A data model determines how customer data is organized for marketing queries, audiences, and campaign analysis. A composable customer data platform (CDP), for example, can identify customers by a primary key—a unique data field—and link other tables to it. This lets teams use the data already held in a marketing data lakehouse without replacing a complicated model just to activate it. section 4.3.1 section 5.4.2
Start with the audiences you need, not a model designed for every future campaign. Teams can query their first audiences manually with SQL, test two or three valuable use cases, and generalize the model as demand grows. Early results help build support before operational scalability becomes the priority. section 3.1
Choose a modeling approach for the data and questions at hand:
- Entity-relationship (ER) modeling suits structured data and may be familiar to teams using relational databases. section 5.4.2
- Dimensional modeling supports dashboards and targeting user cohorts. Star schemas are generally easier to query; snowflake schemas can reduce redundancy but require more joins. section 5.4.2
- Data vault modeling supports complex, changing data through hubs, links, and satellites that aid auditing and versioning. It does not replace access controls, encryption, or governance. section 5.4.2
A lakehouse may also hold unstructured content, such as call center transcripts or social media text. Its data model can therefore combine relational schemas for structured records with metadata tagging, search indexes, or other ways to organize less structured content. Establish naming conventions and schema evolution policies early so new sources do not become another silo. section 5.4.2
Model around an initial marketing goal
A travel e-commerce company scoped its first-party data rollout around churn and recent purchasers while testing activation across ads and messaging. Those limited use cases gave its data and marketing teams a shared roadmap rather than requiring a comprehensive model before launch. section 3.1
