DECISION TECHNOLOGIES IN DATABASE MARKETING: PART V
by Gene M. Ferruzza, Senior VP, Decision Technologies

Data Visualization

Understanding the human factor is critical for delivering data-mining results in ways that are actionable by the business user. Only when information is understood and actionable can it be considered beneficial in marketing programs. Sometimes it is useful for the user to visualize data-mining results in order to use them appropriately in marketing programs. Because data visualization often makes data understandable, it is considered an important basic data-mining technique and often is the final stage of the data-mining process.

Data visualization tools usually are graphical. Charts and graphs make data characteristics more obvious and apparent. Graphs are most useful for presenting results involving four or fewer variables. Thus, because far more than four customer characteristics generally are represented in data marts, the number of variables must be reduced for the purpose of data visualization. The business or marketing analyst can choose the proper set of customer characteristics to view, or data-mining processes can be used to reduce the number of variables to look at.

On-Line Analytical Processing

On-line analytical processing (OLAP) is a process that produces rule-based queries and summarizations. However, OLAP is as much a data organization technology as it is a complex query technique. Closely associated with multi-dimensional data modeling, OLAP has driven the development of concepts now considered "best practice" for data-mart development.

There are similarities, even across industries, in how the majority of database marketing programs and analyses need to "slice" the data. Such slices are "views," more commonly referred to as "dimensions." Thus, a dimension corresponds to a view of the data, usually related to the business application, that the user is likely to need as part of data analyses. Typical dimensions are product and service, customer, time, geographic region, business unit, marketing communications, and so forth. In OLAP, both the database and the queries are organized and optimized in multiple dimensions.

OLAP data mining techniques generally are rule-based, with multiple premises (dimensions), and they result in display of summary information in graphical or tabular form. A typical multi-dimensional query would be to select all customers purchasing product Z from January through March of 1997 in the Midwest who received a particular promotional piece, and then to list the customers by store. This query could be adjusted to, for example, list all customers by total purchases or even by propensity to purchase in the future.

A specific instance of a dimensional model often is referred to as a "cube." The cube is a useful representation of the cross-dimensional characteristics of OLAP. A model of more than three dimensions can be represented as a hyper-cube. Another feature of most OLAP tools is that they incorporate data visualization techniques that allow the user to navigate through the data by manipulating graphs on the computer screen. This type of "drill- down" analysis allows the user to select (click on) one element of a graph and navigate down to increasingly detailed views of the data. In fact, in many OLAP tools, traditional queries are replaced by visual navigation.

Segmentation

The term "segmentation" is used loosely, especially in database marketing. Any analytical process performed on a customer database may be used to segment the customers (i.e., sort them into categories). Segmentation techniques are based on the use of customer behavior data, third-party customer demographic or psychographic data, or primary research data obtained directly from prospects or customers.

The classical approach to customer segmentation, used mostly by market researchers, is to categorize the customer base through the use of attitudinal data obtained through primary research. The customers are asked questions to elicit attitudinal responses that aid in decisions on how these customers should be treated, and why. Attitudinal categories then are used to segment the customers into homogeneous groups based on their responses.

In contrast, database marketing uses customer behavior and third-party demographic information. As discussed above, behavioral data are collected through routine interactions between the company and the customer. Demographic and psychographic data are generated as a result of the customer's interactions with other organizations or institutions and are collected by a third party.
---
Part VI of this series will appear in next week's edition of D S * .
---
Contact Gene Ferruzza at gmf@cmsnet.com