Product benefits
How it works
STEP 1
We build your audience
STEP 2
You ask the questions
STEP 3
You get answers
* As one client put it: "It takes longer to write the question than it does to get the answer"
STEP 4
Dig deeper
Former Chief Data Advisor to the Prime Minister
What are synthetic audiences?
Synthetic audiences are large-scale simulations of real people. We build them from your customer data or high-quality real-world data. The simulations are grounded in social science research and validated against real-world benchmarks. Query them directly through our platform and get answers to unanswered questions in seconds.
How accurate are they?
As accurate as traditional research methods. Our synthetic audiences match what real people say 96% of the time and have been independently validated by Professor Michael Muthukrishna at LSE. We measure this through hold-out testing: we generate synthetic responses, then compare them against real survey data the model has never seen. We have run tens of thousands of these evaluations. Every new audience is benchmarked this way before it goes live, and we communicate these benchmarks directly so you know exactly where the model is strong and where it has limits.
How do synthetic audiences work?
Human behaviour is shaped by a combination of individual circumstances, cultural background, and environment. Traditional research accounts for this by weighting survey samples by age, gender, and region - these are proxies for the factors that actually drive how people think and act. LLMs are trained on an enormous volume of human-generated text. That makes them, in effect, a model of human beliefs and behaviours across billions of people - the most comprehensive representation of collective human thought ever assembled. The problem is that this representation is averaged and opaque. When ChatGPT answers a question, it draws on everything, everywhere, all at once. It has no specific audience in mind and builds generic personas without real-world grounding. Electric Twin solves that problem. We use your real customer data - surveys, focus groups, and customer interviews - combined with machine learning to constrain and direct that broad human representation toward your specific audience. The result is a synthetic population that thinks, responds, and behaves like your actual customers.
How do you build a reliable model from limited data?
This is one of the first questions we work through with every client, and the answer is usually straightforward: if you have data on your audience, we can almost always build a reliable model from it. Your data anchors who the audience is — how they behave, what they think, who they are. The wider context comes from the models underneath, which carry a vast body of human knowledge, and the processes by which we build personas. Together they produce responses that hold up against real-world testing. The first conversation with our team is about what data you have, what shape it’s in and what we can build from it. We'll tell you upfront where the model will be reliable and where its limits are, so you know exactly what you're working with from day one.
Read all FAQs