Synthetic Populations & Their Impacts on Public Relations

What if the best market research started by IGNORING what people say?

What if, instead, you started modeling them based on proven behaviour?

Right down to the regional level?

And what if political polling was done this way too 

A new term is showing up in research and strategy circles with major implications for communicators: synthetic populations. This is not a cheap AI focus group. It is a data-built population model that reflects how people are distributed and behave at scale using high-quality inputs like official statistics, mobility patterns, and registration data, rather than relying only on interviews and surveys.

That matters because self-reported data is often aspirational, incomplete, or socially filtered. Synthetic populations offer another path: estimating market potential, testing where campaigns should start, understanding regional differences, and pressure-testing assumptions before rollout. The real question is not just what synthetic populations are, but what happens when strategy shifts from asking people to modeling populations.

Listen For

3:07 What’s the difference between a synthetic panel and a synthetic population?
5:13 How can a synthetic population be realistic without using real individuals?
8:49 Why do surveys over-claim luxury brands? And how does official data correct it?
11:58 What did Germany’s flat-rate transit ticket reveal about commuting by region?
14:15 Could synthetic populations change how political polling is done?

Guest: Eike Hartmann, Vice President Custom Research & Insights Business at Statista+

Website | LinkedIn

White Paper on Synthetic Populations

Doug

Substack | Website | LinkedIn

Farzana

Substack | Website | LinkedIn

Are you a brand with a podcast that needs support? Book a meeting with Doug Downs to talk about it.

Apply to be a guest on the podcast

Connect with us

LinkedIn | X | Instagram | You Tube | Facebook | Threads | Bluesky | Pinterest

Request a transcript of this episode

Transcript

Emily Page (00:00):
What people say can be useful, but it's not always true. What people do leaves a pattern. Smart strategy learns to read the pattern.

Doug Downs (00:21):
The year was 1854 and London was hot that summer. The streets were crowded. People were dying fast. They said it was bad air. They said it was the smell. They said cholera moved in the fog. John Snow was a doctor and he wasn't buying any of that explanation. He walked the streets. He looked at the houses. He counted the dead and he marked them on a map. One mark and then another, and then many more. The marks gathered around one place, a water pump on Broad Street. Snow didn't need a speech. He didn't need a theory that sounded good in a room. He looked at the pattern. He looked at what people were doing, where they were going, and what they were using. And then he took the handle off the pump. That's the famous part. But the real story is this. He found the truth because he stopped asking what people believed and started tracing what they did.

(01:21):
 Sometimes the breakthrough is not the answer people give. Sometimes it's in the pattern they leave behind. Today on Stories and Strategies, we take the handle off bad assumptions and ask what happens when market research stops asking people and starts modelling populations.

(01:57):
 My name is Doug Downs.

Farzana Baduel (01:59):
And my name is Farzana Baduel and our guest this week is Eike Hartmann, who is joining us today from Hamburg. Hi, Eike.

Eike Hartmann (02:07):
Hi.

Emily Page (02:08):
How are things in Hamburg?

Eike Hartmann (02:10):
Quite good. Spring just started this week.

Emily Page (02:13):
Oh, it's such a beautiful city. Now, Eike, you are vice president of custom research and insights business at Statista, helping organisations turn complex data into practical insight for strategy, market understanding, and also decision making. And you work at the intersection of research, analytics, and innovation, including emerging approaches like synthetic populations for more granular audience and market modelling.

Doug Downs (02:42):
Eike, I loved ... I think it was a newsletter or a social media post where Statista mentioned synthetic populations. And right away, my little brain goes, bing, bing, bing. Never seen that before. Never seen that before. So what are synthetic populations? And is this just synthetic personas, synthetic panels? Is this just AI generated respondents? What is this?

