How to Get Real Insights from Synthetic Personas

Across IIEX Europe, one theme kept surfacing across conversations about AI: synthetic personas are moving from novelty to practical research tools. They're helping researchers explore ideas faster, pressure-test concepts earlier, and better understand customer segments before investing in traditional research.

But nearly every speaker also offered the same caution.

Synthetic personas can produce answers that sound thoughtful, nuanced, and remarkably human. That doesn't necessarily make them accurate.

As AI becomes more capable, the challenge is no longer deciding whether synthetic personas have value. It's knowing when to trust them, when to validate them, and when real human research remains essential.

As Adam Bai, Chief Strategy Officer & Chief Client Officer at Panoplai, explained during the "Digital Twin & Synthetic Masterclass", the real risk isn't that AI produces obviously incorrect answers. It's that the answers sound convincing enough to believe.

"The outputs are shallow, stereotypical, sycophantic, invariable, and dangerously believable."

That final phrase, dangerously believable, became one of the defining ideas across IIEX Europe. Synthetic personas aren't replacing research. They're becoming another research tool, one that delivers the greatest value when paired with strong human judgment.

Quick Answer: Are Synthetic Personas Reliable?

The answer is yes, but only when they're used for the right problems and validated against real human evidence.

Synthetic personas work well for:

• Early concept exploration
• Understanding established customer segments
• Testing messaging directions
• Generating hypotheses
• Exploring likely attitudes

They should not replace human research when organizations are:

• Validating innovation
• Studying emerging behaviors
• Making high-risk strategic decisions
• Measuring real market behavior
• Looking for genuinely new signals

Across IIEX Europe, the organizations using synthetic personas successfully shared one thing in common: they viewed them as a way to strengthen research, not shortcut it.

What Are Synthetic Personas Actually For?

One misconception is that synthetic personas are intended to simulate an entire market.

That isn't how leading research teams describe them.

According to Adam Bai, a digital twin is "an interactive living simulation either of an individual or of a precisely defined customer, audience, consumer segment that you can actually ask questions of, engage with in real time, test future propositions against."

That definition shifts the conversation.

Synthetic personas aren't intended to replace participants. They're designed to help researchers explore ideas more quickly between rounds of traditional research.

Philips shared a similar philosophy.

Rather than treating AI personas as substitutes for consumers, their teams use them to better understand audience segments and improve concepts throughout innovation.

As Sehnaz Arasan, Consumer Insights AI Platform Manager at Philips, explained in "How Philips Accelerates Consumer-Centric Innovation with AI",

"The use case for us is really all about optimization and early feedback. It's really part of the process of generating faster but better, also more consumer-centric insights."

The emphasis is on optimization, not replacement.

Where Synthetic Personas Add Value

Across several IIEX Europe sessions, researchers described synthetic personas as particularly valuable during the earliest stages of research.

Instead of replacing consumer interviews, they help teams decide which ideas deserve further testing.

Organizations are using them to:

• Explore multiple concept directions
• Compare messaging ideas
• Understand established customer segments
• Generate hypotheses
• Refine discussion guides before fieldwork

The GLP-1 Consumer Shift presentation offered another interesting example.

Researchers found that AI moderation can sometimes encourage participants to discuss sensitive topics more openly because they feel less judged than when speaking to another person. AI can also adapt follow-up questions in real time, allowing conversations to probe individual responses differently.

These capabilities don't eliminate qualitative research.

They make qualitative research more focused.

Where Synthetic Personas Break Down

If there was one message repeated throughout IIEX Europe, it was this:

Synthetic personas are excellent at extending what we already know.

They're much less reliable at discovering what nobody knows yet.

Risham Nadeem, Director of Innovation at C Space, explained in "Keeping It Real: How Synthetic Humans Can Help Us Stay Human" why.

"Synthetic data isn't great for innovation... these models are only able to look at historic data for historic patterns. They can't project forward in that way. They can't identify new signals in that way. When topics are new or ideas are emergent, synthetic data is probably not a fit."

That's an important distinction.

AI learns from patterns.

Innovation often comes from breaking patterns.

Context also matters.

Adam Bai reminded attendees that what is representative for one business decision may be completely inappropriate for another.

Similarly, Philips deliberately designs synthetic personas to challenge researchers rather than simply agree with every prompt, recognizing that AI has a natural tendency to reinforce assumptions unless teams intentionally design against it.

How Do You Validate Synthetic Personas?

Perhaps the strongest consensus at IIEX Europe centered on validation.

No speaker suggested trusting synthetic outputs on their own.

On stage with Risham Nadeem, Robin Queripel, Global Senior Insights Manager at Sage, summarized during "Keeping It Real: How Synthetic Humans Can Help Us Stay Human" the challenge perfectly.

"You have to have a robust human data set or existing research to act as an anchor, because how else are you going to identify whether or not the AI is hallucinating?"

That idea surfaced repeatedly across sessions.

The organizations finding success aren't replacing human research.

They're comparing synthetic outputs against interviews, surveys, and existing customer knowledge through parallel testing.

Validation also means asking better questions about AI itself.

Adam Bai challenged attendees not to accept vendor accuracy claims at face value.

"If you hear a vendor say something like, 'Our synthetic data is 95% accurate,' the next question you should always ask is, 'What does that mean?'"

Instead of focusing on headline accuracy numbers, Bai encouraged researchers to evaluate whether synthetic personas demonstrate behavioral realism, consistent preferences, meaningful qualitative depth, and relevance to the specific business decision they're trying to make.

Philips approaches validation as an ongoing process rather than a one-time exercise.

Their synthetic personas are continuously enriched with fresh qualitative research so they evolve alongside changing consumers instead of becoming static representations of the past.

What Good Practice Looks Like

Across all four sessions, a practical framework began to emerge.

Successful organizations don't start by asking how many synthetic personas they can create.

They start by asking where synthetic personas genuinely improve the research process.

Good practice includes:

• Starting with narrow, well-defined use cases
• Running synthetic and human research in parallel
• Continuously refreshing personas with new qualitative data
• Assigning researchers responsibility for interpreting outputs
• Treating AI as directional evidence rather than final proof

Rather than competing with human research, synthetic personas become another layer of evidence that researchers can compare, question, and refine.

What This Means for Insight Teams

Perhaps the biggest shift discussed at IIEX Europe wasn't technological.

It was professional.

As synthetic personas become more capable, the value of insights professionals increasingly comes from knowing when AI deserves confidence and when it requires skepticism.

Researchers become validators.

Interpreters.

Decision stewards.

The organizations gaining the most value from synthetic personas aren't necessarily using the most advanced AI.

They're applying the strongest research discipline.

Conclusion

Across IIEX Europe, synthetic personas were rarely presented as replacements for traditional research.

Instead, speakers consistently described them as accelerators for learning, exploration, and iterative decision-making.

That distinction matters.

The teams seeing meaningful results aren't collecting less human data. They're using synthetic personas to make every interaction with real people more targeted and more valuable.

Adam Bai captured that philosophy well when discussing adoption strategies.

"You wanna go deep and narrow first... This is about supercharging human data and collecting better quality human data, perhaps less often, but perhaps better quality."

That may be the clearest takeaway from IIEX Europe.

The future of synthetic personas won't be measured by how convincingly they imitate people.

It will be measured by how effectively they help researchers ask better questions, make better decisions, and deepen, rather than replace, their understanding of real human behavior.