11.5 Data Warehouses and Data Marts

Since running analytics against transactional data can bog down a system, and since most organizations need to combine and reformat data from multiple sources, firms typically need to create separate data repositories for their reporting and analytics work—a kind of staging area from which to turn that data into information.

Two terms you’ll hear for these kinds of repositories are data warehouseA set of databases designed to support decision making in an organization. and data martA database or databases focused on addressing the concerns of a specific problem (e.g., increasing customer retention, improving product quality) or business unit (e.g., marketing, engineering).. A data warehouse is a set of databases designed to support decision making in an organization. It is structured for fast online queries and exploration. Data warehouses may aggregate enormous amounts of data from many different operational systems.

A data mart is a database focused on addressing the concerns of a specific problem (e.g., increasing customer retention, improving product quality) or business unit (e.g., marketing, engineering).

Marts and warehouses may contain huge volumes of data. For example, a firm may not need to keep large amounts of historical point-of-sale or transaction data in its operational systems, but it might want past data in its data mart so that managers can hunt for patterns and trends that occur over time.

Figure 11.2

Information systems supporting operations (such as TPS) are typically separate, and “feed” information systems used for analytics (such as data warehouses and data marts).

It’s easy for firms to get seduced by a software vendor’s demonstration showing data at your fingertips, presented in pretty graphs. But as mentioned earlier, getting data in a format that can be used for analytics is hard, complex, and challenging work. Large data warehouses can cost millions and take years to build. Every dollar spent on technology may lead to five to seven more dollars on consulting and other services.R. King, “Intelligence Software for Business,” BusinessWeek podcast, February 27, 2009.

Most firms will face a trade-off—do we attempt a large-scale integration of the whole firm, or more targeted efforts with quicker payoffs? Firms in fast-moving industries or with particularly complex businesses may struggle to get sweeping projects completed in enough time to reap benefits before business conditions change. Most consultants now advise smaller projects with narrow scope driven by specific business goals.D. Rigby and D. Ledingham, “CRM Done Right,” Harvard Business Review, November 2004; and R. King, “Intelligence Software for Business,” BusinessWeek podcast, February 27, 2009.

Firms can eventually get to a unified data warehouse but it may take time. Even analytics king Wal-Mart is just getting to that point. Retail giant Wal-Mart once reported having over seven hundred different data marts and hired Hewlett-Packard for help in bringing the systems together to form a more integrated data warehouse.H. Havenstein, “HP Nabs Wal-Mart as Data Warehousing Customer,” Computerworld, August 1, 2007.

The old saying from the movie Field of Dreams, “If you build it, they will come,” doesn’t hold up well for large-scale data analytics projects. This work should start with a clear vision with business-focused objectives. When senior executives can see objectives illustrated in potential payoff, they’ll be able to champion the effort, and experts agree, having an executive champion is a key success factor. Focusing on business issues will also drive technology choice, with the firm better able to focus on products that best fit its needs.

Once a firm has business goals and hoped-for payoffs clearly defined, it can address the broader issues needed to design, develop, deploy, and maintain its system:Key points adapted from Davenport and J. Harris, Competing on Analytics: The New Science of Winning (Boston: Harvard Business School Press, 2007).

• Data relevance. What data is needed to compete on analytics and to meet our current and future goals?
• Data sourcing. Can we even get the data we’ll need? Where can this data be obtained from? Is it available via our internal systems? Via third-party data aggregators? Via suppliers or sales partners? Do we need to set up new systems, surveys, and other collection efforts to acquire the data we need?
• Data quantity. How much data is needed?
• Data quality. Can our data be trusted as accurate? Is it clean, complete, and reasonably free of errors? How can the data be made more accurate and valuable for analysis? Will we need to ‘scrub,’ calculate, and consolidate data so that it can be used?
• Data hosting. Where will the systems be housed? What are the hardware and networking requirements for the effort?
• Data governance. What rules and processes are needed to manage data from its creation through its retirement? Are there operational issues (backup, disaster recovery)? Legal issues? Privacy issues? How should the firm handle security and access?

For some perspective on how difficult this can be, consider that an executive from one of the largest U.S. banks once lamented at how difficult it was to get his systems to do something as simple as properly distinguishing between men and women. The company’s customer-focused data warehouse drew data from thirty-six separate operational systems—bank teller systems, ATMs, student loan reporting systems, car loan systems, mortgage loan systems, and more. Collectively these legacy systems expressed gender in seventeen different ways: “M” or “F”; “m” or “f”; “Male” or “Female”; “MALE” or “FEMALE”; “1” for man, “0” for woman; “0” for man, “1” for woman and more, plus various codes for “unknown.” The best math in the world is of no help if the values used aren’t any good. There’s a saying in the industry, “garbage in, garbage out.”