Walk into any AI meeting and you’ll hear the same buzzwords
Synthetic data. Digital twins. AI-simulated respondents. Virtual consumers. Synthetic personas.
In pitch decks, conference panels and procurement meetings, these terms get thrown around as if they mean the same thing. They don’t – and that matters.
It’s not ‘just semantics’ – the lack of clear language is actually a business problem
Take “digital twin” – a term that comes from engineering. It’s a high-fidelity computational model of a physical system continuously updated with real-world sensor data. It’s a digital copy of a real world system, built to test scenarios, practice or test…just like a Formula One car or Flight simulator. When applied directly to a human beings, it implies something equally rigorous: a dynamic, data-rich model of an individual’s behaviour, calibrated against real behavioural signals.
But in market research circles, the same term often gets slapped onto something far simpler – a statistically profiled archetype built from survey data and demographic segmentation. While this tool has real potential value in research, it is a fundamentally different thing. Treating them as equivalent – giving them the same label – sets unrealistic expectations and hides the very difference that’s critical to determine fitness for purpose and use cases for application. It also confuses or at worst massively obfuscates the validation requirements required.
The problem is massively amplified by market dynamics and incentives. Various suppliers competing for airtime and budgets have every reason to use the most impressive-sounding language – often at the expense of accuracy, which is deeply ironic for an industry predicated on accuracy. Buyers under pressure to demonstrate innovation and leadership within their organisation also often tend to reward ambition – who isn’t to some degree attracted by ‘shiny new’?
The result? A linguistic muddle and a ill-defined playing field with few, if any, established rules.
But what should we actually be talking about?
Synthetic research is the new frontier for the research and insights industry. It offers a transformational opportunity to extend the toolkit we have to offer genuine value; to get evidence into decision-making where it wasn’t possible previously due to feasibility, speed or costs.
It also carries real, useful capabilities: the ability to generate scalable, privacy-safe research assets, to stress-test hypotheses before committing fieldwork budgets, and to simulate population behaviour under hard-to-recruit conditions.
But those capabilities – like all research techniques – come with some limitations, and those limitations vary enormously depending on the specific approach being used. At minimum, practitioners need to distinguish between:
- Generative data augmentation: statistically extending real datasets
- AI-simulated responses: using language models to predict or emulate how a profiled group or persona might respond
- Behavioural modelling: building predictive models of decision–making from observed behavioral data
Each of these has distinct use cases, data requirements, and validation thresholds. Conflating them is like calling a spreadsheet and a neural network both “computers”; technically defensible, practically useless.
What’s next?
Synthetic research represents a genuine methodological frontier, but its value will only be realized if the field establishes the rigour it deserves. The technology is not the limiting factor; definitional clarity is.
Until researchers, suppliers and commissioners share a common taxonomy for what these approaches are, how they work, and where their limitations lie, robust evaluation and appropriate application will remain out of reach.
But there’s reason to be optimistic…cautiously. The sector hasn’t yet made peace with imprecision: there are many of us who are already uncomfortable with vague terminology. The task now is to channel that discomfort into actual standards, norms and an agreed taxonomy for synthetic research capabilities.
Explore synthetic research with Virtual Personas by Savanta
Virtual Personas by Savanta combines real-world research with the power of AI to deliver affordable, reliable consumer insights you can act on immediately.
Ready to give it a go? Sign up to the platform for free today, or reach out to our team if you have specific questions or would like a demo.