Summary

We added the Internet visit information to the data warehouse and Analysis Services database and built Integration Services packages to load the data from the e-commerce system's database. We created a mining model to add customer segmentation to the cube, and another model to supply product recommendations to the Web site. We decided not to use the e-commerce application's built-in BI features because the objectives required extensive data from the existing data warehouse, which was not available in the e-commerce database.

Marketing activities, such as Internet advertising and direct mail, can now be targeted more effectively at customers based on their use of the Web site and their customer profiles. The high-performance cross-sell feature on the Web site is recommending additional DVDs that the customer might like to purchase, hopefully leading to additional items sold per transaction.

Because many of the interesting measures in the Internet activity fact data are measures of the time elapsed between two dates, we added a set of time span calculations to the fact views that were used by the cube. To calculate the time up until the present day, the data load date was used rather than the current system date.

We used the Microsoft Clustering algorithm to create a customer segmentation mining model and data mining dimension so that analysts can use clusters of customers such as Newbies and Store-Only Buyers. We used the Microsoft Association Rules algorithm to create a product recommendations model, and added a DMX query to the e-commerce Web application to suggest a list of possible DVDs that a customer might also purchase.

The product recommendations mining model was deployed to a production server that can be accessed by the e-commerce Web application, and the security was configured so that the Web application can query the mining model. A new operations task is to periodically reprocess the product recommendations mining model so that it is kept current with the latest data.