MAINFRAME > Business Strategy > BI and Analytics

Retailer Optimizes Analytics to Drive a More Customized Client Experience

Personal Shopper
Illustration by Bruce Morser

A customer’s experience is incredibly important in how he or she views a brand. For many, their first experience will be online, likely through a mobile device. Looking at customer behavior can help determine future interest, improve customer satisfaction and increase top-line revenue and growth.

Predictive analytics can examine various parameters of the business and gain a more accurate picture of customers and their expectations. This helps the business make better decisions faster and improves the efficacy of future customer engagements. Predictive models can provide support for human decisions or, in some cases, be used to automate an entire decision-making process (e.g., the development of selling recommendations for online transactions or purchases).

Predictive Analytics in Retail

A classic example of predictive analytics at work is retail. There, predictive models are created based on historical data to foresee a customer’s next action to improve the experience.

A European retailer improved top-line business results by 5 percent by utilizing predictive analytics on the mainframe at the point of sales. With it, the retailer developed “suggestive selling” recommendations based on past buying behavior and the contents of a customer’s current shopping cart. This retailer analyzes hundreds of thousands of purchase transactions daily across several locations.

With such a large volume of historical data, it sometimes took up to a week for the retailer to upgrade even a single predictive model. This dramatically limited the retailer’s capability to run models on full data sets and change, adopt and improve models efficiently and interactively—something it had to do to gain more insight into its current data and optimize the product assignments and marketing campaign in a timely manner.

Customer Study: Optimized Infrastructure

The European retailer believed that it could drive more top-line revenue with more accurate and timely predictive models, but was limited by its existing infrastructure because analyzing historical purchase data was restricted to a few months and the frequency of model refresh was limited to weeks. The most demanding case took a week even with a small subset of historical purchase data of a few months.

The enterprise entered into a First-of-a-Kind project ( which brings together clients with researchers to test technologies on real business problems. In the existing infrastructure, customer and historical purchase data resides in DB2* for z/OS* with IBM DB2 Analytics Accelerator. From there, data sets are selected by an analyst and moved into the IBM SPSS* Modeler server for further processing.

The predictive analytics process in the customer infrastructure consists of three steps:

  1. In the data preparation phase a data analyst copies historical purchase and customer data from a data warehouse on the mainframe to the SPSS Modeler on a distributed platform. The data is then prepared via a series of steps (e.g., cleansing, aggregations, joins, transforms, etc.) all within the SPSS Modeler server.
  2. In the modeling phase the models are calculated in SPSS Modeler based on the data prepared in the first step and the predictive model nugget is created.
  3. The scores of the model are generated and analyzed by the data analyst.

Dr. Otto Wohlmuth received his Ph.D. degree in Computer Science in 2000 from the University of Hamburg-Harburg, Germany. Since 1998, he worked at IBM Research and Development in Boeblingen as architect and then as manager in server and software development.

comments powered by Disqus



2019 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.


AI-Driven Technological Progress

All Together Now

A Centralized Business Analytics Environment Delivers Greater Value

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store
Mainframe News Sign Up Today! Past News Letters