Schillewaert: Yeah. Very good. Um, so if, if, if you [00:05:00] think about it, the traditional like methods survey, panel research, the, the traditional research toolbox, how do these tools, these new tools differ from, from those more traditional tools you would say?

[00:05:12] Koen Pauwels: I would say, uh, I think the major difference is, as I said, convenience and speed. Um, and I’m very much influenced by my good friend Mark Ritson, right, who basically kind of stated that, Hey, you know, there are some, you know, caveats with how representative and how realistic that data is. Uh, but on the other end you get it way faster so you can much faster.

[00:05:35] Koen Pauwels: Put it in your production system to get insights, to get marketing strategy, to try out different options, to build scenarios. And so I, I really see it as a way to first of all, you know, speed up the process and number two, uh, make room for some really crazy ideas that you and I have also traditionally in marketing.

[00:05:56] Koen Pauwels: And that may be just kind of sweat by the wayside because [00:06:00] it would be too expensive or too long to investigate whether they’re viable. And now you can test them out a bit faster. 

[00:06:07] Niels Schillewaert: So would you say that there’s actually instances where this is a method where you would do research and otherwise you would follow your gut completely.

[00:06:15] Niels Schillewaert: Is that kind of where you see the, the difference in a way? 

[00:06:19] Koen Pauwels: That’s a great way to put it. So it lowers the threshold of being a bit more structured and systematic versus, uh, following your instinct because you have to make a decision right now and you just don’t want to, uh, spend the time or, you know, talk to a market research specialist, uh, to get, uh, to get the hundred percent information on the data.

[00:06:38] Niels Schillewaert: Yeah. Excellent. Very good. Um. When we think about the conference that, that I just mentioned in the introduction, uh, what was the, what stood out for you the most about how companies are already today, uh, using these digital twins and synthetic or synthesized data in, uh, in practice? Yeah. 

[00:06:56] Koen Pauwels: Well, what really stood out to me is how wide [00:07:00] prejudice in the sense that at the conference there were several very traditional market research, uh, you know, uh, techniques such as conjoint analysis, uh, such as segmentation, targeting, positioning.

[00:07:11] Koen Pauwels: Such as new product kind of, uh, viability assessment and all of them benefited in some way, shape or form, form artificial data. So I came much more into the conference as a skeptic. Uh, but I left as, as you know, much broader view of, of all the different, uh, parts that that. That it basically touched in marketing.

[00:07:32] Koen Pauwels: And originally, actually before the conference, I thought that this was the very last thing that, uh, AI would not be good at. Um, I mean, I could see AI making lots of my processes easier also in companies, uh, but I. Could now see it kind of replacing, actually asking a representative sample of consumers what they think about something.

[00:07:51] Koen Pauwels: And so I left with Fielding, Hey, you know, I have to adjust my opinion. And there may be several, several conditions where [00:08:00] kind of you need good and enough information in a relatively fast way to make better decisions. 

[00:08:05] Niels Schillewaert: Yeah, absolutely. And it is amazing how. You know, it’s still, it, it’s not there yet in terms of creative interpretation or, or, or maybe not as good in critical thinking, but it is also doing already much more than just the tedious jobs.

[00:08:18] Niels Schillewaert: Right. It’s, it’s, it’s, it’s kind of in the middle right now. Um, so that’s interesting. Um, thanks for help helping, you know, set that stage. Um, let, let’s dive a little bit deeper in, in a couple of lessons learned from, from the conference at, at Columbia University. Um. Do, keep in mind our audience is, is fully learning, uh, and is primarily also a, a research practitioner’s, uh, audience.

[00:08:42] Niels Schillewaert: Yeah. So, you know, not academic. Um, so let’s keep it grounded and, uh, but also meaningful and, and, and applied. Um, you mentioned before that, you know, when we were talking that one of your observations is that synthesized or synthetic data. Should augment human insight rather than [00:09:00] replace it. Can, can you give a, uh, an example of, of where that worked?

[00:09:04] Niels Schillewaert: Uh, um, that was mentioned at the conference? 

[00:09:06] Koen Pauwels: Yeah. So I, I really like the example. Uh, this is my colleague just across the river here, right? A at Australia, at Harvard Business School. So she has fascinating research about, um, whether synthetic data, digital twins can basically replace, uh, human, uh, based conjoint analysis, right?

