To achieve with information purposes at several levels of detail for analysis and to help to ETL processes efficiency, a Business Intelligence Implementation have to count with several types of data storage with distinct functions and designs. These are:
- ODS or Operational Data Store: An off-line copy of operational systems’ data with additional features that aggregates historical versions of copied data and integrates the ER models of several applications’ data. This kind of repository cannot generate strategic information, but operational reports.
- DataWarehouse: A dimensional and aggregated representation of ODS retrieved data. Normally this repository contains enterprise-level data with no distinctions among organization-units data.
- DataMart: A dimensional and aggregated representation of organization-unit level data. A group of datamarts could be feed by enterprise-level datawarehouse, then for practical purposes, the differences between the first and the last are size and data specificity.
- OLAP Cubes: A kind of multidimensional database engine specialized on giving high performance analytical ad-hoc operations.
- Metadata Repository: This is data about the stored data. A well-maintained, detailed and organized data dictionary, which described all aspects above values and meanings of all data stored in all repositories of a Business Intelligence implementation. This is particularly important facing the final user perspective: this key repository gives meaning and sense to all possible values and results to any report, query and graphic.
Depending on design considerations, the type, skills and necessities of final users and the kind of analysis and decisions supported by a specific Business Intelligence implementation, the applications and users could have access over one or more kinds of repositories.