Why have a mart?

Prospective clients frequently ask us - what is a data mart or warehouse and why do we need one?

A data warehouse is a collection (a database) of used data structured in a way optimized for reporting and analysis. By used data we mean that the data originated in other systems (source systems) and was extracted, transformed and then loaded (ETL) into the data warehouse on a regular recurring basis.

  • Without a warehouse reports and analyses are executing directly against the source systems. What happens if a poorly written query drags down the performance of a key source system and suddenly users of the source system can't enter new sales orders or change production schedules?
  • The warehouse provides the ability to integrate information from different processes or even different systems. Almost always the information that is truly unique and strategic to an organization comes from a combination of systems and processes. For example, no one gets a real strategic advantage just from having a good sales order entry system. The strategic advantage comes from being able to efficiently take orders, add value through production and then deliver a superior product or service. If the sales order process goes well but the product or service isn't delivered well - there is no strategic advantage. Information that links sales to production to the financial results can allow managers to monitor and improve the entire value chain.
  • A data warehouse that is dimensionally modeled for reporting and analysis not only provides great ease of use and strong performance - it also matches the expectations of reporting toolsets. For example, connecting the IBM Cognos business intelligence tools to a normalized, transactional source system is a challenging modeling exercise that may yield unpredictable performance. In contrast, a dimensionally modeled data mart is a perfect fit the toolset and can be set up quickly with predictable query performance.

Although there is no precise definition most authorities agree that a data mart is a subject area centered, dimensionally modeled subset of a data warehouse.

In our experience the most valuable sort of data warehouse is one that is built up over time as a collection of data marts. Although this approach isn't optimal for all organizations we believe it to give the best value for our mid-market client base. In this approach we are following the ground-breaking work of Ralph Kimball.

A data mart requires more time, more effort and more cost. There are many reasons why a mart is almost always a good idea. But sometimes we recommend that we start without a mart - building a business intelligence system of reports and analysis directly on top of source systems like an accounting system, manufacturing system or human resources application.