[00:09:25] Koen Pauwels: I love conjoint analysis. It’s the one tool that marketing really gave to the world, right? Completely independent of any other, uh, discipline. And it’s so fantastic because as you know, it, it helps to assess kind of new product ideas. It puts respondents in an actual decision like situation, which is supposed to give, uh, more predictive, uh, insights than, for instance, asking them in a survey.

[00:09:48] Koen Pauwels: And so, uh, so one of her big findings is. Um, basically synthetic data is very good in, uh, in predicting preferences, uh, of humans as [00:10:00] long as the training data actually has very good human insights. And so the typical example is to, uh, imagine, uh, toothpaste flavors. That work well on foods, but work horrible in toothpaste, right?

[00:10:14] Koen Pauwels: Such as, you know, cheesecake or, you know, Thanksgiving, Turkey. And so AI devoid of humans were, was actually way too enthusiastic about it. So AI is overly enthusiastic about trying new stuff and is also overly rational. Whereas humans are not, we get an ick at certain flavors. And so once it was trained on human data, though, it did very well in predicting also new flavors.

[00:10:40] Koen Pauwels: So within a certain context, it was pretty good as long as the training data worked well. However, when she tried to expand it to a different category. Even when it was very similar attributes. So she was looking from, you know, the, the smartphone to, uh, to tablets for instance. Uh, AI went completely wrong even though it [00:11:00] should have been relatively obvious.

[00:11:01] Koen Pauwels: So there are kind of contexts that are close enough to your training data with actual human respondents where it worked well, and then places where obviously a human evaluator would be like, no. It’s just, it immediately a horrible idea to put that flavor in toothpaste that AI just misses. 

[00:11:20] Niels Schillewaert: Yeah. Yeah.

[00:11:20] Niels Schillewaert: Okay, good. Uh, well, luckily, I would say, um, it seems like AI still needs a human eye, right? So, uh, that’s, uh, that’s an important it is. We 

[00:11:30] Koen Pauwels: we’re not out of a job yet. I think every conference in the last two and a half years, I’m asked, Hey, aren’t you afraid that we’ll replace you as a market pro marketing professor?

[00:11:37] Koen Pauwels: And I’m like, if only I would love to retire Neils and leave it to ai, but it’s just not there yet. 

[00:11:44] Niels Schillewaert: Yeah, well, Stanford did something, uh, you know, pretty, uh, pretty changing. But anyway, that’s a different, that’s a different discussion. Uh, what do you think happens when companies or humans, executives in companies would try to rely only on synthetic data instead of [00:12:00] their own human judgment?

[00:12:01] Koen Pauwels: Yeah, that’s a very pertinent question because, you know, uh, my own research, and this is, you know, Europe, uh, uk if I can still call it Europe, and the US actually shows that the more senior your position in a company, the more you’re likely to over rely on ai. And so, and so, it’s the, uh, the more powerful people.

[00:12:19] Koen Pauwels: I, you know, and we are looking at why this is. That are actually overusing AI more than the more junior people. And, and there is a big danger in that, I think, for two reasons, right? So, so number one, uh, you know, research just out in general of marketing, not by me, has showed that the least you know about gen ai, the more you’re going to use it.

[00:12:39] Koen Pauwels: So, so there’s a danger that the people who are least aware of the potential biases and caveats actually use it the most. Number two, uh, you know, gen AI continues to be, you know, riddled with errors, I think specifically for more strategic decisions. And so if you are the person in your organization making more strategic decisions, the human [00:13:00] element is, is still so important.

[00:13:02] Koen Pauwels: So, um, you know, thinking about, hey, you know, what are some of the, of the errors or the ways, uh, of thinking that, uh, gen AI or synthetic data can lead me astray is very important. I would say the flip side, how I’m using it personally also in my academic research is I use gen AI to counter argue my, my arguments.

[00:13:22] Koen Pauwels: So, so, so I’m like, Hey, what would. What would the consumer, or why, why would my retail partner or in academia, what would the reviewers say against my thesis or my direction? And so if you really interact with it in a longer conversation, it’s way better than just taking its first answer as a given. 

[00:13:41] Niels Schillewaert: Yeah, absolutely.

[00:13:42] Niels Schillewaert: That’s interesting. So if I interpret that, it’s more about collaborating with AI rather than relying on it and, and be critical and make sure that you. Ask it to be critical about your own findings. Um, and that’s a good, that’s a good, that’s a good takeaway. Um, [00:14:00] and you also mentioned, you know, it seems to depend on like the type of decisions, like when it becomes more strategic or, you know, it gets more dangerous, which triggers with me is, is AI fitted for every type of decision, for every type of project?

