Synthetic audiences are fast becoming marketing’s new secret weapon: AI-built stand-ins for real consumers that can react, respond and “think” on demand. From slashing the cost of focus groups to testing creative territories overnight, FutureWeek caught up with four industry leaders to find out how marketers are using synthetic audiences right now, and the benefits they can (and can’t) bring.
Chris Woodward, Executive Director, OLIVER
“The use of synthetic personas and focus groups has been a standard part of how we work with our clients since we launched our proprietary platform earlier this year. In simple terms, our platform collects multiple datasets to build the profile of any target audience we want to conduct research against. We draw on multiple datasets (which typically includes CRM data, online behaviour data, bespoke research insights the client might already hold, product reviews and the like). Our platform then trains on this data using machine learning algorithms – such as generative modelling and cluster analysis.
“From this, we create synthetic personas. We can then begin conducting research using these personas. This approach typically costs around 20 percent the price of conventional focus groups and also takes a fraction of the time. Is it as good as a more traditional approach? No, it’s not. It’s about 70-85 percent as good, but the reality is that it often enables us to get a sense of whether a hypothesis is sound and therefore warrants further investigation. What’s more, given the cost and time saving it provides, it’s deeply attractive to clients, especially when conducting research with hard-to-reach audiences.”
Cynthia Vega, Global Analytics & AI Director, Kantar
“With AI agents and large language models now capable of producing human-like responses, it’s easy to mistake synthetic audiences for real people, but they’re not. At the end of the day, it’s data generating data. A synthetic audience is an artificially created group that mimics how humans are likely to behave. In insights generation, these audiences can give individual or group-like answers to new questions, allowing brands to explore new scenarios with confidence, more quickly and in real time. They’re built (ideally) on a baseline of high-quality behavioural and survey data from real people. This provides the AI with a deep understanding of a persona and its preferences, allowing it to infer behaviours.
“We’re using synthetic audiences to help brands bring segmentations to life and shorten the innovation build process, giving instant conversational access to around 7 million digital twins. The answers are validated at scale against real consumers, our ‘ground truth’. Unlike real-life audiences, synthetic audiences can also provide ‘slow thinking’ responses, or more thought-through answers and they allow us to manage desirability bias. A key advantage is access, as brands can explore new scenarios at all times. Of course, there are still limitations. These include the risk of hallucinations, challenges around diverse representation, and the uncertainty around how much commercial confidence advertisers can place in this data today.”
Tom Smith, CEO, GWI
“Synthetic audiences are AI-built stand-ins for consumer groups, acting as a shortcut to understanding audiences without endless surveys. But here’s the catch: not all synthetic audiences are created equal. They’re only as smart as the data they’re trained on. If an AI persona is trained on poor, outdated or even synthetic data, they’re more likely to give generic insights, miss cultural nuance, and reinforce biases. You can even risk feedback loops that reinforce the persona’s own errors if it’s used in isolation.
“But what about the human that’s using AI? That person’s role is critical to catch AI missteps, but when you’re working with a synthetic persona you need to ensure humanity is in its DNA. Without that layer of human truth built into their very foundation, the use of synthetic personas risks a marketing world where everything sounds the same. In other words: content built for robots, not people.”
Dr. Ben Warner, Co-founder, Electric Twin
“I worked in government during COVID and we were trying to understand behaviour then and build policies to try and mitigate the harms, but we couldn’t do it because we didn’t have that understanding. We could have done this with something like Electric Twin. Focus groups and surveys are very valuable tools, but they’re very expensive and very slow. If you’ve ever done them, you know the feeling of realising what questions you should have asked afterward, and suddenly you have to wait another eight weeks, you have to double the price and have to pay all over again.
“Electric Twin means you can actually ask the next question. Once you’ve asked a question, you can even follow up with a certain group of people in your audience. You can ask, “Do you prefer X or Y?” and then for people who answered X, you can follow up and understand more about why they chose that answer. It enables you to work in the way you want to work, rather than in the way we are forced to work because of how expensive and slow market research studies are.”
