Platform selection

Market convergence diagram. At the center, the warehouse as source of truth, with Snowflake, BigQuery, Databricks and Fabric. From above, the suites connect, each through its own bridge: Adobe through Federated Audience Composition, Salesforce through its zero-copy partner network, Oracle through Fusion Unity and APIs, and SAP through Business Data Cloud. From below, the agile architectures connect: Tealium through CloudStream, HubSpot through connectors, and composable through reverse ETL. Every route ends at the same warehouse.

Decision tree for choosing an architecture. The first question is whether there is enterprise budget and more than twelve months. If the answer is yes, the next question is whether millions of profiles need unifying across legacy systems: if that is also yes, the answer is an integrated suite; if it is no, an engagement-centric architecture, which suits an owned channel and short cycles. If the first answer is no, the question becomes whether the warehouse is already the source of truth: if it is, the answer is warehouse-native, with an in-house data team; if it is not, composable, which demands engineering in house.

Comparison of the four customer data architecture archetypes, with the same four questions asked of each: where data is captured, where the profile lives, where segmentation and decisions happen, and how activation works. First row, integrated suite: capture through SDKs and connectors from web, app and batch; the profile lives in the suite itself, in real time; audiences and decisions are resolved inside the platform; activation goes out through a destination catalog covering edge, streaming and file. Second row, engagement-centric: capture through SDK and API plus warehouse loads; the user profile lives inside the tool; segments and orchestration belong to the channel; activation is the owned channels, push, email and in-app. Third row, warehouse-native: capture is scheduled loads into the warehouse; the warehouse is the source of truth; compute happens in place and identity has to be brought in separately; activation is reverse ETL to each tool. Fourth row, composable: identical to the previous one in capture, warehouse and activation, differing only in that pieces are assembled rather than computing in place, with identity also separate. The last two rows look alike because they share almost everything.