[00:14:15] Niels Schillewaert: Um, you think in your opinion. 

[00:14:18] Koen Pauwels: No, and, and I would go back to our joint experience in there was the Dutch reach research networks right in the nineties, uh, Niels, where we had, uh, the Great Be Wier in in Rotterdam. And so here’s some great papers with Ger Berger about types of thinking. And so, uh, so AI is, is definitely not good yet in what, what he calls analog thinking.

[00:14:39] Koen Pauwels: So a lot of stuff we do as, as human decision makers also in marketing is based on an analog regioning, which means we take our insights from, from other. Places, other categories, you know, other industries we worked, maybe parts of our lives, even art or books we read and we say, Hey, this situation looks similar.

[00:14:59] Koen Pauwels: It’s [00:15:00] almost like a grand chess master, right? Recognizing a situation on the board, and it’s not identical to the past, but it reminds us and it makes us make. Great decisions, which some people may call gut instinct, but it’s really based on our experience and using this analog reasoning. And so, and, and the conjoint analysis of of Ala Australia is a great example.

[00:15:21] Koen Pauwels: Uh, humans are still way better at, at this analog reasoning. And thinking through kind of which analogies would apply and would apply less. Uh, so for much more optimization slash execution decisions, uh, I really like gen AI because it forces you as a human to be more consistent. So AI kind of excels in consistency.

[00:15:43] Koen Pauwels: You know, not having the Monday morning feeling, uh, you know, not being irrationally exuberant or irrationally negative depending on your mood. And so I always thought from the conference, Hey, I love my digital twin, but not so much in market research, but to help [00:16:00] guide me as a consumer, making better decisions like, which subscription should I keep, which one should I cut versus, uh, you know, making more big, strategic life changing decisions.

[00:16:11] Niels Schillewaert: Interesting. I like that. Um, how about this is something that, that, you know, triggers me or bugs me, or that I often think about is, is synthetic data good for, for decision making? When, and, you know, there there’s a big issue in, in market research around data quality. And it always been, it has always been an issue, uh, or a topic to pay attention to.

[00:16:35] Niels Schillewaert: Um, how about synthetic data? How do I know if it’s good? 

[00:16:39] Koen Pauwels: Well, I mean, the proof is always in the pudding, right? So I’m a, I’m a big, uh, fan of experiments, right? Uh, as, as you know, I’m an econometric modeler, but my framework is meme, model, experiment, model, experiment in the sense that, uh, anything that can help you get buy in for a cool experiment with your own company, right?

[00:16:58] Koen Pauwels: Whether that’s [00:17:00] estimating a marketing mix model or asking AI and, and, you know, and. Iterating till you make sure that, you know, the biggest risks are probably out in the open are asking your favorite advisor. So everything that can get you the, um, the motivation and the buy-in to actually experiment and, and, you know, and, and see what happens if you apply to your own company is, is very useful in my regard.

[00:17:23] Koen Pauwels: Right. Um, but, uh, but, but you do have to be, you know, you do have to make sure that you actually kind of try it out. So, so, so think about kind of, uh, hey, does, does the kind of of information and the training data, does that align with the context that I’m gonna apply it to? And then of course, you know, how are we gonna test this out, and how certain do we have to be.

[00:17:48] Koen Pauwels: In that sense, I’m a bit kind of fan of, you know, you don’t need a hundred percent information to make a decision. Right? At Amazon ads, we always said 70% is plenty. Uh, and most companies that come to me for [00:18:00] marketing optimization, they, they never follow my optimal advice. They just want directional advice.

[00:18:05] Koen Pauwels: So if I say, Hey, this is how much you should spend overall. This is how much you should price, let’s say 20% more. They don’t increase price by 20%, they increase it by 2%. See what happens, and then adjust. And so most decisions that we make, uh, first of all, you know, they’re two way doors. We can always adjust them.

[00:18:24] Koen Pauwels: Number two, uh, you know, we will only move in the direction that the data tells us or that Jenny tell. Or that some consultant tells us. So I would say most decisions, uh, that you can easily kind of adjust and you just need to buy in for the decision. I think it’s a great companion in our toolbox.