Eike Hartmann (03:07):
Yeah, so there's different terms being used and there's the same term used for different things. So let me try to clarify things. So I think what is pushed really hard in the market research industry is synthetic panels, which is basically you have a panel maybe of a couple of tens of thousands of people. Those are virtual people, so digital twins of actual people that have been trained based on the answering patterns in quantitative or qualitative market research studies using machine learning and AI, and then enabling those digital twins to answer new questions, which is of course providing a lot of benefits if it's much more cost efficient and quicker if the quality is good. So that's the key question that a lot of people are asking. And the term synthetic population is also used for that. But when we talk about a synthetic population, we mean something different because our synthetic population does not only cover a sample of a couple of thousand or tens of thousand people, but the entire population.

(04:21):
 We have Germany as the pilot market. So our synthetic population is a digital twin of 80 million people with a lot of information about those people that we can leverage in market research projects.

Emily Page (04:37):
So let me get this straight. So each member of the synthetic population is in essence a digital twin of an actual individual, a living person that you have previously engaged with and therefore what happens about consent?

Eike Hartmann (04:56):
Yeah, so that could have been impossible. So that's what's possible with a panel that you just replicate and put it into a digital twin. So our synthetic population of the entire population is not based on actual people,

(05:13):
 But all the people have realistic patterns and our panel, or not our panel, our synthetic population is based on public statistics like census, other official statistics that we have disaggregated. And as said, each person is realistic and the entire population meets the statistical patterns of the entire population. So even if you drill down into a narrow subgroup, men between 50 and 60 in one county that drive BMWs, we aim to have the right amount of people and also have all the other data features relevant for that subgroup being as close to reality as possible, but it's modelled in the end. So it's not based on actual people because otherwise you would have asked 80 million people and that's impossible.

Emily Page (06:14):
Yeah. That's so fascinating. So I've got a couple of questions about the financial impacts. So does that mean that it's going to become easier for other organisations who normally don't have the budget to embark upon these huge focus groups? Is it going to be more accessible now, the ability to test concepts, which would normally for the big FMCG and the big corporate? So is it something that's going to really democratise research for smaller organisations with lesser budgets?

Eike Hartmann (06:46):
I think so. So hopefully for the industry, there's also more questions being asked, otherwise we run into some issues and our synthetic population at the moment is static. We are aiming to bring it to life soon. So you can't interact, you can't ask the population any questions that's not inherited into the dataset at the moment. But on the other hand, the inherited dataset has a very high quality. So it meets our very high quality standards and the data that it's incorporated is almost true, as close to reality as possible, but making our synthetic population, bringing it to life would be the next obvious steps because then you don't have a panel of a couple of thousand people, but the entire population you could interact with.

Doug Downs (07:48):
So scalability, but also what I've always found with surveys is you ask people, "Well, how many days a week do you exercise?" "Oops, I exercise four times a week." And we know behaviourally, no, they don't. No, they don't. Gyms, I suppose if you did a study of gyms and the population compared to how many club members they had, you could prove that. That's part of what you're able to do with the synthetic population, right? It's separating the, "Yeah, I work out four times a week" from the ... I'm there once a month maybe.

Eike Hartmann (08:21):
Yeah, that's a perfect example. And we combine different sources and each source has its limitations, including even official statistics. So we found some interesting data even in the German census with three year olds living alone in single households, et cetera.

Emily Page (08:46):
God, you Germans are really advanced.

Eike Hartmann (08:49):
That's right. That's right. And of course, if you ask people stuff, there might be a bias, there might be interesting patterns in how they answer questions. So one of my favourite examples of data that we already incorporated in the dataset is which car people own. So if you run a huge survey, even with high quality standards, you usually have a bias towards luxury cars. So people state that they drive a BMW, and Mercedes, et cetera, and less luxury cars like Chinese cars or Eastern European brands are slightly underrepresented. So what we do is we take the official car registration statistics of the German car registration authority, so we know how many BMWs there are in a county, and then we use survey data to distribute those cars among the right people. But there might be that men drive some cars more than women or older people have a preference for specific brands.

(09:58):
 So we think that those relative perspectives are true, but the absolute numbers are not true. And combining those two sources makes it even better.

