ComposableStack.AI CDP
Back to agent
Capabilitycapability.audience-segmentation

Audience Segmentation

The process of computing cohort membership from customer profile attributes — behavioral events, computed scores, demographic data, and consent flags. Packaged CDPs offer UI-based, no-code Segment Builder interfaces; composable CDPs implement segmentation as SQL or dbt models running in the CDW. The central architectural question is whether the segmentation engine runs inside the CDP (packaged) or inside the data warehouse (composable). Segment freshness — how quickly a new customer lands in the correct segment — is the primary latency and cost driver.

confidence 76%v1reviewed May 11, 2026audience-segmentation, segmentation, cdp-core, cdw-native, sql-segmentation, no-code, segment-builder, composable, packaged, real-time-segments, batch-segments

Audience Segmentation

Audience segmentation is the core CDP operation: given a pool of customer profiles, identify the subset that matches a defined set of criteria. Every downstream activation — email campaign, reverse-ETL push to ad platforms, next-best-action inference — begins with a segment definition.

Two architectural models.

Packaged CDP segmentation (UI-based): The CDP maintains its own profile store and provides a no-code visual interface (AEP Segment Builder, Salesforce Data 360 Data Explorer, Oracle Unity Audience Studio) for defining segment rules. Segment computation runs inside the CDP's processing layer. Data analysts and marketing operations teams build segments without SQL knowledge. The tradeoff: segmentation logic runs only on data inside the packaged CDP's profile store — data not ingested into the CDP cannot be segmented.

Composable CDP segmentation (CDW-native): Segments are defined as SQL queries or dbt models running on the cloud data warehouse. Any data in the CDW (from any source, not limited to the CDP's ingestion connectors) is segmentable. Engineering-driven: requires a data engineering team comfortable with SQL; no-code UI does not exist natively (Hightouch and Fivetran Activations provide audience builder UIs on top of CDW SQL, partially closing this gap). Segment membership is computed on a schedule (typically hourly to daily batch) or triggered on event arrival.

Segment freshness. Segment freshness — the latency between a customer taking an action and appearing in the correct segment — is the primary latency and cost variable:

Operational considerations.

Overlap management: Customers can qualify for multiple segments simultaneously. CDP architectures must define priority or exclusion rules (no-overlap, ranked priority, first-match) to avoid conflicting activations (e.g., win-back email and VIP promotional email delivered on the same day to the same customer).

Suppression integration: Segments must be evaluated against suppression lists (opt-outs, DNC, recently-converted customers) before activation. Suppression is a first-class operation in every segmentation architecture; it is not an afterthought.

Consent-gating: In regulated industries (healthcare, financial services), segment membership may need to be filtered by consent status. HIPAA-covered organizations may only include profiles with appropriate consent or de-identified status in paid-media activation segments.

Vendors providing this capability. Composable: vendor.hightouch (SQL + UI audience builder), vendor.census (now Fivetran Activations — SQL + dbt). Packaged: vendor.adobe-experience-platform (AEP Segment Builder, real-time streaming segments supported via Adobe Streaming Segmentation), vendor.salesforce-data-360 (Data Explorer + AI-computed Einstein Segments), vendor.oracle (Oracle Unity Audience Studio). Emerging CDW-native: vendor.databricks (Delta Live Tables + Databricks SQL for audience computation; activation via Lakebase reverse-ETL).

Sources

Related

This node →

  • prerequisite-ofcapability.reverse-etl— Audience segmentation produces the cohort membership list that reverse-ETL delivers to downstream activation systems. Segmentation is the computation; reverse-ETL is the delivery. Segmentation is a logical prerequisite of reverse-ETL activation.
  • requirescapability.identity-resolution— Effective audience segmentation requires a resolved profile as its input: segment criteria must evaluate a single unified customer view, not disconnected event streams from different devices or channels. Identity resolution quality directly bounds segmentation accuracy.
  • applies-to-latency-tierlatency-tier.intra-session— Real-time streaming segmentation (stream processor evaluating each event on arrival) achieves intra-session segment membership latency — the customer lands in the correct segment within seconds to minutes of the triggering event.
  • applies-to-latency-tierlatency-tier.inter-session— Micro-batch segmentation (CDW SQL query running every 15–60 minutes) achieves inter-session segment membership latency — the standard freshness floor for composable CDP architectures without a dedicated stream processor.

← Referenced by

  • prerequisite-oftech-dim.dev-team.backend-data-engineering— Composable CDP audience segmentation is implemented as SQL queries or dbt models in the CDW — authored and maintained by the data engineering team. The absence of this team in composable architectures means segmentation must stay in the packaged CDP UI.