AI analytics for BigQuery, with every query in your own project
Mitzu is an AI analytics platform that runs on Google BigQuery. Ask about the customer journey, from GA4 traffic and Firebase app events to signups, orders and revenue, and Mitzu answers with a query that runs as a job in your own Google Cloud project, with the SQL shown. Your data stays in BigQuery, and the SQL comes from a deterministic engine, not from the language model.
Your data team defines the metrics once on its BigQuery tables. KPI dashboards, funnels, retention, A/B test results and answers in Slack or your own AI assistant all use those definitions.
How zero-copy AI analytics works on BigQuery
Someone asks a question
In the Mitzu app, by mentioning @mitzu in Slack, or from Claude, ChatGPT, Cursor or another assistant through Mitzu's MCP server. For example: “Which traffic sources bring users who buy twice within 30 days?”The agent builds an analysis specification
It looks up events, metrics and properties in Mitzu's semantic layer, which is indexed from your BigQuery datasets, and fills in the analysis: steps, conversion window, breakdowns and date range.The engine writes the SQL, BigQuery runs the job
Mitzu's deterministic query engine compiles the specification into BigQuery SQL. The job runs in your project under the Mitzu service account, on the pricing model your project already uses.Aggregates come back, with the query attached
Mitzu receives counts, rates and time series, not event rows. The agent explains the result as a chart and a summary, and the exact SQL is one click away.
Stays in BigQuery
- GA4 and Firebase export tables, nested event parameters included
- Orders, subscriptions, revenue and CRM tables
- dbt models and scheduled transformations
- Row-level access policies and column-level masking, enforced by BigQuery
Stored by Mitzu
- Semantic-layer metadata: event, property and table names, plus sampled filter values (at most 500 per property)
- Saved insights, dashboards and cohort definitions, stored as definitions, not user lists
- Cached aggregated results, so dashboards open without new BigQuery jobs
Verified SQL you can find in your own job history
When a language model writes BigQuery SQL from scratch, the query runs and the number looks reasonable, but the method can be wrong. A funnel that ignores step order or time window, a retention chart that mixes cohorts, or GA4 sessions counted twice through an unnested parameter all produce plausible results.
Mitzu's agent never writes SQL. It fills in a typed analysis specification, and the same engine behind Mitzu's insights page compiles it, with funnel, retention, journey and experiment methodology implemented once. Ask the same question next week and you get the same SQL.
Every query is open to review in three places: Show SQL on any chart, the Query Admin tab with BigQuery-reported bytes processed and run time, and your project's own job history, where each job appears under the Mitzu service account. See how the agent works for the full division of labour.
An analysis specification, as the agent builds it
- analysis
- retention
- cohort event
- first_open
- return event
- purchase
- granularity
- weekly, 8 weeks
- breakdown
- app_info.version
Questions BigQuery teams ask, across the customer journey
Acquisition
“Which GA4 traffic sources bring users who purchase again within 30 days?”
Mobile
“Which app version has the worst week-1 retention on Android?”
Revenue
“Orders fell on Tuesday. Was it traffic, checkout conversion or one market?”
Product
“Do accounts that invite a teammate in week one convert to paid more often?”
Experiments
“Did the new paywall variant lift trial starts without raising refunds?”
Monitoring
“Each morning, email the marketing team if paid-social signups drop more than 20% day over day.”
What setup on BigQuery involves
Create a service account in Google Cloud
Give it four roles: BigQuery User, BigQuery Data Viewer, BigQuery Job User and BigQuery Read Session User, then create a JSON key. Mitzu needs no write access to your data.Enter the connection in Mitzu
Set the project ID, then paste or upload the JSON key. Mitzu tests the connection withSELECT 1. Step-by-step: BigQuery connection guide.Let the Configuration Agent map your datasets
It recognizes GA4 and Firebase export tables, Segment and Snowplow layouts and your own event tables, maps user identifiers and timestamps, and handles nested STRUCT, ARRAY and JSON columns. Indexing then samples property values for filters.Review metrics and roles, then invite the team
Mitzu drafts metric definitions such as orders, average order value or active users from your tables and dbt models; the data team reviews and owns them, and sets Mitzu roles for who sees what. Then connect Slack and MCP clients.
One configured event table is enough for a first funnel or retention chart, and a GA4 export usually is that table. If you are starting from GA4, read using GA4 with BigQuery for product analytics.
Cost, security and governance on BigQuery
Billing you already understand
Mitzu's jobs are billed to your project under its existing on-demand or capacity pricing, like any other BigQuery workload. Mitzu does not resell compute.
Fewer bytes scanned
Cached aggregated results and dashboard refresh windows avoid re-running fresh queries, and sampling and time-resolution settings bound how much data each job reads.
BigQuery policies apply
Mitzu queries as its service account, so row-level access policies, policy-tag masking and dataset-level grants on that account apply to every chart and agent answer.
Roles on every surface
Mitzu roles decide what each person sees in dashboards, the agent, Slack, MCP clients and scheduled emails. Nobody needs their own Google Cloud access.
What the AI model sees
Catalog metadata and aggregated results only. Raw event rows never leave BigQuery for the model, and your data is never used to train models.