[00:18:43] Niels Schillewaert: Fabulous. Um, good. So that’s kinda like a second, uh, learning, choosing the right decision as well as the right project or the right problem. The right role for AI to play in our, in our decision making. Um, we touched upon it, [00:19:00] you, you mentioned something in the beginning as well, and it’s also a, a, a topic, uh, that I think surfaced at the conference is all about bias, right?

[00:19:08] Niels Schillewaert: Yeah. Um, how, how does that play out with digital twins? Um, what, what type of bias should we be careful about? 

[00:19:18] Koen Pauwels: So, I mean, the key things that coming outta the research in the conference is number one, as I said, uh, you know, your digital twin or digital persona for your segment is just more rational than, uh, the consumers that they represent.

[00:19:31] Koen Pauwels: So they’re way more consistent in their choices. Um, they’re less situation dependent. Um. And so kind of on average that may work out, right? So if you launch a new product, uh, and then you, you know that I always say marketing is more like chemistry. You never know what each molecule is going to do. You have some average, uh, tendency of, of, of consumers.

[00:19:53] Koen Pauwels: So AI may well represent the average kind of, you know, what people are gonna do. Um. You [00:20:00] know, in a more rational decision world. Uh, the other thing is that, uh, and that may very much affect new product success, that AI tends to be much more enthusiastic about trying new stuff than actual users are. And in that sense, ironically I see it more like in a survey.

[00:20:15] Koen Pauwels: So if you ask people in a survey, specifically, when you introduce your new product, Hey, are you gonna buy it? People typically overestimates their buying propensity. And so, uh, I have a sense that, that, that chain AI is, is very much like that. Uh, it’s, it’s a little bit aspirational in terms of like, Hey, we wanna try new things.

[00:20:35] Koen Pauwels: We wanna see if, if some, some, some, uh, product launch is better than our current product. In reality, people are, you know, creatures of habits. And so, uh, so basically you have to adjust for some inertia. That, that actual consumers have. And then finally, and this is uh, of course as expected, uh, gen ai, synthetic data, digital twins, they’re just way more enthusiastic about, uh, artificial intelligence than [00:21:00] consumers are.

[00:21:00] Koen Pauwels: So they tend to kind of think very positively about the great opportunities that AI presents to the worlds versus, uh. All of the caveats. So there’s been lots of research that shows that if you, uh, offer something to consumers and then later on disclose it was made by AI or supported by ai, they evaluated less.

[00:21:21] Koen Pauwels: Even when you could show rationally it’s better for them, right? It’s faster, cheaper, or even higher quality. And so, and so that kind of, you know, you could call that the human bias, uh, against ai. Uh, that’s of course largely absent. Uh, when you’re asked this to AI itself. 

[00:21:38] Niels Schillewaert: Interesting. It’s AI bias and human bias.

[00:21:41] Niels Schillewaert: Again, you know, um, research is all about limiting bias, right? So, uh, very much so. It’s one of, it’s one of the ground rules. So it doesn’t change in a way. Um, interesting what you say is, you know, AI is there to please, it’s very rational. It’s, it’s, it’s more enthusiastic [00:22:00] about AI and probably technology at, at, at large.

[00:22:04] Niels Schillewaert: My reaction would be. Why not combine AI with real human input? Like not only in the decision making, but also in studying phenomena. Uh, what’s your, what’s your take on that? Combining synthetic data, twins with company data that’s already there. Um, new primary data that’s fed in. Um, could you explain or, or tell me your opinion on that one?

[00:22:27] Niels Schillewaert: Yeah, 

[00:22:27] Koen Pauwels: well I think it’s really kind of the, the view of the future, right? Uh, so, so first of all, kind of combining a more general of the shelving with your own data is, is absolutely crucial. Uh, so at Northeastern University where I work, uh, we have, uh, a clot. And so that’s a nice firewall. And then. Claude also really knows me, and I have several tabs open in different conversations because it is pretty good, uh, after a few iterations to really capture my style or my way of thinking, and, and I give it very specific feedback, [00:23:00] right.

[00:23:00] Koen Pauwels: And then yes, in that more secure environment, I can combine it with, with my own data to come up with even better solutions. Uh, why, why are humans so important in this one? As I said, there are still a lot of blind spots in ai, and I don’t think it’s, it’s merely a matter of technology, right? Technology and test will always say to you, oh, but this is the worst version and it’s gonna be so much better in the future, and new technology is gonna change all the blind spots.

