Adjusting the Analytical Process
The data warehouse and decision support software (DSS) literature of books, articles, conferences, lectures and presentations all describe the process of creating, deploying, and managing a large DSS database. There is tremendous temptation to focus on the technology used to build and manage a data warehouse, especially since it's complex and interesting. But it is crucial to keep in mind that the data warehouse or OLAP cube exists for one, and only one, purpose -- to help end users make better, more informed decisions. Given the fast pace of business today and the pressure to improve the productivity of knowledge workers and the quality of their work, users often find a well-designed system can be invaluable. It lets them execute their normal workload in a fraction of the time, freeing them to delve deeper into analyzing, understanding and solving complex problems.
Transforming an organization into an information-centric entity involves more than providing access to data. It includes providing the infrastructure that enables users to take advantage of data. This infrastructure should have databases and tools, as well as applications that deliver information to users in ways that are meaningful to them. Any transformation should include training and examples of the kinds of queries and analysis they can perform to obtain greater value from the data.
Unfortunately most companies have, up until now, focused on the technology used to deliver data to users. They haven't focused on institutionalizing the analytical process to allow the organization to extract the greatest value from the data.
There are five basic forms of information-centric analysis that users can perform. Companies need to create processes and procedures to help users understand how to apply these forms of analysis through DSS applications and training. The five forms of analysis are benchmarking, limiting, highlighting, trending and communicating.
Benchmarking is the process of comparing two sets of facts. It's important to perform benchmarking with the right set of data. For example, if you simply benchmark year-over-year sales and see that sales have gone up 100 percent, you might feel that everything is going well. If, however, the market your product is in has grown 500 percent, the reality is that your company is losing market share since it isn't growing as fast as the overall market.
Limiting is the process of examining increasingly lower levels of detail. Limiting, also know as drill down, restricts the scope of an analysis to a finer level of detail. This is what many people think OLAP analysis is. Rather than focusing on the technology required, however, the goal of the analysis is to investigate the root causes of performance problems that are buried in the data.
Highlighting attempts to expose key issues in the data. Ranking, exception reporting and segmenting are three types of highlighting analysis.
Trending gives you a historical perspective on what led to a certain situation. You can trend any kind of periodic data. There are several types of trends you can examine, such as simple trends, rolling trends, gain analysis and smoothing trends. Trending is also useful when you want to attempt to make predictions about the future. Good trending software and data will let you make projections with reasonable confidence that the analysis will be true.
You must keep in mind, however, that extraneous factors that are not part of the data you are trending can have a major impact on the ability of your trends to predict future behavior. For example, if the stock market dropped by 30 percent within the next year, what impact would this have on your business? A major war or sharp downturn in the economy could also cause market growth to become negative beyond your expectations.
Communicating your results requires making the analysis crisp, easy to read and simple to understand. The judicious use of color can help highlight exceptions or major areas of concern. Most good analysis reports use graphics to help the brain grasp concepts visually. Clearly describe all assumptions and label all major report elements. Obscure and pedantic charting should be avoided at all costs. --Robert Craig is vice president of marketing at WebXi Inc. (Burlington, Mass.), and a former director at the Hurwitz Group Inc. Contact him at rcraig@webxi.com.