What’s the big deal with synthetic data?

For instance, based on surveying 1,000 real people from a population, algorithms can create additional data that looks and behaves like it came from actual consumers, and even gets to increased statistical precision within sub-groups of the population.

Another form of synthetic data is digital twins, where a model is trained on actual attitudes and behaviors from a real respondent and then used for subsequent predictions and insights – with permissions, privacy and regulatory safeguards built in.

Now, take it down another level – explain that again, but to a non-marketing person

Take the first use case above. Imagine a household brand wanted to better understand what a group of people thinks about a new product or service. They’ve already conducted a survey with 800 adults. Now, say they want to analyze responses from men under the age of 25 but only have 20 responses in this group. This would not be statistically valid.

So, rather than spending more money and waiting for more time for these people to answer, synthetic data can be used to make an educated guess on how this group may respond, based on analysis of response patterns for the 800 people for whom they do have reliable data. It’s all about predicting responses, based on real data patterns, to make faster decisions.

Why should we care about this now?

The world moves quickly. CMOs and insights teams need to move at lightning speed to innovate and stay competitive. But getting real, high-quality data from enough people can be slow, expensive and sometimes impossible for niche audiences. Additionally, with increasing privacy regulations and consumer concerns about data collection, synthetic data offers a way to speed up research and get richer insights without needing to collect more personal information.

AI and machine learning advancements, especially generative AI, have dramatically improved the quality and capabilities of synthetic data generation. Forbes predicts that synthetic data could become a $2.34bn industry by 2030, making it an essential tool for staying competitive in a data-driven landscape.

What’s the most exciting thing happening in this space?

When used correctly, synthetic data accelerates our understanding of human behavior, especially for underrepresented audiences. That’s really important in today’s world.

Organizations have also started to create digital twins of customers. Virtual models such as these can be used to predict how customers are likely to react to new products and campaigns, for example, and it has the potential to revolutionize how brands create ideas and tailor strategies. The real excitement lies in fine-tuning these tools using high-quality, specific data sets that make predictions more accurate.

What should marketers avoid?

That synthetic data doesn’t need human data for training, or that off-the-shelf large language models (LLMs) can answer any question autonomously. The market research industry influences decisions worth billions of dollars, for brands, governments, the media – you name it.

For synthetic data to be truly useful and accurate, it requires a solid foundation of high-quality, relevant and recent human data to train the algorithms and then a robust framework to validate them. Relying solely on generic pre-trained models will lead to biased, stereotypical and ultimately unreliable insights. Fundamentally, synthetic data is an extension of human data – not a replacement.

How does it work in practice?

Imagine a brand tracking survey where you have limited responses from a crucial but hard-to-reach demographic – something like Gen Z buyers of luxury cars in Germany, say. Instead of spending time and money trying to recruit more of these specific individuals, we use our high-quality, fraud-proof human panel data to train a synthetic data model. This model then generates new rows of synthetic respondents that accurately reflect the characteristics and preferences of that under-represented subgroup. This boosts the statistical power for that audience segment, allowing marketers to make more confident decisions about how to target and engage them, without compromising the overall sample quality.

What’s the risk of sitting this one out?

There is a risk of falling behind competitors who are already making the most of these capabilities to gain deeper, faster insights. You might miss opportunities to understand niche audiences, optimize campaigns or develop new products by limiting yourself to traditional data collection methods.

Of course, using synthetic data incorrectly carries significant risks. This includes using misleading insights and making poor decisions based on information that is generated from biased or irrelevant training data or poorly trained models. There would be erosion of trust if insights are consistently off and wasted resources if solutions are implemented without understanding their limitations. So beware.

What can marketers do now to level up?

1. Educate yourself and your team: Start by understanding the basics, common use cases and the critical questions to ask. Like any new technology, you need to know what you’re talking about and not every use of synthetic data is going to be right for you – so encourage your team to read, listen and engage with experts to understand what’s best for your business.

2. Focus on your data quality: Before even thinking about synthetic data, ensure your existing human data sources are held to the highest quality standards and are representative, diverse and fraud-free. Synthetic data is only as good as the real data it learns from.

3. Run a pilot with a trusted partner: Don’t jump in head-first. Identify a specific, manageable problem where synthetic data could offer a solution, such as boosting a small subgroup of people in a low-risk survey. And then work with a reputable partner who has strong foundational data and rigorous validation processes to run a controlled pilot.

Which partners, platforms or technologies should be on our radar?

There are specialized synthetic data providers, companies that specialize in generating synthetic data, particularly those with strong validation frameworks and expertise in specific industry verticals.

Kantar’s own approach is built on the foundation of a high-quality panel of data, as well as our AI Lab that is responsible for research and development. In addition, our Qubed anti-fraud technology and various AI tools ensure we are always making informed decisions and keeping information safe.

We’re also doing a lot of R&D on digital twins, and so is the wider industry. Technology companies such as Microsoft are also building software libraries and agent-based frameworks that are designed to simulate how humans might make decisions and interact with one another. These haven’t necessarily been customized for market research applications yet, but given the pace at which the technologies are being built, they probably will be soon.

Advice for a marketer starting out with this?

Start with a clear problem to solve instead of just generating synthetic data because it’s new and exciting. Ask yourself, ‘What specific insight am I trying to address?’ Is it about reaching more people, understanding specific segments better, or testing scenarios that would be impractical with traditional research? Then, ensure you have high-quality human data as your foundation before exploring synthetic options.

Synthetic data is a tool to extend and enhance human insight, not replace it. Start small, with controlled experiments where you can validate synthetic against human responses before scaling up to more critical business decisions.

For more explainers on all things data, visit the Data for Drummies content hub.