Identity Graph

Identity Graph stitches permitted identifiers from connected sources into canonical users keyed by vendo_user_id. Open Data → Data Preferences → Identity Graph.

Identity runs before other models that use the User entity, so mapping changes can affect profiles, metrics, audiences, and downstream exports.

Identification Rule

Choose the signals that make a resolved profile identified. A profile matching any enabled signal is identified; a profile matching none is anonymous in supported dashboards and audiences.

This classification is separate from graph linking. A record can participate in the graph without meeting the workspace’s identified-user rule.

Source Mappings

For each contributing app:

1. select the source table;
2. map source fields to built-in or custom identifier types;
3. choose the supported normalization for each value;
4. save and verify the mapping with sample records.

Available built-in types include email, phone, device, cookie, customer, user, click, name, and OneSignal identifiers. Use a custom type only when its domain meaning and normalization are documented.

Trust and Priority

The source trust order controls which source wins when overlapping data competes. Advanced identifier priority controls the golden-record preference for email, phone, name, and platform ID.

Changing either setting can alter resolved outputs. Record the reason and test known conflicts before saving a production change.

Safety Limit

Max identifiers per profile prevents unusually large connected components from entering the golden record. A profile over the cap is excluded rather than merged silently. Investigate the mapping that caused the expansion before raising the limit.

Safe Change Procedure

1. Capture current coverage and a set of known profile examples.
2. Change one mapping, normalization, trust, or priority rule.
3. Run or wait for identity processing.
4. Compare merges, splits, unidentified profiles, and oversized profiles.
5. Verify Customer 360 and one downstream consumer.
6. Revert or correct the rule if the examples no longer match the source of truth.

Never map two same-shaped columns together without proving that they represent the same real-world identifier.

Related

• Customer 360
• Audiences
• Data Model
• Data Quality