A New York company teaching machines to answer the questions a spreadsheet never could - and to prove the answer before you act on it.
Somewhere inside a retailer you have heard of, a merchandising manager types a question into a chat box: why did margins dip in the Northeast last week? She is not a data analyst. She has never written a line of SQL. Thirty seconds later she has an answer, a chart, and a list of the exact tables and filters that produced it. No queue. No "the data team will get back to you." That small, unremarkable moment is the entire point of Zenlytic.
The product is named Zoe, an AI data analyst that lives on top of a company's data warehouse. Zenlytic is careful about the noun: Zoe is an analyst, not a copilot. A copilot waits to be asked. Zoe asks its own follow-up questions, digs deeper, and hands back work you can check.
For two decades the industry sold a tidy promise: dashboards for all, insight on demand. The reality was less photogenic. Dashboards multiplied. So did the questions they couldn't answer. Anything off the beaten path went back to a small, overworked data team, and the rest of the company learned to wait - or to guess.
Then large language models arrived and a hundred startups bolted "text-to-SQL" onto a chat window. Type a question, get a query, get a number. The trouble is that a confident wrong number is worse than no number at all. In analytics, a hallucination isn't a quirky anecdote - it's a revenue figure someone takes into a board meeting.
Zenlytic was started by Ryan Janssen and Paul Blankley. Janssen, the CEO, came from the venture and engineering side; Blankley, the CTO, holds a master's in data science from Harvard and, in the gaps between shipping code, can be found snowboarding and rock climbing around Denver. They had built and sold data products before. They had also watched plenty of them collect dust.
Their bet was that the answer was not LLMs alone, and not the old semantic layer alone, but both - wired together. The language model brings comprehension: it understands a messy human question. The semantic layer brings consistency: it knows what "revenue" actually means at your company, every single time. One without the other is either eloquent and wrong, or correct and unusable.
The assistant is named Zoe after the Sesame Street character - a small joke that the original large language models were the ones that taught us all to talk. It is the rare AI product whose name admits it learned from a puppet.
Connect a warehouse - Snowflake, BigQuery, Databricks, Postgres - and Zoe goes to work. It reads the tables, builds a semantic model, and starts answering questions in plain language. Every response arrives with its receipts: the sources, the filters, the lineage, all visible to someone who has never opened a database.
Citations, filters and lineage attached to every result, so non-technical users can trust - and audit - the number.
Zoe Self-Learning connects, finds the right tables and builds the semantic layer in under an hour. No months-long setup.
Beyond charts, Zoe generates presentations, Word docs and Excel models grounded in your business logic.
Ask questions from Slack, Microsoft Teams or email. Governance is versioned and reviewed like code, via Git.
Founded in New York. Janssen and Blankley set out to build BI that people would actually use.
$5.4M seed. Bain Capital Ventures and Primary Venture Partners back an early bet on unifying BI and product analytics for commerce brands.
$9M Series A, led by M13. Bain Capital Ventures, Primary, Company Ventures, Correlation Ventures and 14 Peaks join to scale Zoe.
Zoe Self-Learning ships. The AI analyst onboards itself onto a warehouse in under 60 minutes, no data engineering required.
A pitch about trustworthy analytics should, ideally, come with evidence. Zenlytic's shows up in two forms: the logos that adopted it, and the numbers it reports against the alternatives.
Bars scaled for drama, not for a peer-reviewed journal - figures are Zenlytic's own, against text-to-SQL tools. Read them the way you'd read any founder holding a microphone: interested, but with one eyebrow raised.
There is the matter of compliance, too, which is unglamorous right up until it isn't: SOC 2 Type II, GDPR and HIPAA, plus role-based access control. The boring stuff that lets a Verizon say yes.
The mission Zenlytic repeats is simple to say and hard to do: make analytics usable by everyone, technical or not, without lowering the bar on whether the numbers are right. The investor case is the same sentence read from the other side - the company that makes self-serve BI trustworthy gets to sit between every business question and the warehouse that answers it.
If agents like Zoe deliver on even half the promise, the data analyst's role shifts from query factory to something closer to editor - defining what the metrics mean, reviewing the semantic layer like code, and trusting the agent to handle the long tail of "quick questions" that used to eat the week. That is either a threat or a relief, depending on how much you enjoyed being a human SELECT statement.
Back to that Tuesday morning. The merchandising manager already has her answer and has moved on to acting on it. The data team never saw the question, because it never needed to. The backlog that defined a generation of BI is, for that one query, simply gone. Multiply it by a few hundred questions a week, and you have the change Zenlytic is betting the company on.