[00:23:27] Koen Pauwels: And I’m like. Absolutely not. I mean, jet CPT five is worse than Jet Gcpt four, for instance. So we have hit diminishing returns, in my opinion, in what, in how much, uh, you know, AI can really improve. And so I think we will continue to need to add, uh, human touches to it. Uh, as I said in the conjoint analysis, adding human respondents, uh, really helped to, uh, to anchor AI to others.

[00:23:52] Koen Pauwels: Uh. To what is much more reasonable. And then of course, human judgment is absolutely key. I would say to the extent that it’s [00:24:00] still human consumers making the decisions. Of course, we have this whole area of, of, uh, of machines, uh, making, buying decisions. And I think in, in marketing, we’re typically not educated to think about that one.

[00:24:12] Koen Pauwels: Uh, and so for, you know, ai, uh, influencing ai, I think that’s pretty straightforward. But I think we definitely have to combine with human data to, uh, to get more appropriate, uh, offers decisions for, for human consumers. 

[00:24:25] Niels Schillewaert: Yeah, I love to hear that. Uh, uh, that’s where the power is, right? I remember when. You know, Kasparov was beaten by, by big Blue, you know, but then if you would combine Big Blue with Kasparov, it would, you know, be even stronger.

[00:24:42] Niels Schillewaert: So that’s really where the power lies. It is. 

[00:24:44] Koen Pauwels: And, and Gary was actually here last year at Northeastern, and he, he, he always, uh, markets himself as, I’m the first person who lost his job to ai. Uh, but then he talks about how he used it to also help train the next generation. In a [00:25:00] combination of, of his human instinct and, uh, and a high, uh, insights.

[00:25:03] Koen Pauwels: Yeah. 

[00:25:04] Niels Schillewaert: Yeah. Yeah. Good, good. Was there anything, I know it’s a buzzword, but it’s, it’s something that people often encounter, uh, retrieval, augmented generation, rag, uh, was there anything to mention there in that, in that perspective? In, at the conference? 

[00:25:18] Koen Pauwels: I, I would say for this audience, and you can read up on it.

[00:25:21] Koen Pauwels: So it, it came out as something that was, uh, that, you know, it’s very popular and, and really helps to, uh, to kind of, uh, iron out some of the caveats and the issues that we talked about. 

[00:25:33] Niels Schillewaert: Okay. Good. Well, people can reach out to you then on that topic. I, I wanted to mention it ’cause it’s something that, you know, often comes up as well.

[00:25:40] Niels Schillewaert: Um. Rounding off last, uh, lessons around, you know, practical, practical implementation. I know you have a framework based on your research and, and your experience that outlines like key steps of how. You know, companies should implement digital twins and digital data. Um, can you explain a little bit and, um, and, [00:26:00] and discuss in a nutshell what it’s about?

[00:26:02] Koen Pauwels: Yeah, of course. Yes. So, so, so first of all, like in any traditional kind of marketing decision, I start with the goals, right? So if you’re an innovate. Is this getting ideas for a new product? Is it, um, you know, aiming to come up with, with the right positioning, the points of parity, points of difference that I still teach?

[00:26:20] Koen Pauwels: Or is it really to kind of get, uh, input on the best marketing mix to launch it? Um, and then, so based on that one, kind of based on the goal, how good of, you know, quality data do you really want and need, uh, to make that decision? Uh, so my framework then goes to contextual appropriateness. So, uh, so the context of what you’re gonna do, does that align with, uh, with the strengths of, of synthetic approaches?

[00:26:45] Koen Pauwels: Uh, so as we said, digital twins are much more rational and tech friendly than, than actual consumers are most actual consumers. Um, and so you’ll have to kind of then, you know, understand what the likely biases are to correct for. And then of course the [00:27:00] validation strategies. So do we have, you know, clear, uh, metrics for success?

[00:27:04] Koen Pauwels: I’m, I’m big on metrics and so you know, what has to happen for, you know, our. Our, our big endeavor to succeed. And, and how will we know that the synthetic data answer is good enough? What is the benchmark really that we’re gonna use? Uh, and then of course, uh, as we talked, uh, earlier, is how are you gonna augment that, uh, with actual human insights, uh, maybe bringing some your own data.

