Overview
Subsets can push a daily snapshot of your experimentation data into a cloud storage bucket that you own. Files land as gzipped CSV in a Hive-partitioned layout that any warehouse or query engine reads directly - Spark, Trino, Snowflake, BigQuery, Athena, Redshift Spectrum, Databricks, DuckDB, pandas.
Three feeds are available, enabled independently:
| Feed | Contents | Grain |
|---|---|---|
experiment-participants | Every participant of every non-draft, non-archived experiment - treatment, control and suppression variants included | experiment × customer |
audience-memberships | Every audience and every customer assigned to it | audience × customer |
subscription-risk | Churn-risk score per customer, 0-1, highest across their scored subscriptions | customer |
Every row is keyed by customer_id - your customer's id in your CRM integration - so the files join directly onto data you already hold. The data contract specifies the exact layout, columns and consumption rules; a _schema/SCHEMA.md file delivered into your bucket documents your specific integrations and table definitions.
Each feed delivers one file per CRM integration per UTC day - a full snapshot, not a diff - with at most 24 hours between snapshots. Access is keyless in both directions: no credentials are exchanged, stored or rotated. Subsets only ever writes to your bucket and never deletes; your own lifecycle rules decide how long history survives.
Setting it up
Setup is self-serve in Settings → Data feeds: connect a bucket (one destination per organization), pass a one-time validation test, and enable the feeds you want. Per-feed delivery health is visible in the same tab.
Amazon S3 is the supported destination. Settings → Data feeds is the setup guide; Amazon S3 destination lists what it asks for and the security details for your cloud team.