8.2 Segmentation Methods: Streaming, Batch, Edge & Sequential Logic

8.2 Segmentation Methods and Sequential Logic

An audience definition describes who qualifies. Its evaluation method describes when and where that rule can be evaluated. Journey entry is a third choice. Keep those layers separate.

Batch evaluation

Batch supports the broadest audience rule complexity and evaluates on a scheduled job. It is appropriate for large historical windows, complex aggregates, and sequences that are not eligible for streaming or edge evaluation.

Batch is not inherently “bad” or obsolete. If the use case is a daily renewal audience based on twelve months of behavior, scheduled evaluation may be correct. Coordinate Read Audience with completion of the batch job.

Streaming evaluation

Streaming continuously evaluates audiences whose rules meet streaming eligibility as qualifying profile and event data arrives. It is useful when membership changes should trigger near-real-time action.

Not every rule is streaming-eligible. Complex historical aggregation, unsupported functions, and some sequence structures can force batch evaluation. The audience builder indicates eligibility; do not infer it from the name.

Audience Qualification is a natural journey entry for streaming membership transitions. End-to-end timing still includes ingestion, Profile, segmentation, and journey processing.

Edge evaluation

Edge evaluates eligible audiences on the Adobe Experience Platform Edge Network for same-session digital experience decisions. Edge eligibility is the most restrictive because the rule must use data and logic available at the edge.

Use it for web or mobile personalization where a request needs an audience decision during the interaction. A complex year-long purchase sequence does not become edge-capable merely because the desired page response is fast.

Sequential audience logic

Sequential logic defines an ordered pattern: event A followed by event B, optionally within a time window and with additional constraints.

Example: “Viewed Product X, then added Product X to cart within 30 minutes, and did not complete purchase.” The builder must preserve the product relationship and the order. A simple audience using “has view” AND “has add” could qualify someone whose add happened first or concerned another product.

A sequence can use time constraints such as “within one day” and event-level filters. Test:

• correct order;
• wrong order;
• event outside the time window;
• different product or entity;
• missing second event;
• repeated events.

Evaluation method depends on whether the exact sequence is eligible for streaming. If not, use batch and design journey timing accordingly.

Audience sources

Audiences can come from rule-based Segment Builder definitions, uploaded files, federated or partner sources where configured, and Audience Composition. Composition can combine, exclude, rank, split, or otherwise shape source audiences into activation groups.

An uploaded CSV audience is useful for an approved one-time or externally prepared list. Map an identity namespace carefully, validate rows, and monitor import. It does not become a streaming rule, and Read Audience incremental semantics may not work like incremental qualification of an AEP-evaluated audience.

Audience Composition can support direct-mail or controlled splits, but composition processing and publication must finish before downstream use.

Journey entry mapping

Choose identity namespace, reentrance, and schedule after selecting the audience.

Monitoring

Check audience estimate and actual membership, last evaluation status, evaluation method eligibility, source freshness, and representative profiles. For sequences, inspect event chronology rather than only the final membership flag.

If a batch audience is empty, verify source data and evaluation. If a streaming audience is unexpectedly batch, inspect rule eligibility. If an imported audience contains fewer profiles than rows, check invalid identities, duplicate values, namespace mapping, and import errors.

Decision framework

1. Define the business rule without choosing speed.
2. Determine required history and sequence.
3. Check eligibility for edge or streaming.
4. Use batch when the rule requires it.
5. Select Audience Qualification or Read Audience according to entry behavior.
6. Test membership and timing end to end.

Sequence verification

Build a five-profile sequence test: correct order within window, correct order outside window, reverse order, same event types for different products, and missing completion. Inspect event timestamps and entity keys for each result. This validates the meaning of the audience before debating whether the finished rule is eligible for streaming or must remain batch.