[00:27:28] Koen Pauwels: Go out on the street and talk to some potential customers, which is still extremely valuable. Uh, and then of course, uh, you know, thinking about not just the decision you’re gonna make, but over time, improving decision making. Uh, how will the insights that you get and the comparison really, uh, help you improve the process over time?

[00:27:47] Koen Pauwels: And, and how will you feed back into learning about when to use one versus the other one? 

[00:27:53] Niels Schillewaert: Fantastic. Um, learning as we go. It’s a new technology, it’s one on steroids, but, um, I, I like your, I [00:28:00] like your framework and I’m sure that people can reach out to you if they wanna know more about it. Um, that’s, uh, for sure.

[00:28:05] Niels Schillewaert: Also something the MRII can, uh, can facilitate Please before we wrap up, ’cause we touched on a lot of information here. Yeah. Um, in your opinion, what’s the next big thing? What’s the next frontier for synthetic data? Um, any new news cases you see that are coming up? 

[00:28:23] Koen Pauwels: So, um, I mean, I’m very excited for the follow ups in marketing decision making, so I’m just very excited for, uh, you know, leveraging it, uh, you know, even with the current technology to really kind of be much faster in our creativity, in our decision making.

[00:28:40] Koen Pauwels: Creativity is such an interesting point to me. A lot of people are afraid, right? That we will see all these bland ai slop generated things. But I think AI has so much promise to improve our creativity. I mean, back when I took a creativity class in our, in our Belgium, right, uh, we were actually taught to brainstorm like a hundred ideas and [00:29:00] then come up with what was common and all these hundred ideas, and our next idea should not have that common factor.

[00:29:05] Koen Pauwels: Right. So the example was like a, how do you harvest apples in the, in CIN in Belgium? And all of our ideas had to do with going up to the apple and grabbing it. So then the professor was like, well, can you come up with something that doesn’t have that? And then you come up with things like shaking the trees or, you know, uh, basically genetically engineering the, they fall down perfectly by themselves in the basket and so forth.

[00:29:29] Koen Pauwels: And so I, I think AI is just very good. Kind of summarizing everything which you did in your life and saying, Hey. They all have these things in common. Are you happy with that or do you want to break the pattern? And so I, I have not seen it being used very effectively in that way, but I think it can really helps us realize where we are kind of stuck in our own thinking and, and, you know, hopefully make our human creativity shine more.

[00:29:54] Niels Schillewaert: Yeah. Very interesting. And it also goes back to what you referenced to earlier to be, to [00:30:00] make sure it plays the role of being critical about yourself. Yeah. Basically it gets you to the added value much quicker than before. Um, so we’re not tired by the time we have to, uh, you know, make an added value decision if you want.

[00:30:12] Koen Pauwels: That’s a great way to summarize it. Yeah. 

[00:30:14] Niels Schillewaert: Yeah, yeah. Koen, we’re at, we’re pretty much at the end and before closing, if you would have to leave the audience of marketing insights professional with professionals with one. Piece of advice about synthetic data. What is it? 

[00:30:27] Koen Pauwels: So I would say try it out an experiment.

[00:30:29] Koen Pauwels: Uh, don’t get stuck to just one approach or maybe one vendor. Um, I did that mistake when, uh, you know, after Chad CPT was launched. I tried it out right away because of course it had huge implication for teaching and research at our university. Uh, but you know, I got quickly disappointed and kind of dropped it for a few months until I got a much better solution that I have been using very happily.

[00:30:52] Koen Pauwels: So I would say don’t be. Too quickly discouraged, kind of, you know, try, try out different approaches, uh, see if, if something [00:31:00] fits you better and helps you to make a better decision. And so in that sense, it’s no different from any kind of marketing tool or market research tool that you have basically incorporated or, you know, maybe rejected after a while in your career.

[00:31:14] Niels Schillewaert: Fantastic. Thank you so much, Koen, for sharing all these learnings and, uh, it’s good to hear that AI is here to help. Uh, we have to be critical and, and purposeful. Um, so thanks again, um, to everyone. Synthetic data is not going away. It’s one of the big topics of the conversation in our industry. Uh, but Koen couldn’t help to separate the hype from the facts.

[00:31:35] Niels Schillewaert: Uh, so thanks again. Thanks to everyone for tuning into the Insights and Innovators. See you next time. 

[00:31:42] MRII Announcer: Thanks for joining the Insights and Innovators podcast for Market Research Institute International. Click subscribe to never miss an episode and visit us@rii.org for more market research insights.