Blog

Chart the dbt metrics your data team already models, such as MRR and revenue booked, in Mitzu next to your product data. Roll-up rules and setup, step by step.

Experiments is new in Mitzu. Your experimentation tool or your own code keeps assigning users to variants; Mitzu reads the exposure event it leaves in your data warehouse, measures every variant with the saved insights your dashboards already use, checks that the test itself is healthy, and records the decision. This guide covers why we built it, how it works, and a step-by-step walkthrough from the first exposure event to a finalized decision.

What agent analytics is, how it differs from LLM observability and evals, which AI agent metrics to track, and an event model for agent sessions.

Analyze A/B tests from Optimizely, VWO or LaunchDarkly on your warehouse: the exposure table, metric attribution, SQL, lift, intervals and health checks.

What guardrail metrics are, examples by type, how to pick two or three per A/B test, set thresholds before launch, and read a result that is not significant.

What sample ratio mismatch (SRM) is, why it invalidates A/B test results, its common causes, and how to detect it with a chi-square test in SQL.

What a metric tree (KPI tree) is, SaaS and e-commerce examples, how to build one in five steps, and how to walk it to find which input moved a KPI.

What a semantic layer is, how it works, the main types and tools (dbt, Cube, AtScale, Looker, Databricks, Mitzu), and a worked metric example with its SQL.

Databricks AI/BI Dashboards and Mitzu both build KPI dashboards on your Databricks warehouse. Compare governed metrics, the AI, self-service and the limitations of a Databricks dashboard, and see what Mitzu adds: product analytics, experiment analysis and an agent that explains why a number moved.

A/B testing statistics in plain English: p-values, significance, z-test vs t-test, confidence intervals, sample size and power, with worked examples.

If your events already land in Snowflake, BigQuery, Databricks, or Redshift, migrating off Mixpanel does not mean re-instrumenting your product. Here is the concrete path: what changes, what stays, and what to do with historical events.

An analytics agent answers data questions by running multi-step analysis on your warehouse. See how agents work, the main types, and what makes one trustworthy.

Text-to-SQL (NL2SQL) tools and BI chatbots have a language model write the query; analytics agents run investigations on a deterministic engine. Where text-to-SQL works, where it breaks, and a checklist for evaluating NL-to-SQL tools.

Analytics agents are only as trustworthy as the context they run on. Why a product-analytics semantic layer prevents hallucinated SQL and methodology errors.

A structured framework for enterprise data and analytics leaders assessing AI analytics agents: methodology trust, warehouse governance, semantic depth, diagnostic depth, security, and governance of permissions, shared definitions and audit, with a buyer checklist.