Data Sources

What is a Data Source?

A Data Source is a read-only connection from GrowthBook to your warehouse or database. GrowthBook uses this connection to run analysis queries on demand — it never writes data back. Supported engines include BigQuery, Snowflake, Postgres, MySQL, ClickHouse, Redshift, Databricks, Athena, PrestoDB, and more.

Connecting a data source

Step 1 — Open Data Sources

1. Log in to GrowthBook.
2. In the left sidebar, navigate to Analysis then click Data Sources.
3. Click Add Data Source.

Step 2 — Choose a connection type

1. Select your warehouse type (e.g. BigQuery, Snowflake, or Postgres).
2. Enter the connection details your warehouse requires:
   • BigQuery: paste a service account JSON key and choose the default project.
   • Snowflake: enter account, warehouse, database, schema, username, and password.
   • Postgres / MySQL: enter host, port, database name, username, and password.
3. Click Test Connection to verify GrowthBook can reach your warehouse.
4. Click Save.

Step 3 — Write the experiment assignment query

The assignment query is the most important query in GrowthBook. It answers the question: “Which user saw which experiment variation, and when?”

A typical assignment query selects four columns:

Column	What it contains
experiment_id	The GrowthBook experiment key (e.g. checkout-redesign)
variation_id	The variation index the user was assigned to (0 = control, 1 = variation A, …)
user_id	A stable user identifier (user UUID, anonymous ID, etc.)
timestamp	When the assignment happened (UTC)

Example for a BigQuery events table:

SELECT
  experiment_id,
  variation_id,
  user_id,
  timestamp
FROM
  `myproject.analytics.experiment_viewed`
WHERE
  timestamp >= '{{ startDate }}'
  AND timestamp <= '{{ endDate }}'

GrowthBook substitutes {{ startDate }} and {{ endDate }} automatically when it runs the analysis.

Step 4 — (Optional) Add identifier joins

If your warehouse uses different ID types (anonymous cookie ID vs logged-in user ID), add Identifier Joins so GrowthBook can stitch them together across the assignment and metric tables.

No warehouse? No problem

If you are just getting started and don’t have a data warehouse, GrowthBook can work without one for basic feature flags and targeting. You can add a data source later when you are ready to run statistical experiments. GrowthBook also supports uploading CSV data for lightweight experimentation.

🧪 The assignment query is the foundation

The experiment assignment query tells GrowthBook who saw which variation and when. Every metric and statistical result flows from this query — if it is wrong or incomplete, all downstream analysis will be wrong. Verify it with a direct SQL preview before saving.

What does GrowthBook do with the data in your warehouse?

It copies events into its own database for faster queries It runs read-only SQL queries against your warehouse on demand It streams events in real time via a webhook It stores raw events in GrowthBook Cloud

Which four columns must the experiment assignment query return?

user_id, feature_key, value, timestamp experiment_id, variation_id, user_id, timestamp session_id, event_name, properties, timestamp cohort_id, bucket, user_id, created_at

What are {{ startDate }} and {{ endDate }} in the assignment SQL?

JavaScript template literals that must be escaped GrowthBook template variables substituted automatically at query time SQL comments that are stripped before execution Column names in your events table

What should you do before saving a new Data Source connection?

Create at least one metric Click Test Connection to verify GrowthBook can reach the warehouse Run a full experiment first Export the schema to a CSV

Submit

Mark as complete