Come Fly with Me! Why Simulations Leave Synthetic Behind

We’ve been talking about simulations in the context of insights for over three years. We’re often asked what we mean by simulations, how they differ from synthetic data, and why the distinction matters when it comes to making better decisions. So, let’s break it down.

Four modes of insight

We can group today’s research and insight approaches into four boxes.

REPRESENTATION —Traditional Research  

Representation is the goal of traditional research: sampling, quotas, and method design all work to mirror the population. This foundation delivers trustworthy segmentations, surveys, qualitative insight, and personas rooted in lived experience. But these outputs are static, slow to update, siloed, and limited in scale - out of step with how quickly customer expectations now shift. And layered on top is a well-documented respondent quality crisis.

Representation remains the most trusted and low-risk foundation for decisionmaking, but its value is increasingly limited as traditional research struggles to keep pace with today’s rapid decision cycles.

IMITATION — Synthetic Data, Synthetic Personas & GPTs

Imitation tools - including synthetic datasets, synthetic personas, and GPT-style generative models - generate artificial versions of people or markets, producing fast, scalable outputs. But as Ipsos warns, they can be “plausible yet completely wrong,” reproducing patterns in existing data rather than the complexity of human behaviour. They’re helpful for early drafting or hypothesis development, but risky for real decisions: most synthetic systems aren’t validated, lack cultural grounding, and can’t reliably confirm truth.

That makes them high-risk and low-value for strategic use.

REPLICATION — Digital Twins

Digital twins aim for precision: detailed, data-rich reconstructions of individuals, systems, or markets. They’re excellent for modelling operational flows or forecasting where inputs and rules remain stable - Google’s Digital Twin guidance reinforces this. Their drawback when applied to market research is rigidity. They’re costly, slow to evolve, and unable to absorb cultural change, emotional nuance, or the day-to-day variability of human behaviour. Meaning they can never be fully rounded or reliably up to date.

They replicate what is, but struggle with what could be, limiting their value in an industry where uncertainty and imperfect data increasingly define decisionmaking.

GENERATION — Simulations  

Simulations sit in the generation space - where insight doesn’t just mirror the world but creates new understanding. A simulation is a dynamic, adaptive system that blends validated human data, cultural context, and machine reasoning. It doesn’t imitate people or replicate systems; it performs like reality, evolving as conditions shift. Recent Stanford research shows that simulations grounded in human interviews can replicate behaviour with 85% human-level accuracy - far beyond synthetic shortcuts.

By combining validated inputs, traceable logic, continual updates, and human oversight, our simulations remain transparent, auditable, and fit for decision making. Built to behave like the real world, simulations allow teams to safely test ideas, explore consequences, and see decisions play out in context. And because they evolve with new evidence, they stay aligned with changing expectations and emerging behaviours. The result: uniquely low-risk, high-value intelligence for decisions that matter.

Think of a flight simulator. A pilot doesn’t just read a dashboard, they step into an environment that behaves like the aircraft, practising manoeuvres before taking off. Simulations work the same way: not dashboards or static models, but training cockpits for decision-making. You can try strategies, see outcomes unfold, and rehearse the future before it arrives.

What this means for the industry

Traditional research tells us what was. Most AI tools simply speed up what we already do. But evidence-based decision-making is failing, largely because insight still isn’t embedded where decisions actually happen. Simulations create something different: a decision environment that behaves like the real world.

Insight stops being retrospective and becomes an active tool for shaping what comes next.

This shift is needed now. ESOMAR’s 2025 Insight Activation work shows fewer than half of business decisions today are evidence-based, largely because insight remains static; trapped in decks, dashboards, and disconnected studies rather than embedded in the workflows where decisions happen.

Simulations correct this. Instead of producing knowledge that decays on delivery, they embed intelligence inside the environments where teams think, plan, and act. They enable continuous learning, rapid experimentation, and cumulative knowledge, a living system rather than isolated projects.

For researchers, the role expands: from reporting findings to designing, calibrating, and governing intelligent systems. And for the industry, expectations rise - insight shouldn’t just describe the world, it should help teams understand how it behaves and how their choices influence it.

Simulations move us from explaining the past to rehearsing the future, a change with the potential to redefine the purpose, value, and impact of insight altogether.

And we think that’s incredibly exciting