The Tail That Wags the Dog

Where does business intelligence and decision support begin?

How BI Fits

BI draws most strongly from the data-driven category, integrating databases structured for analytic queries (data warehouses), OLAP-style slice-and-dice analysis and aggregations, and data mining to attempt to identify patterns that match a relatively small set of models. BI models are more descriptive than explanatory: They excel at showing "what" rather than "how."

I don't know of any widely available BI tool that, for instance, understands that random queue arrival times are most often described by a Poisson process, but every OLAP tool supports aggregations along a time dimension that will show arrival-time summary statistics in tables of numbers and in charts. Similarly, the charts of share prices and trading volumes that you'll find on numerous Web sites, despite their log scaling, moving averages, and comparative graphing, paint a very incomplete valuation picture if you don't consider model-derived financial-performance indicators that can serve as the basis for buy-and-sell decisions. Table 1 directly compares a number of BI and DS characteristics.

Although BI has been slow to incorporate more sophisticated DS modeling techniques, it is expanding rapidly in other DS directions. First, BI techniques and tools are being applied to new domains, such as Web clickstream analysis. Second, BI systems through portals are drawing from distributed data sources that may include the spectrum of media formats found on the Web. The use of XML as an interchange mechanism is part-and-parcel of this evolution. These BI portals present a consolidated view of source data; finding a single structure to model these sources would be a much harder task. Next, the reach of the Internet and the availability of lower-cost, commodity BI systems - spearheaded by companies such as Cognos Inc. and Microsoft - and groupware systems have brought communications-driven DS under the BI umbrella. And the emergence of "closed loop" BI - direct application of BI-derived knowledge in operational systems - further blurs BI and DS boundaries.

BI as a data-analysis practice has grabbed center stage, pushing DS off to the wings - a case of the tail wagging the dog. BI techniques and tools do satisfy the bulk of analytic needs for the majority of users, but nonetheless, many specialized DS modeling and analysis approaches are not found in BI systems. As these high-value DS possibilities - currently beyond BI's scope - grow in popularity, they will doubtless be incorporated into those systems. DS will continue to blaze the trail for BI's evolution.

Rate This Article Rating: 1 (best) 2 3 4 5 (worst) Comments: Optional e-mail address:

Seth Grimes [grimes@altaplana.com] is a principal of Alta Plana Corp., a Washington, D.C.-based consultancy specializing in analytic computing systems and demographic and economic statistics.