From Research to Reinvention: Prayag Narula on Startups, Setbacks & AI Innovation

The uncomfortable pivots, the moments you get “dumped” by your own company, and the stubborn grit it takes to get to the end of a day when nothing is working.

Prayag’s story begins where many product stories end: in research, not as a supporting function but as an identity. He came to the US to study research at Berkeley, taught design research there, and expected to take the academic path all the way to a PhD and professorship. He even joked that part of his dad still expects him to finish. But the more interesting detail was what drew him in the first place. With a computer science background, he became obsessed with HCI, the question behind every feature: how are people actually using this?

Before “HCI programmes” were a given, he went to Helsinki, Finland, to learn from the best place to study interaction at the time: the Nokia orbit. It was an era when Nokia held an enormous market share, produced serious research, and also demonstrated a failure mode that still feels painfully familiar. They would do brilliant work, then put the findings in a closet and lock the door. Prayag delivered the line as a laugh, but it landed like a warning. Insight does not matter if it cannot travel.

That theme, how insight moves through a company, becomes the spine of Marvin.

How a researcher ends up in marketing

Ryan pulled the conversation into Prayag’s first company, LeadGenius, which operated in demand generation and marketing. It is not the obvious destination for someone trained in research. Prayag’s answer was blunt and oddly encouraging: he would not recommend it unless you have to, but selling is also a form of probing. If you can ask good questions and listen carefully, you already have more of the “sales gene” than you think.

The real twist, though, was that he did not start in marketing. He started in AI back in 2011 and 2012, trying to build datasets for machine learning systems to train on. The market was not ready. He watched Scale AI later do something similar and become enormous, and he used that gap as a practical lesson: timing is a variable you cannot control, but it may be the most decisive variable you face.

So LeadGenius pivoted. They kept the same underlying technology and moved it into a market that actually had demand: sales and marketing data. It was a classic founder paradox. Prayag spent a decade teaching people how to do sales while learning sales in public.

When Ryan asked for the biggest lesson from that first time as a founder, Prayag did not talk about fundraising or product market fit. He talked about leadership, and he talked about it in the least romantic way possible. He became CEO at 27, with no “real job” in product, and realised that titles do a lot of signalling, especially early. People hear “CEO” and assume leadership. But what sustains leadership is not the label. It is the doing. Showing up, building something, talking about it, and doing the work that nobody else is doing yet. In his view, that is most of leadership.

It echoed one of Ryan’s own refrains from his earlier talk: leadership is a choice, not a position. Here, the nuance was that leadership is also behaviour. It is an action you repeat until other people start trusting it.

Getting “dumped” and deciding what you really want

If the first half of the fireside was about identity, the second half was about reinvention. Prayag left LeadGenius after ten years. He described it like a breakup, laughing as he said it was “consensual” in the way people say breakups are consensual. He got dumped. Then he took a break.

When he told his wife he wanted to start something new, she did not ask what it was. She asked for the deadline. How long would he try before he had to go get a salary again? There is something grounding about that detail. The myth is that founders are powered by pure vision. The reality is that most founders are also negotiating time, responsibility, and household risk.

So why Marvin?

Because of a problem he carried forward from LeadGenius. In the company, people kept asking the same question: what have customers said about this? And there was no good answer. Feedback was scattered across support systems, call recordings, Salesforce, NPS tools, and whatever else had been bolted on over the years. There was no single place to go when you needed the truth of the customer voice, and no simple way to connect decisions to what users had actually said.

Marvin, in Prayag’s words, is a research repository and a research copilot. The phrasing matters. The repository is about centralisation and access. Copilot is about helping people do the work better. The intent is not just to store insight, but to make it usable in the flow of product decisions.

There was also a second reason, more personal and arguably more sustainable. Prayag wanted to work with his “tribe” again. After a decade in sales and marketing, he wanted to build for researchers, designers, and product builders, the people he genuinely enjoys spending time with. If you are going to devote another 10 to 15 years to something, he suggested, the community you build with becomes part of the product.

The awkward truth about research in startups

Ryan pushed into a spicy topic: in many startups, research is one of the later hires. Prayag’s answer was contrarian. At LeadGenius, the first product hire was a researcher. But the outcome explains why the conventional pattern persists. In a small company, a researcher’s mandate expands quickly. You start with understanding problems, then you are pulled into shaping solutions, aligning teams, and translating learning into action. Eventually, you are doing product management, whether or not your title says so.

In other words, startups do not avoid research because it is unimportant. They avoid it because the organisation cannot yet contain the role. Research becomes a responsibility spread across everyone: the CEO, PM, sales, and design. The challenge is not whether research happens, it is whether it happens with enough structure to be trusted.

That set up Ryan’s next question: if you democratise research, what happens to quality? Is bad research better than no research?

Prayag answered with a story from graduate school. He asked his qualitative research professor how to do proper research when he could not even afford to pay participants. Her answer: Talk to the person next to you. Guerrilla research, done with intent, can still be useful. Not every project needs a flawless methodology to produce a signal. But he drew a clear line. If you are simply seeking validation, asking leading questions, or fishing for compliments about your design, you will not get value. If you meet a baseline of honesty and curiosity, “messy” research can still move a product forward.

AI as a coach, not a replacement

From there, the conversation moved naturally into AI. Ryan asked whether AI could help correct bad questions, create guardrails, and steer teams toward the real problem. Prayag’s answer was an unambiguous yes, and he referenced one of Marvin’s features: giving feedback on interviews. The AI analyses the conversation and suggests improvements, like pointing out leading questions or highlighting when the interviewer talked too much.

Then came the surprising behavioural insight: people were more willing to accept feedback from an AI than from peers. AI feedback felt less personal, less political, less likely to bruise ego. It is a small point, but it hints at a new role for research teams. Beyond producing insight, researchers may increasingly act as coaches who help others seek truth, while AI handles a portion of the routine guidance.

Ryan raised a bias next: can AI remove bias from research?

Prayag had a spicy take. Trying to remove bias completely is futile. Bias exists in framing, recruiting, context, and language. What matters is addressing it, naming it, and being transparent about what might have influenced results. In that framing, bias becomes less of a methodological purity test and more of a trust practice.

Synthetic users and the future of “junior work.”

On synthetic users, Prayag was pragmatic. Synthetic training data works well in many AI contexts, and much of what modern systems learn from includes synthetic data. But synthetic users are not a replacement for real humans. Where they can help is in two places: stress testing a discussion guide before you talk to real participants, and augmenting weak quantitative signals to reach statistical significance, used carefully and with awareness of limitations.

The fireside ended where many AI conversations end, with the job question. Is AI coming for your job?

Prayag’s answer was reassuring, but not complacent. AI does not have judgment and taste. Taste makers will be fine. The real concern is how people become taste makers. If AI performs “junior work” extremely well, organisations may lose the apprenticeship pathways that help early career people build judgement over time. That is the shadow side of efficiency: you can automate the ladder out from under the next generation.

His advice to a scrappy UX generalist at a startup followed the same logic. Your job is not only to talk to users. It is to help everyone else in the company have better conversations with users, and to build habits that keep learning visible. Learn quickly through communities like conferences and by reading research papers that are often more accessible than people assume.

Underneath all the jokes, the conversation left a clear takeaway. Reinvention is not a moment. It is a practice. It is building systems that keep truth from getting locked in a closet, and building the resilience to keep showing up long enough for the work to compound.

Want to watch the full talk?

Get access to the full talk and exclusive notes in the UXDX USA 2025 Post Show Report: https://uxdx.com/post-show-report

Or make your Product Career AI-proof by downloading our free eBook: https://uxdx.com/ebook/career-compression/