AEP-as-Edge-Node

Problem. An organization is heavily invested in AEP — licenses, integrations, internal training — but its data strategy is moving toward a composable stack centered on a cloud data warehouse. The AEP guardrails (constraint.aep-first-gen-export-500kb, constraint.aep-second-gen-export-1500kb, constraint.aep-adhoc-query-timeout-10min) actively push back against using AEP as the upstream source of truth.

Strategic shift. From "AEP as the single source of truth" to "AEP as a downstream edge node."

Implementation.

• Do the heavy lifting in the cloud data warehouse first: complex identity resolution, massive JOINs, feature engineering.
• Feed AEP only the highly aggregated, lightweight profile traits and pre-computed audiences it needs for activation.
• Avoid sending raw event streams to AEP — they consume export quotas without yielding proportional value.

Why it works. It honors the existing AEP investment (activations, destinations, integrations) while moving the data-intensive work to where compute is unconstrained. Egress quotas become non-binding because the data flowing into AEP is already aggregated.

Tradeoff. Identity resolution effectively moves out of AEP — for organizations relying on AEP's identity stitching, that's a non-trivial replacement. Reverse direction: it concentrates a critical capability in the CDW, increasing the data team's responsibility (source.packaged-vs-composable-md).

Alternative pattern. If the organization is willing to abandon CDW-centricity, "AEP as source of truth" remains viable for use cases that fit within the export quotas — but the agent should surface this as the explicit alternative, not the default.