Custom clustering dimensions for persona sets

Persona sets can now cluster on your own data — first-party traits defined on your account and properties from any event stream used to define the cohort — not just Faraday Identity Graph attributes.

Previously, clustering was restricted to a curated set of FIG attributes. That restriction keeps a persona set universally applicable: because we have those attributes on everyone, anyone can be assigned to a persona. But if you're personalizing after a first purchase, the restriction is overkill — and it can suppress model quality. Which product line someone bought first, or whether they used a discount code, is often far more useful clustering signal than their FIG demographics.

In the Dashboard, new controls under Advanced on the persona set form let you pick clustering attributes, clustering traits, and clustering stream properties. Note that persona sets built on custom traits or stream properties only apply to people present in the underlying first-party data — anyone outside it can't be assigned a persona.

Via the API, use the new clustering_dimensions property on create and update persona set. The modeling_* properties are now deprecated — if both are supplied, clustering_dimensions takes precedence. See the persona set documentation for full details.