Why architecture matters in synthetic research
If language is the first problem facing synthetic research, architecture is arguably the more consequential one. The way a synthetic representation is built defines the boundaries of its knowledge and reasoning. Those boundaries aren’t just a technical detail; they directly affect the quality of the decisions made from the insights generated.
This brings us to the distinction that matters most: are you building from the ground up, or from the top down?
Digital twins: built from the bottom up
A genuine digital twin starts with observed behavioural data – transactional records, digital interactions, passive measurement, longitudinal survey participation – and assembles a predictive model calibrated to that real-world evidence. Think of the computational rigour that goes into an F1 car simulation or aircraft flight model: systems that are continuously tested, updated, and challenged against real-world performance data.
For digital twins, validity comes from the quality and breadth of the underlying data and the robustness of the modelling methodology. Crucially, a real digital twin needs to be updatable and testable against new real-world data. If it can’t be challenged and recalibrated, it can’t honestly claim the label – and by extension, neither can the supplier offering it.
Synthetic personas: built from the top down
Synthetic personas work in the opposite direction. They begin with a conceptual framework – a segmentation schema, a jobs-to-be-done structure, or a set of attitudinal typologies – and populate that framework with plausible synthetic attributes.
This type of persona does not represent any specific individual’s behaviour. It’s a coherent, illustrative archetype designed to represent a type of person. This is a well-established and genuinely useful research tool. But it is not – and can never be – a substitute for empirically measuring real human responses. And treating it as such is where the real trouble begins.
Why confusing digital twins and synthetic personas is harmful
This is more than a semantic debate. The commercial confusion between these two approaches creates real, measurable problems:
- Clients commissioning “digital twin” outputs sometimes receive sophisticated personas, and make decisions at a level of confidence the underlying methodology doesn’t support. That’s not just poor research; it’s a governance failure.
- Legitimate persona-based work is often unfairly discredited when judged against validity standards designed for empirical modelling. Both the method and the supplier lose out – and the client loses a genuinely useful tool.
Both approaches have real value. The damage comes from misapplication – when clients apply the wrong expectations to the wrong method. In an industry predicated on rigour, that’s a deeply ironic failure.
There’s an ethical dimension too
Bottom-up digital twin construction relies on substantial personal data, raising meaningful questions that too few conversations in our sector are actually grappling with: consent, data governance, what it means to create a proxy identity for a real person, and who owns that representation.
Top-down synthetic personas are built from non-personal aggregate information. That creates a different – but not absent – set of ethical considerations, particularly around representational bias and the risk of encoding existing inequalities into research archetypes.
What this means in practice
While digital twins and synthetic personas are often conflated, they are fundamentally different; one is bottom-up and empirically modelled, the other is top-down and conceptually constructed. Both have value, but confusing the two approaches creates real risk – applying the wrong expectations to the wrong method can result in measurable damage.
The key is transparency. Suppliers should clearly state which epistemological approach underpins their offering, so buyers can assess how it should be validated and applied. Understanding whether they are commissioning top-down conceptual construction or bottom-up empirical modelling enables buyers to ask better validation questions – and make more informed decisions.
Curious about synthetic research? Try it for yourself.
Virtual Personas by Savanta combines validated behavioural science and decades of real consumer data to deliver reliable audience insight at speed. Try the free version and you’ll gain instant access to eight pre-built personas – so you can run interviews, test ideas and explore your audiences straight away, no commitment needed.
Sign up for free today, or reach out to our team if you have specific questions or would like a demo.