Strella

It is Tuesday afternoon, and somewhere a product manager just got back two hundred customer interviews. Not transcripts to read next month - synthesized themes, the exact clips where people hesitated, and a short answer to the question that started it all. No recruiter was emailed. No researcher stayed late. The interviews were run by Strella, and they finished before lunch.

That is the trick Strella is selling, and it is a strange one. The company built an AI that conducts qualitative interviews - voice conversations, up to ninety minutes, with follow-up questions that adapt to whatever a person just said. Then it turns the pile of recordings into something a busy team will actually use before the decision is already made.

Everyone agrees you should talk to customers. Almost no one has the time.

Lydia Hylton and Priya Krishnan had both lived inside the bottleneck. As consultants and product managers they watched the same ritual repeat: a company decides it needs to understand its users, then discovers that doing so properly takes about eight weeks. Recruit participants. Schedule them. Run the calls. Transcribe. Tag. Synthesize. By the time the deck is ready, the roadmap has moved on without it.

So teams cheat. They send a survey, which is fast and shallow and quietly riddled with fraud and bots. Or they skip research entirely and call it "moving quickly." The depth of an interview and the speed of a survey have always been a trade. You picked one. You lived with the consequences of not picking the other.

Caption: somewhere in this sentence is the eight-week research process that drove two founders to quit and build a robot.

Two friends from high school, betting an interview could be automated without being cheapened.

Hylton and Krishnan met as teenagers and had wanted to build something together more or less ever since. Hylton, a former Bain consultant, took the CEO seat. Krishnan, who had done time at Fitbit and DoorDash, became COO. Their wager was specific: large language models had finally gotten good enough to do the part of research that humans are worst at - listening patiently to the hundredth person and still asking a sharp follow-up.

The counterintuitive discovery came early. People are more candid with the AI than with a human interviewer. No one is performing for a stranger on a video call. The machine never sighs, never rushes you, never telegraphs the answer it wants. The feedback comes out cleaner.

It recruits, it interviews, it listens, it summarizes. You read the highlight reel.

In practice Strella runs the whole arc of a study. It recruits participants from a global panel and schedules them. A multilingual AI moderator runs the conversation, probing for the "why" behind an answer the way a good researcher would. Then real-time synthesis turns hours of talk into themes, instant highlight reels, and customizable reports a stakeholder can poke at without watching a single recording.

There is a quieter feature that tells you who the buyers are: you can skip the AI moderator entirely and just upload your own human interviews for Strella to analyze. The company is not religious about how the talking happens. It cares about the part where the talking becomes a decision - which is, conveniently, the part everyone else hates.

The numbers do the arguing here.

A research tool is easy to demo and hard to trust. So look at what the early customers did with it. The clearest data point belongs to Duolingo, which used Strella for video concept testing and watched a six-week project collapse into two days - fast enough to put results in front of the C-suite while the idea was still warm.

Caption: the orange bar is what eight weeks of good intentions looks like. The yellow one is Tuesday.

Caption: a customer list that runs from owl-shaped language apps to oat milk. Curiosity, it turns out, has no industry.

Make customer understanding a habit, not a quarterly fire drill.

Strella's stated goal is to democratize customer research across an organization - to take the tedious middle of the process and let AI carry it, so that "we should really talk to users" stops being a thing people say and start being a thing people do. The founders frame it as keeping research human by removing the parts that were never the human's job in the first place.

There is a healthy skepticism to entertain here. An AI that interviews people and then tells you what they meant is a lot of trust to hand a model. Strella's answer is human oversight kept in the loop and a deliberate bias toward letting customers be candid. Whether that holds at the scale of thousands of interviews is the open question - and exactly the one its investors are paying to watch.

If research gets this cheap, the excuse for not doing it disappears.

The interesting consequence is not faster decks. It is what happens when talking to a hundred customers costs almost nothing. Research stops being a special occasion and becomes ambient - a thing you do continuously, the way you check analytics. The companies that win the next decade may simply be the ones who stayed closest to what their customers were actually trying to do.

Which brings us back to that Tuesday. The product manager with two hundred interviews before lunch did not do anything heroic. She just asked, and got an answer in time to use it. Strella's bet is that once you have felt that, the old eight-week wait stops looking like rigor and starts looking like what it always was - a tax on curiosity that nobody needs to pay anymore.