Recently I've heard some interesting statements from people who are trying to decide whether OLAP or a relational-reporting tool is best for supporting their business intelligence (BI) needs. One person believes that because relational-reporting tools are a mature technology, they provide all the data he needs without specialized analytics. Another person wonders whether he can structure his OLAP cube to replicate the same reports that he previously generated from a relational-reporting tool. It's easy to get caught up in a particular technology and forget when it's appropriate to use that tool. OLAP and relational-reporting tools are both useful, but they work differently. You need to understand the strengths of each technology so that you can decide which one is best for your specific requirements.
Most people recognize that relational reporting is a more mature, well-understood, and widely used technology than OLAP. In addition, relational reporting scales well to large numbers of users and has a lot of flexibility. Relational reporting focuses on the problem of making a fixed number of reports available to large numbers of users. This solution provides advanced formats for online or hard-copy reports and lets you schedule and cache reports. Formatting, scheduling, and caching all make sense when you want a finite set of reports that will be reused many times.
But relational reporting isn't a slam-dunk solution because since relational reporting was originally created, the data needs of decision makers have become increasingly complex. Leaders must make decisions more quickly and take more information into account for every decision. Generally, more people are involved in making decisions today than in the past. An organization that had 25 decision makers 50 years ago might have 2500 decision makers today.
Relational reporting is a great answer to the problem of delivering data to your users, but sometimes, simply delivering data isn't enough. OLAP goes beyond simple delivery and helps you present large amounts of data to many users in ways that let them manipulate the data so that they can see trends and relationships more quickly. Can you imagine what an online search would be like if Internet search engines served up only a small fixed number of reports like most relational-reporting tools do? Search engines would be virtually worthless because the Internet contains too much information to include in a finite number of reports, and the contents and structure of the information change too frequently. Most business problems that have these characteristics are best solved when you use an OLAP application.
Figure 1 illustrates the process that decision makers go through to make a decision. The decision-making process is divided into two major phases: the data phase and the human phase. The data phase consists of all the steps in data delivery, and the human phase includes the steps that the decision maker goes through after receiving the data.
To better understand these steps, imagine you're a corporate department head and you receive a quarter-end expense report. On the report, you notice a huge jump in travel expenses for your department over the past quarter. You request another report that shows line-item details about travel expenses, and you figure out the problem is that employees are using multiple methods to book travel. After consulting with the accounting department, you institute a company policy that requires travelers to book their trips through the company travel agent to help ensure that travel costs return to normal levels.
Several things happened on the way to this decision. First, the company captured travel-expense data near the time the expenses occurred. Then, a period of time passed before the quarterly expense report was generated and distributed to you. This is the extraction, transformation, and loading (ETL) lag that Figure 1 shows. After you received the report, more time passed while you reviewed and discovered the problem; this time is the discovery lag. Finally, you needed more time to retrieve a line-item report and arrive at a decision to change the travel policy. This last time period is the decision-making lag.
In this example, relational reporting addresses only the data-delivery part of the process. Fortunately, to make your travel-policy decision, you needed only a limited amount of data; so two reports provided the information you needed to discover the root cause of the problem. For a more complex problem—say your gross margins in Europe have been slowly eroding over the past five quarters—you might require an enormous amount of data to find the root cause of the problem.
The part of decision making that relational reporting doesn't address is the human phase. Three primary drivers—problem abstraction, business-logic handling, and the discovery process—determine how long the human phase of the process will take. Let's see how OLAP applications address each of these drivers.