On a Monday morning, someone at Sweetgreen asked a question. By Thursday, the restaurant chain's insights team had an answer from customers. This was not the usual corporate miracle in which a meeting produces another meeting. Brian Davia, Sweetgreen's head of consumer and business insights, said his team had run a study with Listen Labs and returned with data and insight before the week was out. The answer arrived while there was still time to use it.
The short version
- Listen Labs recruits people, interviews them with AI and turns their responses into reports, clips and searchable evidence.
- Its customers include Sweetgreen, Microsoft and Anthropic; the company reported more than one million interviews by January 2026.
- The key move is the follow-up question: software can probe a vague answer in thousands of individual conversations.
- The economics depend on whom you need to reach, what they must be paid, and how carefully the findings are checked.
The scene explains the business better than the phrase “AI research platform” ever could. Research is valuable when it can change a decision. The usual machinery - writing a guide, recruiting people, finding a time, moderating, transcribing, coding, presenting - often delivers its verdict after the decision has been made. Listen Labs compresses those steps into a single workflow. A team supplies a question; the software helps shape a study, finds participants, conducts adaptive interviews and produces themes with the underlying recordings attached.
The failed product inside the successful one
The founders, Alfred Wahlforss and Florian Juengermann, had built an AI avatar app before this. Juengermann has described the discovery plainly: a working product did not guarantee that they understood why anyone would want it. His account says his first three startups failed. The avatar project made a particular failure visible. Their problem was not only software engineering; it was getting close enough to users to hear the difference between a polite reaction and a real need.
That is why Listen Labs is arranged around a conversation rather than a form. A survey can efficiently collect an answer such as “fine.” An interviewer asks what “fine” means, what happened just before the feeling, and what the person did next. Listen's AI moderator is designed to make that second move across many interviews at once. It can show a concept, ask about a reaction, follow an interesting detail, and leave behind video, audio or text that a human researcher can inspect.
The company says its participant network reaches 30 million prequalified people. That number is capacity, not a guarantee that every niche buyer is instantly available. Listen also offers controls for suspicious or low-effort responses, a necessary feature in a market where a respondent may be paid for participating. The final output includes summaries, themes, clips and presentation materials. The practical advantage is not merely that a machine writes a report. It is that the researcher can get from a question to inspectable evidence in a shorter cycle.
A salad chain offers the better demonstration
Sweetgreen had a problem familiar to any company with many locations: what works in one market may fail in another. Davia said traditional methods forced a choice between quick, thin data and slower, richer interviews. Listen let his team ask more people while keeping the texture of individual responses. According to the company's case study, research informed the chain's wraps, a macro-tracking feature and a 100-gram protein bowl. Sweetgreen tested across more than 300 locations and said it could do more research without a larger budget.
“I have sat in a Monday morning meeting with an executive team and been posed a question, and by Thursday morning I have data back.”Brian Davia / Sweetgreen
The case also supplies the cost question with some useful restraint. Sweetgreen's published account says it obtained five times as many responses for about half the research cost on the studies it ran. Those are its own figures, not a universal price list. Listen's public guidance describes a credit model beginning around $25 per interview, while enterprise terms, difficult audiences and participant incentives change the bill. Traditional qualitative studies can involve much larger recruiting and moderation costs. The savings arrive when a team can replace several handoffs with one workflow - and when the sample is credible enough to support the decision.
Microsoft used the platform for more than 150 multilingual interviews in a project gathering customer stories for its fiftieth anniversary. Anthropic describes a different benefit: using interviews and quantitative evidence in one program instead of running separate studies in sequence. These examples explain why Listen sells to product teams, marketers and professional researchers alike. One wants to decide what to build; another needs to test a message; a third must defend the quality of the work.
The number and the reason
In 2026, Listen added Gabor-Granger pricing tests. The method presents buyers with different prices to estimate a demand curve. Listen attaches the buyer's own explanation to each point. If someone accepts $30 but rejects $40, the product can ask what changed. Was it a budget limit, a missing feature, or the belief that a rival product is good enough? A chart can say where the curve bends. A conversation can tell a team which lever might move it.
That combination is the company's clearest place in the market. Qualtrics and survey tools excel at structured answers. Dovetail and other repositories organize research that has already happened. AI interview rivals such as Outset and Strella also automate conversations. Listen's pitch is the whole trip: study design, recruitment, moderation, quality checks, analysis and deliverables in one place. It is a useful distinction for teams that have been stitching together agencies, panels, video calls and spreadsheets. It is less decisive when a company already has a strong panel, specialist moderators, or a study whose small sample requires careful human judgment.
Research, after all, can move quickly and still be wrong. A sample can miss the relevant customer, an incentive can attract the wrong one, and a neat theme can flatten a dissenting voice. The remedy is visible in Listen's better product choices: clips attached to claims, screening before interviews, and a researcher who checks whether the conclusion survives the raw material. Automation buys time for judgment. It does not absolve anyone of it.
A company that markets by asking for a reaction
Listen has also used unusually theatrical ways to attract attention. A cryptic San Francisco billboard, reportedly bought for about $5,000, hid a coding challenge based on the notoriously selective Berlin nightclub Berghain. More than 400 people solved it, according to reporting on the campaign. In 2026 the company said it spent $1 million to place ads on more than 1,500 New York screens. One campaign tested engineers' curiosity; the other played with the gap between what people say and what they mean. Both put the audience to work.
The company announced a $69 million Series B in January 2026, after reporting more than a million interviews, hundreds of enterprise customers and eight-figure annualized revenue. Those are strong signs of demand, but the lesson is smaller and more portable than the funding round. Give a decision a deadline. Ask a real customer about a recent behavior, not a hypothetical preference. Probe the first answer. Then keep the evidence close enough that someone can challenge the summary. Listen Labs made a business of that sequence. Any team can borrow the habit.