REVIEWER: Bill Comeau, senior director for Bell Canada Database Marketing Center.

BACKGROUND: Bell Canada's Database Marketing Center operates on a fairly large scale, supporting almost 1,000 internal marketing clients with data mining, modeling and segmentation, and list management solutions. We produce over 1,500 specific direct marketing lists annually in support of a wide variety of wireline, wireless, Internet and bundled telecommunications offers and channels.

PLATFORMS: We use NT for desktops and some Windows 95.

PROBLEM SOLVED: Our original need was to find ways to improve our direct marketing lists' time-to-market and sales results. We were looking for ways to improve our target modeling capabilities in a competitive and results-driven environment. When we purchased ANGOSS KnowledgeSTUDIO, we already had a suite of 80 predictive models in place. Almost all of these models were developed with a classical statistical approach using SAS software.

PRODUCT FUNCTIONALITY: First, Knowledge-STUDIO is used to access an existing SAS input dataset containing a sample of customers along with a multitude of their varying attributes that may or may not be correlated with a particular offer, e.g., a long distance offer. With a few simple menu-driven commands, the decision tree option is selected and the first step of modeling begins. We can visually examine the strongest factors associated with a particular offer using the tree diagram, bar graphs and pie charts. By looking at the tree from the top and then down each "branch," we can quickly discover what factors help to "split" customers who bought long distance from those who did not. Equally important, we find Knowledge-STUDIO very adept at discovering how different factors interact. For example, suppose "Region" was an important factor at the top of the tree. If Region A showed twice the sales rate as Region B, that would be important. If within Region A we could further find a branch (for example, high overseas callers) that showed four times the sales rate, then that interaction of Region and high overseas calling would be an important factor to consider in the final model. KnowledgeSTUDIO then outputs the factors described at the tree's "nodes" into SAS code that our analysts can then use in the final stage of model development in SAS. However, we have also used the decision tree as a final model in several cases. This has proven very useful when an explanatory or quick model solution is needed. The decision tree has proven useful in examining the important process factors behind product churn, for example, so that changes in business processes can be studied.

STRENGTHS: We have found several strengths of this product and vendor: ease of installation and use, the decision tree algorithm itself, a relatively low cost, compatibility with a larger model-building process and good desktop design. Service support was also good, including user training. Our analysts also still enjoy using it. As one analyst said, "I enjoy watching the trees grow; it's rather entertaining!"

WEAKNESSES: Based on our needs, there are two weaknesses that could be addressed in the future: better documentation and the ability to create new variables within the software. Although the software comes with "bells and whistles" such as reporting, logistic regression and neural networks, we have not really tested those tools fully given our success with the main application and its overall direct marketing process effectiveness.

SELECTION CRITERIA: The main reason we first purchased the software was because we needed to quicken the model exploration and development stage without sacrificing targeting power. ANGOSS's software was chosen because it could easily plug into our existing analytical process and SAS environment. We now use the software regularly as a preferred tool in our model building process, in conjunction with our SAS analytical suite.

DELIVERABLES: The applications of Knowledge-STUDIO have been quite successful in meeting our original business needs: we have reduced our modeling development time by 40 to 50 percent and have also seen an incremental improvement in our model targeting power of 5 to 10 percent based on this software.

VENDOR SUPPORT: When the software was introduced in 1999 to a core group of 6 to 10 modeling analysts, they found it easy to install on their desktop PCs. The team's initial focus was to exploit the decision tree functionality right away and enhance our modeling process. The software was extremely intuitive and easy to use from the outset, but we still needed to make a few calls to the hotline support.

DOCUMENTATION: KnowledgeSTUDIO has built-in online help with hard copies of the user manual available upon request.

Register or login for access to this item and much more

All Information Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access