Mitzu - Agentic product analytics that runs on your data warehouse
by•
Mitzu is an agentic product analytics platform that runs on your data warehouse — built to answer the why, not just the what. Ask complex questions like "why did week-2 retention drop?" and the Analytics Agent investigates across funnels, cohorts, and segments, returning a synthesised answer. No hallucinated SQL: a deterministic engine produces methodology-correct queries every time. Your data stays in Snowflake, BigQuery, Databricks, Redshift, or ClickHouse.
Replies
Mitzu
Hey Product Hunt! 👋
I'm Istvan, founder of Mitzu. We built this because we kept watching the same frustrating loop happen:
A data analyst fields a Slack message: "Why did activation drop last week?" ➡️ Opens a notebook, writes three SQL queries, finds a partial answer, and posts it two hours later.
➡️ The PM asks a follow-up. ➡️ The cycle repeats.
The problem isn't the analyst. The problem is that answering diagnostic product questions (not just "What was DAU?" but "Why did week-2 retention drop in November?") requires two things that haven't lived together until now:
Product analytics methodology (funnels with conversion windows, cohort-bucketed retention, session-level journey maps).
Access to all the data in your warehouse (billing, CRM, support tickets).
Enter Mitzu 🚀
Mitzu is an agentic product analytics platform that runs directly on your data warehouse. Here is what that means in practice:
Zero AI Hallucinations 🧠 — The agent doesn't write raw SQL. Instead, it assembles a precise analysis specification, and our deterministic query engine turns that into SQL using hardened product analytics methodology. Same question, same SQL, every time. No methodology errors slipping through.
Automatic Setup ⚡ — Our Configuration Agent scans your warehouse, identifies your event tables, and builds a semantic layer specialized for product analytics. No YAML, no manual hand-mapping.
Your Data Never Leaves 🔒 — No ingestion, no per-event pricing, and no data silos. Because it runs natively, Mitzu can effortlessly join product events to billing tables or CRM data in the exact same query.
⚠️ One Honest Qualifier
Mitzu requires a modern cloud data warehouse (Snowflake, BigQuery, Databricks, Redshift, or ClickHouse) with event data already sitting in it. If you're still figuring out your data stack, we're not the right fit just yet.
We'd love for you to check it out! I'm happy to answer any questions below about how the deterministic engine works, our semantic layer, the Slack and MCP integrations, or anything else on your mind.
— Istvan & the Mitzu team
ROAARRR
This is sick! Don't all of these warehouses have AI analysis built in? ie snowflake cortex? What advantage does Mitzu have over the in-house solutions?
@nick_barth Cortex and similar in-warehouse AI features are mostly text-to-SQL — the LLM writes SQL against your tables, optionally grounded in a YAML semantic model you author by hand.
Mitzu runs on the warehouse too but is shaped differently in two ways:
The agent doesn't write SQL. It assembles analysis specs — funnel steps, retention windows, cohort definitions — and a deterministic engine turns those into SQL. Same spec, same SQL, every time.
The semantic layer is built automatically by scanning your warehouse, and it's specialized for product analytics (events, properties, entities, sampled filter values). No hand-authored YAML.
Where this matters: ask Cortex "why did week-2 retention drop in November?" and the LLM authors the query — cohort bucketing, conversion window, group-by, all on the model to get right. In Mitzu the methodology lives in the engine, not in the model's head.
Cortex is solid for ad-hoc SQL on warehouse data. Mitzu is built for the diagnostic product analytics questions where methodology has to be right.
MindOne
I've been waiting for this a while now! Congrats!