This large company's analytics software supported over 500 internal clients with predictive modeling projects for improving customer acquisition and retention, identifying up-sell and cross-sell opportunities, increasing revenue, and understanding customer lifetime value. Members of the global analytics team had been using other legacy analytics platforms for years, and they were heavily dependent on them for data extraction, processing, analysis, and reporting. However, there were drawbacks with these systems.

When the team was asked to migrate to the TIBCO® Data Science platform, some longtime users of other platforms had reservations, and first wanted to do a feasibility analysis. They wanted to create a customer lifetime value model for one of the business units, a type of project the team had undertaken dozens of times before. This model would enable calculating the risk and return profile of a client's customer base and provide guidance for maximizing the client's marketing and sales activities.

Because other legacy platforms were merely tools for data extraction and processing, and the TIBCO Data Science platform is an advanced analytics solution, they decided to use their existing Teradata Data Lab for data extraction and processing, and the TIBCO Data Science system for advanced analytics. With this strategy in place, the team began its feasibility analysis, eventually learning that the project was much easier, and delivered greater business value, using the TIBCO Data Science platform. In just a few months, even the most seasoned analytics platform veterans were delighted with the TIBCO Data Science platform, and looking back, wondered why they were anxious about the migration in the first place.

Easier Data Preparation

The first step was to deal with missing values and outliers to ensure a valid data set. "Other legacy analytics solutions require a lot of coding, but with the TIBCO Data Science platform, data preparation is an easy, menu-driven process," says a business analysis senior advisor. "We simply imported the data using the import wizard, selected the variables we wanted to treat for missing values from a list in a dialog box, then used the menus to specify the values they should be imputed to. We could replace the missing values by the mean, mode, median, statistical measures, a specific value, or a random number, all without coding. That made the data preparation easier to implement than other analytics solutions."

Excellent, Easy-to-Understand Graphics

Next, the team used a year's worth of transaction data extracted from Teradata and TIBCO Data Science algorithms to identify natural product groupings and validate business assumptions about purchasing behavior. "The TIBCO Data Science platform automatically created two very valuable tables. One listed the item sets and number of transactions for each, the other showed the association rules, for example, what products are purchased together.

"The TIBCO Data Science platform has excellent graphical capabilities that make it easy to visualize results. Other legacy analytics platforms we were using did not have Random Forest, so we would have needed to use R and write a function to create a graph. The TIBCO Data Science platform, on the other hand, has many auto-generated visualizations for Random Forest, and they are more useful than the graphs we could produce with R.

Easy, Powerful, Cutting-Edge Functions

"We used models supplied with the TIBCO Data Science system to determine both the predicted profit and the customer's probability of survival. The tool generated the actual and predicted values along with residuals, and it also enabled us to easily plot actual values versus predicted or residual values. It even allowed us to save the predicted values."

With these analytic models in place, the team moved to the final deliverable: a statistically valid analysis of customer behavior and recommendations for improving marketing and revenue. "The TIBCO Data Science system provides cutting-edge functionality. We were able to create clusters based on predicted margin and probability of survival, and test those clusters for statistical significance. Other legacy platforms require a lot of complex coding, and you have to change the variables every time you run the code. The TIBCO Data Science system provides the results you need with a click of a button, and the output is very easy to understand and interpret."

The Best Results

"In the past, when we had to code with another legacy analytics solution, we went in with an approach fixed in our minds. One of the very exciting things about the TIBCO Data Science platform is that it puts a range of techniques at our fingertips and makes it easy to play around with them, so we can discover which one delivers the best results. Simply put, it gives us the freedom to think."

The team is looking forward to exploring more of the tool's capabilities, including feeding its output directly into the company's data warehousing environment.