How do you use data modeling to support business intelligence and analytics?

If conceptual data modeling is the first step, concept modeling is the zeroth. This subtlety is important. Concept modeling establishes the business-layer semantics; the common terms and meanings. In any enterprise, you expect a single concept model from which you may then derive one or more conceptualized models; typically focusing on sub-domains. Consider that there is the singular concept of a 'Customer' but within say Marketing, Ordering, and Fulfillment, there may be differing conceptualizations of Customer according to the needs of those sub-domains.

Depending on your database type and use case you will target a specific scheme for your data model. Typically, in the conceptual data model I focus on high-level entities/tables and their relationships. This is true if I'm using UML, ERD, or ORM modeling tools. I can easily use conceptual data models to communicate with business users because I focus on only key business topics and the rules around how they relate.

Generic data types must be present at the logical model layer. Physical data types, meaning, platform specific types, are defined at the physical model level. One misconception about the physical model is that it is synonymous with the physical database. It is not. "Physical" means that the model represents what will literally be implemented on some particular vendor's DBMS.

Attributes definition takes place logical model rather than in a conceptual one. Data types are part of data structures and are finally done physical model.

If the process starts with reverse engineering an existing solution, then the order my be physical first, then mapping to a conceptual or logical (if they already exist).

Data Vault aims to resolve agility as the (3NF) modelling produces complex primary keys with cascading snapshots, while dimensional modelling is challenging to modify for any granularity changes. DV is useful if you assume that the model will require several iterations to the data model, either through the acquisition of different data classes or new insights with existing data and assumptions. The hub-and-spoke nature of Vault between entities means that objects can be loaded and deployed independently. Therefore, when you need to make changes to your model, you typically create a new object rather than modifying an existing one. You can then phase out the old version or stop processing it. The same applies from a processing perspective.

How the data will be used is critical to how the data model is defined. For example, a data model to be used fir integration will not be the same is one for physical implementation, and one for business communication will be different again. Understand the use as much as possible and build the data model to match this.