Data warehousing is the sum of its marts - data marts and data warehouses - includes related articles on products for building data marts and Capital One's data warehouse - Technology Information

Software Magazine, Jan, 1997 by Barbara Francett

Not long ago, data marts were viewed as limited alternatives to fully populated enterprise data warehouses. Today, in the wake of a legion of failed or abandoned warehouse projects that took too long and cost too much, data marts have surged in popularity. Frequently, they serve as more manageable, cost-effective stepping-stones to the data warehouse.

A data mart is best defined as a subject-specific data warehouse. "Warehousing is a spectrum from large and complex to small and simple," says George Zagelow, IBM's project manager for data warehousing solutions. "A data warehouse has more users, more subjects, and offers a chance for users to cross subject-area lines."

What does not define a data mart is its size. Arbitrary ceilings of X number of gigabytes are artificial and impractical, according to industry experts. Some subject-area data marts may be huge, whereas an enterprise data warehouse of summary data could be surprisingly compact.

Complexity, however, is a another matter. The complexity of the data model for a data warehouse increases in accordance with the number of lines of business it serves. A data mart, regardless of size, requires a far simpler data model, since it focuses on a single subject area. The subject of a data mart is generally determined by the department or line of business for which it is created. Typically, the first data marts an organization develops are marketing and sales marts because the customer information they contain is essential to the health of the business.

Data marts can be subdivided into two varieties: independent and dependent. An independent data mart derives its data from a number of sources and operates autonomously. A dependent data mart is fed data from the enterprise data warehouse and is essentially a subset of the warehouse.

The nature of the organization's business drives whether a data mart will be independent or dependent, says Neil Mendelson, senior director of data warehousing at Oracle Corp., Redwood Shores, Calif. "Where a line of business has profit and loss responsibility, there may be no need or desire to integrate [data] across multiple lines of business," he says. "Where cooperation across business lines is an issue, there is no inherent value to independence."

Organizations are increasingly turning to data marts for their decision support needs for two principal reasons: They get simplicity and a quick, demonstrable return on investment. Specifically, the data mart's focus on a particular line of business means the number of data sources tapped are fewer, which makes modeling the data mart a simpler process; the time required to get the data mart up and running may be only a few months; and the cost is much lower -- about a tenth or less than the cost of an enterprise warehouse.

Nonetheless, many organizations will wind up with enterprise data warehouses anyway. "You can't indefinitely proliferate data marts because of the complexity of managing the data flows," says David Gelardi, IBM's program director of decision support systems and OLTP marketing for the RS/6000.

"Therein is one of the dangers of data mart technology. If you do not build an enterprise data warehouse structure underneath, data marts grow into a data warehouse on their own," says David Buch, IT director at Capital One, a credit-card company in Richmond, Va.

More than six years ago -- before the terms "data warehouse" or "data mart" were coined -- Capital One built a subject-oriented decision support environment for marketing and analysis data based on its mainframe Oracle database. This data mart, which contains information about the firm's customers and their purchase habits and histories, has grown from one million accounts to eight million accounts. It currently houses more than one terabyte of data.

"It's still our most active database," Buch says. He attributes the company's success, evidenced by its lower-than-average charge-off rate and growing cardholder base, to its "mass customization" strategy of targeted marketing that treats each of its eight million customers individually. "We target the credit-card customer who will be profitable at low risk," Buch says.

Meanwhile, additional big-volume data marts have grown up at Capital One. The largest of these are customer service and collections data marts, which have been built from the ground up. Each data mart contains over half a terabyte of data.

Now, the integration of these multiple data sources has become a big issue, Buch says. "We are being asked to integrate disparate data sources into an enterprise data warehouse. With such a 'spider web' environment, integration can be very difficult," he says. "We need to take the spider web apart and build an infrastructure. It's like building a skyscraper without a first floor, then going back and putting it in. It's a slow process."

Once the enterprise data warehouse starts rolling out, he says, it will be easy to integrate into other applications. The enterprise data warehouse will then feed multiple data marts. "The data marts will be re-sourced, with one source instead of many, and there will be more of them," he adds.