Profiles, Audiences, and Segmentation
The Real-Time Customer Profile merges records sharing identities into a unified profile, using a merge policy to resolve attribute conflicts (e.g., latest timestamp wins, or dataset precedence). Traps: assuming all source data is merged identically - merge policies are configurable, and the default policy may not be the one used by a given audience. Also know profile fragment vs. merged view: the profile viewer shows both.
Audiences (segments) are defined by PQL (Profile Query Language) or the visual builder. Know the evaluation types:
- Batch segmentation - scheduled, evaluates the full profile store at intervals.
- Streaming segmentation - near-real-time as data arrives; supports a subset of PQL and event-based criteria.
- Edge segmentation - evaluated at the edge for low-latency use cases, limited data scope.
A frequent trap is mixing up streaming and edge: streaming runs in the platform on streaming data; edge runs on the Edge Network for immediate personalization. Another trap: assuming a newly created audience is immediately populated - batch audiences populate on their next scheduled run, and you can trigger an evaluation.
Understand audience lifecycle: draft → published → active/inactive, and that destinations only receive audiences that are mapped and activated. Also know look-alike and account-based audiences (B2B) as distinct constructs.
Rule of thumb: match the evaluation method to the latency requirement and data availability. If the criteria depend on batch-loaded attributes, batch segmentation is required even if the audience is “real-time” in name. If criteria depend only on edge-available events/attributes, edge segmentation can serve sub-second decisions.