Every product team keeps a small graveyard of decisions that didn't work. A feature shipped to fix churn that didn't move churn. A survey that asked ten confident questions and returned ten confident guesses. The dashboards were never wrong, exactly. They showed the drop. They just never explained it. Poth Labs, a two-person company in San Francisco and part of Y Combinator's Summer 2026 batch, is built around that specific frustration.
The pitch is a single line the founders keep returning to: your analytics tell you what your users did, but they rarely tell you why. Poth - the product - is designed to close the gap between the two.
The ideaA brain, not another dashboard
Poth Labs calls its product "the customer brain for your company." The metaphor is doing real work. Most companies do not have a shortage of customer data. They have it scattered across five tools nobody has time to read: call transcripts in one place, support tickets in another, product analytics in a third, survey results in a fourth, a CRM full of half-written notes in a fifth. The signal exists. It just never sits in the same room.
Poth connects those sources into one model. Analytics from a tool like PostHog and transcripts from meeting recorders map to the same customer, so the platform can reason about a person rather than a spreadsheet row. From there it does something most tools stop short of: it forms a hypothesis about why a behavior is happening, then goes looking for evidence to confirm or kill it.
The word "brain" is a claim about memory. A dashboard forgets you the moment you close the tab; it holds no running theory of who your customers are or what you learned about them last quarter. Poth's ambition is the opposite - a model that accumulates. Every transcript, ticket, and event is another data point about the same person, and the platform is supposed to carry that context forward instead of making each team rediscover it from scratch. Whether it holds up under real, messy production data is the entire game, but the framing is clear enough to build against.
"Your analytics tell you what your users did. They rarely tell you why." Matthew Wong, Co-Founder & CEO
The behavior-versus-reason gap
There is a reason this problem has stayed unsolved. The two obvious toolsets each solve one half and ignore the other. Product analytics platforms are excellent at showing behavior at scale - they can tell you that 40 percent of new users never reach the second screen. They cannot tell you why those users left. Traditional research tools can ask why, but they ask it with static questionnaires written before anyone knew what the answer might be. You write the questions once, ship them, and hope you guessed the right ones.
Poth's answer is to make the asking dynamic. Instead of a fixed survey, its interviews adapt to each response, chasing the thread as it appears - the way a good researcher actually conducts a conversation, following the interesting answer instead of marching through a script. As evidence comes in, the platform raises or lowers its confidence in each hypothesis and generates new ones when the data points somewhere unexpected.
Anyone who has run a real user interview knows the useful moment usually arrives off-script. You ask your planned question, the person says something slightly strange, and the actual insight is in the follow-up you improvise on the spot. Static surveys are structurally incapable of that follow-up - they were printed, so to speak, before the conversation started. Poth is trying to automate the improvisation: to notice the strange answer and ask the next question a curious human would have asked. That is a harder target than summarizing feedback, and it is the part of the pitch that separates it from a nicer report.
Product, growth, and the people who sign off
The intended users are the three groups that argue about the same customer from different angles: product teams deciding what to build, growth teams deciding what to fix, and leadership deciding what to believe. Each currently reconstructs customer context by hand, and each arrives at meetings with a slightly different version of the truth. Poth's bet is that a shared, searchable model settles those arguments faster than another slide deck.
Early customers named by the company include Inkle, Sol, Composio, UnCircuit, and Illume - a set of other startups, which fits a young product finding its first believers among teams that already live inside their data. The use cases the company points to are the familiar hard ones: discovery, adoption tracking, churn analysis, win-loss analysis, and account intelligence.
Root cause, not a tidy summary
Plenty of tools will read your feedback and hand back a neat list of themes. Poth's stated distinction is that it does not stop at restating the complaint. It compares feedback against product, operational, and behavioral data to look for the underlying cause - the difference between "users say checkout is confusing" and "users abandon checkout when a specific step times out for a specific segment." One is a summary. The other is a fix.
| Tool | Tells you what | Tells you why | Adapts |
|---|---|---|---|
| Product analytics | Yes | No | No |
| Static surveys | Partial | Partial | No |
| Feedback summarizers | Partial | Restates | No |
| Poth | Yes | Yes | Yes |
"Poth connects that information into a living model of your customers, mapping analytics from Posthog and transcripts from Granola to the same customer."The founders
Two data worlds, one thesis
The founding team is small and technical. Matthew Wong, the CEO, was previously a forward-deployed engineer at Palantir, the kind of role that means sitting next to enterprise customers and watching them fail to get answers out of data they already own. Mojmir Horvath, the CTO, came from Tietoevry, a large European technology group. The two backgrounds - one American, one European, both steeped in enterprise data - converge on the same observation: organizations are rich in customer information and poor at explaining what it means.
That is the thesis Poth is built on, and it is a bet on humility as much as on technology. The product assumes you are guessing about your users more than you would like to admit, and offers to replace the guess with a tested hypothesis. For a category that has spent years shipping prettier dashboards, that reframe - from displaying data to interrogating it - is the interesting part.
There is also something telling in a forward-deployed engineer starting this company. That job is essentially professional empathy for data: you show up at a customer's office, watch where their questions die, and build the thing that would have answered them. It is a role that teaches you the gap between "the data is in there somewhere" and "the person got their answer." Poth reads like an attempt to productize that gap - to give every team the engineer-in-the-room who keeps asking follow-up questions until the reason surfaces.
The marketWhere Poth sits
Poth lands in the crowded space between product analytics and user research, next to survey tools, feedback aggregators, and the analytics giants. Its wager is that the wall between those two categories is the actual product opportunity: nobody has cleanly joined analytics-scale behavior with adaptive, hypothesis-driven interviews inside one agentic system. The business is early - two people, an initial cohort of design partners, and pricing it has not published - so the open question is less about the idea than the execution. Unifying five messy data sources into one trustworthy model is the hard, unglamorous work, and it is exactly where a "customer brain" either earns the name or doesn't.
For now, Poth Labs is worth watching for a simple reason. It picked a problem every product team recognizes on sight, and refused to answer it with another chart.