AI in Research: Speed, Depth, and a Sneaker Wall

Author's note: Penning down some reflections from Marvin's AI research debate held in July 2026. Full disclosure, I also work on the marketing team that put this AI research debate webinar together, so I'm not a neutral bystander. Just someone who sat through it twice —  once helping build it and once as a viewer.

‍

Does AI give you richer insights, or does it flatten results in the pursuit of faster answers?‍

‍

If you weren't there, Owen Sanderson (Senior Director of Research and Design Strategy at IA Collaborative) and Prayag Narula (CEO & Co-founder, Marvin) took the stage to argue this question. Owen came in as the practitioner arguing for depth and human judgment. Prayag came in arguing that speed is no longer optional. Neither of them were wrong.

Does this debate scream ragebait? I kid. Or do I? 

Let’s get into it, and find out!

Watch the event on demand →

‍

Where the room started

Before the debate even started, the audience of more than 100 people voted on which statement best matched our perspective:

• 73.9% of us picked "it's not that simple, there are valid points on both sides." 
• Only 16% were ready to call AI a new era of research productivity.
• Just 10.1% were willing to call it a sea of sameness.

Almost nobody walked in with a hot take. Most of us walked in expecting the answer to live somewhere in the middle.

‍

The human case: what Owen & Prayag brought to the stand

Owen Sanderson has spent years doing research in places most researchers never go. Operating rooms, rural clinics, farmhouses in Kenya. He clarified he is definitely not on the team, “AI is bad,” but believes there are things a researcher standing in a room will always catch that a transcript never will. 

On the other side, Prayag Narula, far from being a skeptic of research – built a company around it. His argument is about timing. A study that takes three months and lands after the market has moved is late. And in a product world where teams are shipping every few weeks, late is the same as irrelevant. AI, in his view, is what keeps research in the room where decisions are actually being made.

Both agreed that judgment is irreplaceable. Where they diverged—and where the debate lived—was how much time teams still have to exercise that judgment.

That framing is what set up everything that followed.

‍

What a researcher sees that AI doesn’t

Owen opened with an interesting story.

His team was working with a hospital in Manhattan, a brand new bed tower, years in the making. The challenge was patient flow: how do you move people from the front door to their physician's office in a way that actually works? They'd done the contextual research inside the hospital. But nobody in that room could imagine something fundamentally better while standing inside the building they already knew.

So Owen's team took them somewhere else entirely. They brought hospital administrators and physicians onto the Scarlet Lady, a Virgin Voyages cruise ship, and watched how the ship moved thousands of people through check-in and onboarding.

A hospital is not a cruise ship. But the staff needed to get outside the setting they were used to seeing every day. Context changed the team's perspective.

That's a hard story to argue with. It's also, as Prayag pointed out, "a once-in-a-career kind of project." Building a hospital is a long-lasting, mammoth project. Owen's client had the luxury of time.

Truth be told, most teams don't have that runway. And I suspect most of you don't either.

But then Owen kept going. And the stories that followed are the ones that stayed with me, because they all made the same point:

The most valuable insights are often the ones nobody says out loud.

At the end of a wellness study session, Owen's team asked a participant to show them around their home, a standard and friendly "just walk us through your space" moment. The participant led them into their bedroom. An entire wall, floor to ceiling, was lined with a hundred pairs of sneakers in individual boxes.

They weren't purchased for resale. Each one represented a moment, a version of who this person was or wanted to be. Health and wellness, it turned out, wasn't just about physical habits. It was about identity, how this person constructed and displayed a sense of self that no survey question about diet or exercise would have ever reached. That story changed the product direction. And it only existed because someone was standing in that room.

The sneaker wall was memorable because it illustrated something AI still struggles with: noticing meaning that no one explicitly articulated.

Owen shared a few more stories that made the same point.

Your product's real use case might not be the one you designed for.
A biodiesel cooling unit was built to help smallholder dairy farmers preserve milk for a better market price. Being in the field revealed that the same unit was just as important for keeping the evening's beer cold. That completely reframed what success for the product actually looked like.

What people say about your product and what they actually do with it are two different datasets.
A participant loved a pharma brand's packaging. Then the researcher asked how she'd dispose of it. She folded it inside-out before throwing it away, so her name and the medication wouldn't be associated with her. She described one experience, but her behavior told a different story.

These stories came from someone standing in a room, noticing things nobody said out loud.

That's the gap a transcript alone would never show you.

‍

Speed wins - until it doesn’t

Prayag told a story about a pricing study run by researchers he respects, that took from January to April. By the time they finished, the market had already moved on, and the results didn’t matter anymore. 

Prayag’s point wasn't that rigor doesn't matter. It's that a single, beautifully done study delivered too late can be worth less than three rougher ones delivered on time.

It’s a familiar tension. 

Patrick Neeman's piece, published about and ahead of this debate, argues that speed was never really the axis. AI took the mechanical half of the job (transcribing, tagging, first-pass synthesis) and left the half that was always ours, deciding what a signal means. I agree! 

Listening to both speakers, though, I realized they had moved on from debating whether judgment matters. They'd settled that almost immediately.

A practical question arose in conversation:

How slow can research actually afford to be?

Owen argued that slowing down creates the space to notice what others miss. Prayag argued that if research can't keep pace with how products are built, it risks being left out of the decisions altogether.

I don't think either of them is wrong.

In fact, I think that's where they ultimately landed:

Research fails when teams use speed as an excuse to stop being curious.

‍

The questions worth sitting with

• Are we producing more research, or better research? One attendee asked this deceptively simple question. If synthesis gets faster, the easy failure mode is just... doing more of it, not doing it better. That's a stakeholder-expectations problem as much as a tooling one.
• What do we actually do with the time we get back? This is a piece of the same puzzle as above. If AI takes the transcription and first-pass synthesis off our plate, is that time going into deeper interpretation, or "shipping faster"? And how do we get credit for the former when stakeholders mostly notice the efficiency gains?
• How much of "AI flattens insight" is actually about AI vs. the organization around it? Someone framed AI as a mirror. It reflects and amplifies whatever habits and incentives were already in the building. That reframes a chunk of this debate as an org-design question wearing an AI costume.

Safe to say, I left with better answers than I walked in with, which feels like the right outcome for a debate that never had a clear winner in the first place.

‍

The vote that closed the debate

By the end, 87% of us landed on "I see both the benefits and the tradeoffs." Concern dropped to 6.5%, uncertainty to 4.3%, and blind trust in where AI is taking research stayed basically nowhere, at 2.2%.

The room didn't swing toward either speaker's corner. It consolidated. The energy in the virtual-room felt like we went from "it's complicated, but I'm still holding two extremes in my head" to "it's complicated, and I've actually made peace with that." More of us landed on holding the tension than on resolving it.

‍

What the transcript won't tell you

AI isn't flattening insights.
We are when we let speed become the excuse to stop being curious.

Owen's worry isn't the tool. It's frequency bias. "Sometimes the things that are infrequent lead to the most salient 'aha' moments in research. If we're just focused on frequency, we're going to miss some of the nuance." The sneaker wall came up once. The beer fridge in Kenya came up once. The packaging folded inside-out came up once. None of them would survive a frequency filter. 

The fix isn't slowing down. It's being intentional about what you hand over. Transcription, tagging, first-pass synthesis — AI's. The hunch, the observation, the thing nobody said out loud — ours. As Owen put it, "judgment's expensive. Let the AI do what it's great at but keep the researcher in the judgment position."

Curiosity is the job! And speed, used well, simply creates more room for it.

‍Watch the full webinar →‍

See how Marvin helps research teams close the listening gap →