Suraj Narayanan Sasikumar, CTO and co-founder of Socialtrait, explains how AI communities with 50+ dimensions per persona can predict brand crises, test campaigns, and reach audiences traditional research can’t touch.
What if you could find out how people will react to your campaign before a single real person sees it?
That’s the question Suraj Narayanan Sasikumar has been chasing since a Hacker News comment blew up in his face. He went on vacation, came back, and found his post buried under negative replies. He hadn’t meant any of it the way people read it. And as an AI consultant with reinforcement learning on his résumé, his first thought was: could I have simulated that reaction ahead of time?
That thought became Socialtrait. Jessie Lizak and I recorded this one outside here in Florida, and it’s one of those conversations I keep coming back to.
How do AI personas predict brand reputation risk?
Suraj Narayanan Sasikumar described Socialtrait’s roadmap as a three-phase master plan. Phase one, built in the Berkeley SkyDeck accelerator before ChatGPT existed, was a brand safety tool. Companies routed their outbound communication through hundreds of AI personas, and the system flagged anything likely to trigger a crisis with a red, amber or green rating.
He used the KFC example: a promotional message in Germany that tied chicken to Kristallnacht, a somber memorial. An intern or an automated system might miss the history. The personas didn’t.
“We were able to predict more than 90% of the cases where the gaffes had occurred or the crisis situation occurred.”
— Suraj Narayanan Sasikumar, CTO and Co-Founder, Socialtrait
Here’s the catch, and it’s a lesson for anyone selling prevention.
“People didn’t want to invest in something because of something you would predict or you would prevent.”
— Suraj Narayanan Sasikumar, CTO and Co-Founder, Socialtrait
Nobody budgets for the crisis that didn’t happen. But the early customers kept reading the persona feedback to improve their messaging, and that pointed Socialtrait toward phase two.
What are AI communities, and how are they used for consumer research?
Phase two is where Socialtrait lives today. Instead of a loose collection of personas, you build an AI community around a target characteristic. Suraj Narayanan Sasikumar’s example: if you’re Nike entering Egypt, you can spin up a community of 18 to 25 year old women who run.
And here’s where it gets interesting. Those personas aren’t one-dimensional. Each one carries more than 50 dimensions, because real people are more than their segment.
“As humans we are multi-dimensional. Some of those aspects might actually be useful for a customer to understand to create their product or customize their product.”
— Suraj Narayanan Sasikumar, CTO and Co-Founder, Socialtrait
Once the community exists, you treat it like a WhatsApp group. Run focus groups, surveys, image rankings, even attention heat maps on a new landing page. The personas are agents, too. Post a question and they decide for themselves whether they care enough to answer.
“We didn’t want to mimic actual people. We want to create an entirely new person who has an entirely new set of characteristics, but we want to give them a behavior that is representative of that particular person.”
— Suraj Narayanan Sasikumar, CTO and Co-Founder, Socialtrait
I told Suraj this reminded me of something Reed Hastings talked about: Netflix running scripts through AI built on its own customer data to predict whether a show will succeed before buying it. Socialtrait is democratizing that for companies that don’t have Netflix’s budget. Small and mid-size businesses have never been able to afford big consulting firms running real focus groups. That’s a big deal.
Can synthetic personas reach hard-to-reach audiences ethically?
This was the story that got me. Socialtrait worked on a project to raise awareness of pre-exposure prophylaxis among people at high risk of contracting HIV. It’s a group that’s very hard to interview in person, but it talks openly and anonymously online.
Suraj Narayanan Sasikumar explained that Socialtrait doesn’t scrape any individual’s data. It learns a behavioral layer from the broad picture and applies it to entirely new AI personas. The early results matched the opinions the client had already gathered on the ground.
“We don’t have to infringe on anybody’s specific privacy because we can learn a behavior layer on top of that and then just from a very big picture perspective understand that behavioral layer and then apply it to a new person.”
— Suraj Narayanan Sasikumar, CTO and Co-Founder, Socialtrait
He also told us Socialtrait turned down political campaign work early on because of how easily this kind of system could be used to manipulate discourse. I respect that. It’s the kind of line more AI companies need to draw out loud.
How do you handle bias in AI-generated audiences?
Jessie pushed on ethics, and Suraj Narayanan Sasikumar gave an answer I didn’t expect. Socialtrait built a patent-pending algorithm to match distributions across many dimensions at once. That lets a customer mirror the real world or deliberately oversample underrepresented groups so their voices are heard.
“We are in the business of creating bias, but we are guiding that bias to look at a specific slice of an audience that a particular customer is interested in.”
— Suraj Narayanan Sasikumar, CTO and Co-Founder, Socialtrait
Honestly, that’s the most candid description of segmentation I’ve heard. Every audience is a bias. The question is whether it’s the right one.
What’s next: connected communities and network effects
Phase three is a social network of AI personas. Communities overlap the way they do in real life: the young runner in Egypt also belongs to a community of women in Egypt, a gym community, and so on. Drop a new sneaker concept into Nike’s core fan community and you can watch it spread through likes, shares and follows across the graph. Suraj Narayanan Sasikumar said internal tests showed significant correlation with real-world data.
On the tech stack side, Socialtrait built almost everything from scratch because the tools didn’t exist yet: load balancer, model router, orchestrator, observability. They’re now moving to established tools like SGLang and vLLM for model serving and Langfuse for observability, and they require the team to use AI coding tools like Cursor and Windsurf.
The bigger picture
At BDEX, we spend our days on real identity data, real people, real signals. So you might think synthetic audiences would make me nervous. They don’t. Simulation and real data answer different questions. Socialtrait helps you figure out what to say and who it’ll resonate with. Clean identity data helps you actually find those people and reach them.
The companies that win will use both. Test the idea against a million simulated consumers, then put it in front of the right real ones. That’s how you stop guessing.
Connect with Suraj Narayanan Sasikumar: Learn more at socialtrait.com, find Suraj on LinkedIn, or send him a DM on X, where he says he’s most active.
Ready to find more of your best customers? BDEX uses clean identity data and AI to help you build audiences that actually resonate. Visit bdex.com and click “Talk to an Expert.”
Watch the full episode: AI-Powered Consumer Insights: Innovation, Ethics, and Impact
This article was adapted from an episode of Deconstructing Data, BDEX’s weekly podcast on data-driven marketing. Tune in live every Thursday at 4:15 PM Eastern on LinkedIn.
About BDEX: For companies that need clean identity data to power their products, BDEX offers unmatched quality and execution. Visit bdex.com and click “Talk to an Expert” to get started.



