Somewhere in a Fortune 100 design org, a researcher is searching for a single sentence a customer said in a Zoom call back in February. She knows the quote exists. She knows the insight matters. She just can't find it. This - the great forgetting of qualitative research - is the small, expensive, deeply human problem that Marvin set out to solve.
Marvin, the operating name of Equafin Inc., is an AI-native customer insights platform that records, transcribes, organizes and - crucially - lets you ask questions of every customer conversation a company has ever had. About 4,000 teams use it. Microsoft uses it. Razorpay uses it. Half of the Fortune 10 healthcare companies use it. The platform claims a 4.8 rating on G2, which is the kind of score you get when your customers are themselves researchers and therefore have unusually firm opinions about software.
For roughly thirty years, the practice of user research has been stuck in a productivity paradox. The tools to collect qualitative data have multiplied - Zoom, Lookback, Dovetail, UserTesting, surveys, sales calls, support tickets, the works. The tools to find anything in it later have not. A user researcher in 2025 still spends most of her week, ironically, doing manual labor: clipping videos, tagging transcripts, summarizing themes, and emailing PowerPoint files to product managers who will, with some confidence, ignore them.
The internal joke in the field is that research libraries are write-only. Insights go in. Insights do not, usually, come out. The result is a peculiar economic waste: companies pay for the most expensive kind of data they will ever collect - direct conversation with humans - and then they file it away where nobody will ever read it again.
Prayag Narula has done human-in-the-loop AI before. His previous company, LeadGenius, raised more than $25 million pairing software with global human workforces to do go-to-market data work. He spent his graduate years at UC Berkeley publishing more than a dozen papers on crowdsourcing, HCI and machine learning - which is to say, he had a head start in thinking about how machines and humans share cognitive labor. His co-founder and brother, Chirag Narula, came from Blinkit, where he led a 30-person design team building one of India's most-used consumer apps.
Their bet, made in 2020 and articulated more fully when Marvin announced a $3.8M seed in early 2022, was almost suspiciously simple: large language models were about to make the messiest data in any company - human conversation - finally searchable. The researchers wouldn't be replaced. The clerical work would be.
Berkeley I-School alum. Previously built LeadGenius. Has the unusual distinction of being both an academic in human-computer interaction and a person who has met payroll for hundreds of people.
Designed Blinkit, the 10-minute delivery app most Indians have on their home screen. Brings a consumer product sensibility to enterprise research software, which is rarer than it sounds.
Marvin does five things. None of them are conceptually new - that is the point. The trick is doing them all in one place, on top of the same searchable corpus, so a marketer can run the same query as a designer and get the same evidence trail.
One library for interviews, sales calls, support tickets, surveys and notes. 30+ integrations pull in data from where it already lives.
Ask the research library a question in English. Get answers with citations back to specific clips and quotes - not just vibes.
An AI interviewer runs structured studies in 40+ languages while you sleep. Useful if you've ever tried to staff a usability test across three time zones.
Thematic coding, tagging, affinity mapping. The unglamorous part of research, made faster.
Dynamic reports, video reels, highlight playlists. Built for the meeting where the VP asks for "the clip."
Talking your book is cheap. Receipts are not. Marvin has the kind that matter to enterprise buyers. The customer roster includes Microsoft, Criteo, Razorpay and Pantheon. Ten Fortune 100 companies pay for seats. Half of the Fortune 10 healthcare leaders are on the platform - a sector that does not, traditionally, hand its customer data to startups with cheerful logos.
The compliance shelf is the unglamorous reason that works: SOC 2, ISO 27001, ISO 42001 (the new responsible-AI management standard), HIPAA, and GDPR. The 4.8-out-of-5 G2 rating is a separate kind of receipt - the kind issued by the people who actually use the software, not the people who buy it.
The official version, smoothed for a website, is that Marvin makes customer understanding accessible to every team. The unofficial version, which you can hear in the company's blog posts and product launches, is more interesting: it is that qualitative research has been gate-kept for too long by people with the time and tooling to do it manually, and that AI - applied without too much drama - finally widens the door.
That implies a slight political stance inside the customer-research community. There are researchers who worry that AI summarization flattens nuance, and they are not wrong to worry. Marvin's pitch is not that AI replaces the researcher's judgment but that it removes the parts of the job that nobody, on their deathbed, will say they wish they had done more of.
The interesting future question for a company like Marvin is not whether AI can summarize an interview. That question is settled. The future question is whether the customer knowledge inside a company can become a kind of institutional memory - a thing the org refers back to, automatically, every time it makes a product decision. If the answer turns out to be yes, the unit of competitive advantage stops being who can run the most studies and becomes who can actually remember what they found.
Back to the researcher in February. In the world Marvin is building, she does not have to remember which folder the quote was in. She types her question into a box. Marvin pulls the clip, in context, with the timestamp, with the four other customers who said something similar in the last eighteen months. She forwards it to product. The decision gets made on Tuesday. The customer, who never knew her name, finally gets the change she asked for.
It is not a revolution. It is a piece of plumbing - the unsexy kind that, once installed, you stop noticing. Which is probably the highest compliment software can earn.