Skip to main content
Adrià García
Menu

Architecture guide

What needs to leave the warehouse?

Keeping orders in the warehouse does not decide the architecture. First establish whether you need the detail, an audience or an updated copy.

Sources checked:

Three data paths

  1. Ingest

    1. Source data
    2. Dataset in AEP
    3. Configured use

    The required data is copied. Ingesting it into the data lake does not automatically add it to Real-Time Customer Profile.

  2. Federate with FAC

    1. Query in the warehouse
    2. Audience or attributes
    3. Use in AEP

    Detailed data is queried at source. Results needed to create or enrich audiences or profiles can still enter AEP.

  3. Synchronise with Data Mirror

    1. Inserts, updates and deletes
    2. Relational schemas
    3. Data maintained in AEP

    Source changes are transferred. Data persists in AEP; this is neither a federated query nor a zero-copy promise.

These options address different needs; they are not a ranking. A design can combine ingestion and federation with explicit scopes.

What Adobe documents

Ingestion and Profile have separate configuration

Schemas describe ingested data. Real-Time Customer Profile uses Profile-enabled data; not every data lake table belongs in the profile.

Adobe: XDM and Platform services

FAC reduces movement of underlying data

Federated Audience Composition builds and enriches audiences by querying external databases. Avoiding a copy of underlying data does not eliminate composition results or warehouse work.

Adobe: Federated Audience Composition

Data Mirror requires a compatible model and entitlement

It synchronises changes through relational schemas and record keys. Availability includes AJO Orchestrated campaigns; CJA access is a limited release, subject to licence and enablement. Confirm access before designing around it.

Adobe: Data Mirror overview

How I would review it

  • Define who needs the data and when. A scheduled campaign and an in-visit response have different freshness requirements.
  • Review query, transfer and storage costs alongside licensing. Identify who operates each connection and handles failures.
  • Test corrections and deletes, not only inserts. Follow their effects through audiences, copies and consuming applications.

An architectural comparison based on public documentation. It does not describe my own FAC or Data Mirror implementation, or guarantee latency, cost or availability for every contract.

Explore through examples

Related services