Oracle Data Mining and Analytics

I firmly believe that a major trend in applications is the incorporation of analytic-enabled functionality. Users want more than just reports or a replay of the past. Users want to have insights and their attention directed to key points. This is where analytics can make a big impact across all types of applications. Notice that I am not proposing exposing analytical capabilities (e.g., data mining and statistical modeling) to users. That, I think, is only effective for a small number of users. What I believe is more meaningful is to present functionality and information to users that are the result of analytics taking place behind the scene.

In line with this trend, Oracle has recently released a new application: Oracle Sales Prospector. This application is targeted at sales representatives. It provides which accounts they should target with specific products or services. It also indicates which deals are most likely to close within specific time frames, and provides accompanying corporate profiles as well as likely customer references.

This is a very cool application and I have wanted to talk about it for some time now. My group has worked closely with the Social CRM group in developing the data mining driving the product, and now that the it has launched I can discuss some of the data mining details.

Oracle Sales Prospector is as analytical-driven as an application can be. Almost every aspect of what you seen in its very nice user interface is the product of data mining models. Oracle Data Mining (ODM) provides the algorithms and the real-time scoring driving the application.

The product functionality is presented to the user in a easy to interact screen (see main screen above). Sales opportunities are presented in a bubble chart at the left of the main screen. Each bubble is an opportunity. An opportunity is a (customer, product) pair. The size of the bubble reflects the expected (estimated) revenue for the opportunity, the vertical placement how likely it is to close that opportunity, and the horizontal placement the expected number of months to close it.


If the user provides no filtering criteria, the application scores all customers and products associated with the sales representative and display the top ones (e.g., ten) ranked by expected revenue in the bubble chart. This is done in real-time and scales very nicely to large numbers of products and customers. One of the benefits of having data mining tightly integrated with a robust and scalable database server like Oracle's.


The user can filter opportunities by product or customer. It is also possible to constraint the opportunities to those with a projected revenue size above a given value and or probability of sale above a given threshold. Once the user specifies the filtering criteria, the application scores all customers and products in real-time and returns those that match the criteria. The real-time aspect of it ensures that the latest pieces of information (e.g., current sales data) are taken into account when assessing opportunities.

The opportunity bubble chart provides a lot of functionality and leverages different types of modeling techniques. It uses, in different ways, regression, clustering, and association techniques to provide scalable, accurate, and transparent recommendations (customer-product pairs) as well as estimates for time to close and expected revenue estimates. Robustness and scalability are important principles to keep in mind when design analytical applications. It is better to sacrifice some accuracy and select more robust and scalable techniques that will behave well with a variety of data distributions and data quality conditions.


For each opportunity (bubble) the application also returns a list of likely references. These are customers that are like the one in the opportunity but that have already purchased the product in the opportunity. Customer segmentation using clustering is used for this part of the application. Similarity between customers is computed taking into account both purchasing behavior and company information (e.g., size and industry).

For more information on Oracle Sales Prospector check out its web site here and the Social CRM blog here.

Labels: Applications, Oracle Sales Prospector, Social CRM