Before pilots ever sit in a real cockpit with 200 passengers depending on them, they crash — hundreds of times — in a simulator. They experience engine failures, crosswinds, bird strikes, instrument malfunctions. They fail safely until they can't fail at all.
Product teams have never had that.
You build something. You launch it. It underperforms. You learn. You rebuild. The cost of that cycle — in time, money, and morale — can be catastrophic for an early-stage company
“What if product teams could test their ideas in a simulation of the real world before they bet the company on it?”
Not a focus group. Not a survey. Not a chatbot pretending to be a customer. A living, social, simulated world — with thousands of AI personas who have memory, relationships, opinions, and a pulse.
That's what we've been building at TestSynthia. Here's the full story.
Stateless vs Stateful: The Insight That Started Everything
When we started building, we obsessed over one question: what makes a human respondent's opinion valuable?
The answer wasn't intelligence. It wasn't demographic accuracy. It was history.
A human participant in a research study has lived a life. They've bought things, been disappointed by things, recommended things to friends, been talked out of purchases, changed their mind about brands over years. Their opinion about your product isn't formed when you ask the question — it was formed by decades of accumulated experience that you're now accessing.
Traditional AI personas, even very sophisticated ones, don't have that. They're what we call stateless.
What Stateless Means
A stateless persona is summoned when needed, responds based on a generic profile, and disappears. It's like hiring someone for a one-hour shift, getting their opinion, and then they die. No memory. No history. No continuity.
What Stateful Means
A stateful persona is persistent. They exist between sessions. They have a history of interactions, a memory of products they've evaluated, opinions that have been shaped and reshaped by conversations and experience. They go to sleep at the end of the day. They come back next week with slightly older, slightly more refined views.
The Conversation Layer
Think about how a real research panel actually works. You bring a group of people into a room. They fill out their responses individually. And then — almost inevitably — they start talking to each other.
That post-survey conversation is where a lot of the most honest signal lives. But traditional research throws it away.
In TestSynthia, they don't disappear.
Once a survey round is completed, the personas don't just file their responses and go silent. They start talking — but not randomly. Conversation clusters form naturally based on shared context.
A 34-year-old product manager gravitates toward other product managers. A CFO finds their way to other CFOs. A first-time parent connects with others in the same life stage. Two personas in the same income bracket and political leaning find common ground.
Just like in real life — when you meet someone new, the first thing you establish is where they're from, what they do, what world they live in. Shared identity is the on-ramp to honest conversation.
World Events: The Simulation Doesn't Pause
In our simulation, the world keeps moving.
We built a mechanism called world events — injections of real-world context into the simulation.
When something significant happens, we can introduce it as a world event: a virus outbreak, layoff news, a major drug gets FDA approval, a recession signal appears in the data.
Opinions shift in ways that weren't scripted or anticipated — because that's what happens in the real world. The personas don't know they're in a simulation. They're just living their lives and responding to what's happening around them.
A live simulation, can be seen in here.
Memory, Reflection, and Evolution
A persona's first response to your product is the least interesting one they'll ever give.
Not because it's wrong — but because it's thin. It hasn't been shaped by experience yet. It's a snapshot of someone who just arrived.
The more interesting question is: who does this persona become over time?
Every persona carries four streams of memory that load every time they're called into a simulation:
Product memories — drawn from every research panel they've participated in. Not the raw responses — compressed, importance-scored summaries of what they experienced. Only memories scored 6 or higher on a 10-point importance scale are passed in the next step. The forgettable stuff fades. The things that actually moved them stay. A persona who was genuinely shocked by a pricing question carries that. One who found a feature irrelevant doesn't burden their future self with it.
Social memories — the residue of their conversations. Not just what was said, but who said it. The system tracks which persona they talked to by name, what was exchanged, and how important that exchange was. When Sarah from accounting told them she switched away from a competitor last month, that memory travels with them into your study.
World event memories — injected as targeted news summaries, routed by job cluster. A persona mapped to healthcare gets events relevant to healthcare. A finance-cluster persona gets different signals entirely. The same world, filtered through the lens of what actually touches their life. Up to five active world events shape their current context at any given time.
Reflection — and this is where it gets interesting.
The three memory streams above accumulate over time. But a persona carrying hundreds of raw memories isn't the same as a persona who has processed them. So once memories cross a certain threshold, a reflection is generated — a synthesised, first-person consolidation of everything accumulated so far. What they've learned. What patterns they've noticed. How their views have shifted.
The result is a persona who doesn't just have more data than when they started. They have better judgment. Opinions that are anchored in experience rather than generated from a profile description alone.
Segments Deep Dive: Who Actually Wants Your Product
Running a simulation across a thousand personas is only half the job. The other half is understanding which slice of those thousand people you should actually build for.
Most research tools give you averages. we give you segments.
Once a simulation completes, the analysis layer breaks your respondents into distinct groups — not by the demographic inputs you configured, but by how they actually responded. A 45-year-old nurse in Texas and a 28-year-old grad student in Oregon might both score your product a 4.2, cite the same top driver, and share the same top barrier.
That's a segment — defined by behaviour and response pattern, not by the demographic box you put them in.
A detailed report on this can be found in here
Flight compensation recovery agent for frequent flyers
Test in the Simulation. Sell in the Real World.
Fifty years of market research orthodoxy is a hard thing to disrupt. We know that.
We believe product decisions deserve better inputs. Not faster garbage. Actual signal — from personas with memory, living in a social world, experiencing the same context shifts your real customers experience every day.
And the ambition doesn't stop at product teams.
When we talk internally about what we're building, we use a specific frame: we're constructing a duplicate of the real world.
Not a perfect replica — that's not possible and not the point. But a simulation that captures enough of the texture of real human social and economic behavior that it can serve as a testing ground for decisions that would otherwise require real-world experiments with real-world consequences.
"The most valuable thing about a flight simulator isn't that it teaches you what to do. It's that it lets you discover what you don't know before the stakes are real."
That's the bet we're making with TestSynthia. The future of decision-making under uncertainty isn't better s or smarter analytics. It's simulation. A world where you can run the experiment before you run the experiment.
The simulation is ready. The question is whether you'd rather learn what works before the launch or after it.
We're live now — If you're making a high-stakes launch decision, we'd love to show you what the simulation sees
Top comments (0)