Seat pricing, optional self-hosting
Mitzu is priced per seat, with no per-event or per-query metering. Teams with strict isolation needs can run a private Mitzu instance in their own cloud.
Mitzu vs BigQuery Conversational Analytics vs GA4 reports
| Capability | Mitzu on BigQuery | BigQuery Conversational Analytics (Gemini) | GA4 reports and explorations |
|---|---|---|---|
| Data it can reach | Any table in your BigQuery projects, queried in place | Any table in BigQuery | The GA4 property, plus limited data imports |
| Who writes the SQL | A deterministic engine, from the agent's analysis specification | Gemini, for each question, guided by instructions and verified queries | No SQL; GA4's own reporting engine |
| Where definitions come from | Drafted by Mitzu from your tables and dbt models, owned by the data team | Table annotations, instructions, verified queries, or LookML maintained by the team | GA4's built-in metrics plus custom definitions |
| Funnels, retention, journeys | Built-in methodology in the engine | Composed by the model per question | Funnel and cohort explorations on GA4 data |
| Joins to orders, billing and CRM | Native: they're in the same warehouse | Yes, for tables in BigQuery | No, beyond what GA4 imports support |
| A/B test analysis | Lift, 95% interval, corrected p-values and guardrails on exposures in BigQuery | No built-in method; the model writes any statistics | No built-in test analysis |
| History and limits | Everything in BigQuery; sampling only if you turn it on | Everything in BigQuery | Explorations limited by the property's data retention; thresholds can hide rows |
| Pricing | Per seat; query compute on your BigQuery bill | Google Cloud usage pricing | Free, or GA4 360 for enterprise limits |
| Where people use it | Mitzu app, Slack, MCP clients, monitoring-agent emails | BigQuery Studio, Looker Studio Pro, Gemini Enterprise, the API | The GA4 web interface |
Conversational Analytics is a good fit for free-form questions across any BigQuery data with deep Google Cloud integration; GA4 stays the simplest view of web traffic. For the full comparison with SQL examples, read BigQuery AI vs Mitzu.
AI analytics for BigQuery FAQ
Does Mitzu copy data out of BigQuery?
No. Mitzu runs query jobs in your project and receives aggregated results such as counts, rates and time series. Event rows are never exported or stored outside BigQuery. Mitzu stores only semantic-layer metadata, saved definitions and cached aggregates.
Does the AI write the SQL that runs in BigQuery?
No. The analytics agent builds an analysis specification and Mitzu's deterministic query engine compiles it into BigQuery SQL. The same specification always produces the same SQL, which you can open on any chart or find in your project's job history.
What permissions does the service account need?
Four roles: BigQuery User, BigQuery Data Viewer, BigQuery Job User and BigQuery Read Session User. You connect with a JSON key for that service account and your project ID. Mitzu asks for no write access to your data.
Who pays for the queries, and how do we limit bytes scanned?
Jobs are billed to your Google Cloud project under its usual on-demand or capacity pricing. Cached results, dashboard refresh windows, and sampling and time-resolution settings limit what each job reads, and Query Admin shows the bytes processed by every query. Mitzu itself is priced per seat.
Does Mitzu work with the GA4 and Firebase BigQuery exports?
Yes. The Configuration Agent recognizes GA4 and Firebase export tables and maps their events, user identifiers and nested event parameters, so funnels and retention work on the export as it lands, without a modeling project first.
Can we join GA4 events with orders, subscriptions or CRM data?
Yes, when those tables are in BigQuery. Because Mitzu queries the warehouse directly, a funnel can start at a GA4 session and end at an order in your billing tables, broken down by a CRM field.
Do BigQuery row-level and column-level policies apply?
Yes, for the service account Mitzu connects with. Row-level access policies, policy-tag masking and dataset grants are enforced by BigQuery when the job runs. Inside Mitzu, roles control what each person can see on every surface.
How is Mitzu different from Gemini and Conversational Analytics in BigQuery?
Conversational Analytics has Gemini write SQL for each question, guided by instructions and verified queries your team maintains. Mitzu's agent fills in analysis specifications that a deterministic engine compiles, with funnel, retention and experiment methodology built in, and adds KPI dashboards, A/B test analysis and monitoring agents on the same definitions.
What does the AI model see from BigQuery?
Catalog metadata, such as event and property names, and aggregated query results. Raw event rows never leave BigQuery for the model, and your data is never used to train language models.
Can Mitzu analyze A/B tests from BigQuery?
Yes. If your experimentation tool or in-house assignment writes exposures to BigQuery, Mitzu reports lift, a 95% interval, corrected p-values, guardrail metrics and health checks on the same metrics as your dashboards.
How long does setup take?
Creating the service account and connecting takes minutes, and a GA4 or Firebase export is enough for a first insight. Reviewing the drafted metric definitions and roles is the part worth the data team's time.
Can Mitzu run inside our own cloud?
Yes. Besides Mitzu Cloud, the Mitzu team can deploy a private instance into your own cloud environment, which can be isolated from the internet. The architecture is the same: jobs run in BigQuery and data stays there.
Try Mitzu on your BigQuery data
Connect a read-only service account, ask your first question, and check the SQL behind the answer.