Synthetic data offers speed and efficiency in research, but it cannot fully replicate real human behaviour. While useful for early-stage ideation, synthetic respondents struggle to capture emotional bias, contradiction, and lived experience, making real human insight essential for high-stakes commercial decisions.
Do You Want an Answer, or the Right Answer?
Synthetic data is becoming an established part of the research toolkit. Its speed, scalability, and cost efficiency make it appealing, particularly in early stage thinking and ideation.
But speed does not always equal understanding.
In this article, Will Morgan, Research Director at Spark Emotions, explores the critical differences between synthetic respondents and real human responses, and why behavioural nuance, emotional bias, and contradiction still matter when decisions carry real commercial weight.
“Synthetic data is here to stay. I’m not here to challenge that. However, what I do want to do is remind those that use it, that human psychology clearly demonstrates why this method has limitations.
For those out of the loop, synthetic data is created by utilising real data points to train an AI model, often so a synthetic respondent can ‘reply’ in a way that mirrors a real person, often called a ‘digital twin’.
This makes the tool excellent for early-stage ideation and drafting, narrowing 100’s of ideas down to the 10’s. But making million-pound decisions on these responses should be taken with caution.
As we all know, humans are biased. Our processes and behaviours are not explicit, but often lie behind the conscious understanding of what we do, and why we do it. This is even more true when it comes to the drivers behind our purchases.
Therefore, models trained on explicit responses often lack the ‘pulse’ of the real human experience, which is that we often can’t (accurately) articulate why we do/like the things we do. This means synthetic respondents often over-emphasise the rational claims consumers make about their behaviours, and downplay the irrational. Yes, real human responses can give frustrating, unclear, and imperfect answers, but this often reflects the difficulty in trying to communicate a response to a question about behaviour that they can’t rationalise.
Simply put, synthetic models cannot simulate the emotional bias, the physical exhaustion or the sudden cravings that drive our purchases. It has no lived experience to create a complete view of a multi-step, complex series of behaviours.
Humans are contradictory: We want to be healthy, but we buy sugary foods. We want to exercise but spend our evenings in front of the screen. Asking a synthetic respondent what types of ‘healthy’ products they might buy risks missing this nuance.
Synthetic respondents also often lack a knowledge of the outside world beyond it’s training and datasets. It lives in a retrospective reality that aren’t affected by the unpredictable winds of what is cool and what’s not. So it bases its reasoning on the predictable world it inhabits, not the unpredictable world outside of it.
Not only this, but as we already know, AI tends to seek the middle ground. It rarely takes risks, and therefore smooths out the fringes and outliers in the data. Outliers can often provide us with the trends of the future, the opportunities that market research creates, therefore we risk missing the spark that might lead to bigger ideas.
Synthetic data’s unparalleled speed and costs make it the perfect tool for the ideation and, but when it comes to go/no-go decisions, you need to know not what a consumer might do, but what they will do.”
Summary
Synthetic data has an important role to play in modern research, particularly when speed and scale are required.
But as this article highlights, understanding behaviour means grappling with emotion, contradiction, fatigue, impulse, and context, elements that cannot be fully replicated by models trained on rationalised responses alone.
For Spark Emotions, the distinction is simple. When decisions matter, when investment is real, and when behaviour needs to be predicted rather than assumed, real human insight remains irreplaceable.
Explore more thinking from the Spark Emotions research team or get in touch to discuss how behavioural insight can support confident decision making.