Doug Downs (10:12):
So for Statista, I don't know if you're beta testing or test running it in Germany, what are the future plans? When does someone in the US get to try out a synthetic population through Statista?

Eike Hartmann (10:25):
Yeah. So US is second on our roadmap. So we started in Germany also based on a project where we've been able to develop the initial dataset, but US is the second country that we will roll out the synthetic population later this year. So that's the plan.

Doug Downs (10:48):
And would you do it market by market, like starting California and then New York or just blanket it?

Eike Hartmann (10:54):
I think we're going to do the entire country. So for Germany, we did the entire country. You need to have good data. So I think the US census is a great asset that we can build on, combine it with proprietary survey data that we have other sources pretty similar as we have done it for Germany. But I think it adds a lot of value if you are able of covering the entire country at once. But it's a lot of work. So that's why we are not launching a new country every week because our quality standards are very high. And it's a lot of work because all the data is ... Each data point is correlated to all the other data points. So complexity increases for each data point that you're adding because how old people are, what they earn is correlated to what car they drive, what values they have, et cetera, and we are aimed to get those correlations right.

(11:58):
 I think the main benefit of our synthetic population, which incorporates the entire population, is also regional granularity. So for each topic where regional granularity is important, even if you have 60,000 people in a synthetic panel, you have two people that live in one city, and we have basically all the people. So in this project that we have run for the German Rail Authority, regional granularity was of the essence because we analysed the commuting behaviour also to impact public policy here and they wanted to know how the ... So there was a flat rate public transport ticket introduced in Germany two years ago. So for about 50 bucks, it allows you to run on every local train and bus all across Germany. And the question was, how did the introduction of that ticket impact commuting behaviour and commuting cost on a local or regional level?

(13:07):
 And that was the question. And of course you could have run a huge survey study asking 500 people in every county, but then that's basically impossible to pay.

Doug Downs (13:18):
Yeah. So with the regional granularity that you're able to offer or the targeting, I got to think that there's some kind of application here to replace the current political polling that we see out there, which amazingly just never seems accurate, not precisely anyway. Is there an application for that, do you think?

Eike Hartmann (13:40):
So I think the only answer would be I don't know. So we haven't tested it, but I think if you get 90% accuracy of data, and if you train a model and maybe include some additional sources like what is the sentiment out there, what are historical patterns, how sentiment influenced actual voting behaviour, et cetera, I did train a model on that might be as good as actual polling from my perspective.

Doug Downs (14:12):
Eike, thanks so much for your time today. Thank you very much. It was a pleasure.

Farzana Baduel (14:17):
So here are the top three things we got today from Eike Hartmann. Number one, the use of synthetic panels and populations will bring in increased speed, accuracy, and lower costs for research. Number two, he distinguished the difference between synthetic populations and synthetic panels and with data on populations, it gives much greater accuracy, specifically when you're looking at different regions and being able to drill down to specific towns and cities. Number three, we are not in Kansas anymore. We are talking about disruption with a capital D. We are entering into uncharted waters and who knows how this application is going to play out with political campaigns, with the rollout of global FMCG new products? It's going to be fascinating, so buckle up.

Doug Downs (15:18):
It feels like we were just introduced to something that three years from now, like everybody knows about it and we're all using it. It's kind of like when GPT came out.

Farzana Baduel (15:27):
Yeah. Yeah, absolutely. It's just, I mean, even the phrase like synthetic populations, it's like God, we're being replaced even in research.

Doug Downs (15:37):
And our behaviour monitored closely. If you'd like to send a message to our guest, Eike Hartmann, we've got his contact information in the show notes. Stories and Strategies, co-production of Kurzen Public Relations and Stories and Strategies Podcast. If you liked this episode, please leave us a rating, a real one, not a synthetic one, possibly a review. Thank you to producers and they're real, Emily Page and David Olajide. And lastly, if you're listening to this episode and you thought of a friend of yours at some point in the conversation, they just kind of popped into your mind. Before you go, would you forward this episode to them? Chances are they'll love it too. Thanks for listening.