Definitions: Data Warehouse

Definition

A data warehouse is a specialized database that is optimized for  analysis, reporting and decision support at both the tactical and strategic  levels. Data warehouses make sense because the data in production systems –  such as ERP systems – is stored and managed in ways that make analysis difficult. Creating new reports is therefore a time-consuming process that requires highly trained programmers who know how and where to access the required data. In contrast, with a data warehouse the process of creating new reports is relatively quick and easy, and can be done by department-level users with no need to involve the IT department. Sometimes the content of a data warehouse is partitioned by function into department-specific databases, often referred to as “data marts.”

Business Applications

The goal of a data warehouse is better business decisions through better business intelligence. Data warehouses can support this goal by providing reports that are more targeted to specific problems (Which suppliers should we eliminate?), more comprehensive (What are the effects of advertising on consumer take-away, warehouse capacity and factory production?), and available sooner – because they can be created “locally,” often by end-users themselves.

Controversies

Vendors of production systems, notably ERP systems, argue that data warehouses are unnecessary, and that their own “bolt-on” business intelligence productsare adequate or even superior. Data warehousing projects are expensive, often running into the mid-six figures, and their value is difficult to quantify because most of the benefits are indirect. Populating a database with data is complex and fraught with both technical and political problems, e.g. who owns the data?

Technical Details

Data warehousing projects require companies to address the problem of inconsistent data. This challenge, which is non-trivial, can be met via a combination of cleansing and ETL(Extract, Transform and Load). The cleansing process relates to problems with the actual data (e.g. the same individual or part listed by two different names). ETL is the process by which data is extracted from the production database, re-formatted to meet the data warehouse's requirements, and then loaded.