It can take a mountain of data to arrive at the elusive "360-degree view" of a customer -- but as AOL Germany found, it's an effort that can really pay off.
With competition fierce among Internet service providers (ISPs), AOL Germany needed to do all it could to better understand customers and target its marketing offers. That was a problem when it had access to only a central, corporate data warehouse in the U.S., according to Fred Türling, director of database marketing. The corporate data warehouse stored 15 basic attributes of each customer -- and didn't integrate any data from local German systems, such as call center, billing or loyalty program data. Limited network bandwidth made any serious analyses difficult and time-consuming, but marketing was begging for much more insight into customers -- for an understandable reason.
"This is one of the most competitive [ISP] markets in the world," Türling said. "For example, in 2005, an ISP flat rate was 20 euros. Now it's down to 0 euros."
That means that customers get basic ISP services such as email free of charge, and providers make their money on line charges, premium features and bundles of products. It also means that marketing must be highly targeted, focusing only on customers who are likely to buy a product. AOL Germany needed a much better understanding of customers than it was getting from just 15 attributes and limited analyses.
So, in 2002, AOL Germany developed a customer data warehouse to act as a central repository for analysis activities. After evaluating 11 vendors, the company settled on Westmount, Quebec-based SAND Technology Inc.'s Analytic Server, in part because of its compression capabilities. A SAND data warehouse is typically one-third the size of an equivalent indexed relational data warehouse or data mart, according to the vendor. And AOL knew it would need lots of space.
"With SAND, we now have a data mart which has more than 700 attributes describing the customer," Türling said.
That amounts to a warehouse that's approximately three terabytes, as opposed to the eight to 10 terabytes in a traditional, uncompressed version, Türling explained. The 700 attributes stored for each customer are compiled from about 70 different source systems, including customer databases, billing systems, call-center activities, click-stream and data related to premium services, such as VoIP or fax services. With systems in different time zones, Türling said, it takes a bit of coordination to get all of the data into the warehouse.
Every night, the different source systems push data to a staging area. An extract, transform and load (ETL) tool with embedded data quality checks (from Lexington, Mass.-based Ab Initio Software Corp.) then delivers cleansed data to the SAND Analytic Server. From there, Türling's team uses a layer of front-end analysis tools, such as Business Objects and SPSS, to create rich customer profiles and support marketing campaigns. Creating enriched customer profiles with accurate, up-to-date and complete data has enabled sophisticated segmentation and propensity modeling, Türling said.
"Now, we have a 360-degree view of customers because we have nearly everything which describes an AOL customer in one database," Türling said. "If you look at up-selling or cross-selling campaigns, [you] see that the enriched profile and up-to-date data have increased the response rate 5% to 20%, depending on the type of campaign."
The new system helped AOL Germany better support its loyalty programs, create a more accurate cancellation prediction model, and select the best users for the "sign on a friend" referral campaigns. It's also been able to better target up-selling and cross-selling campaigns, such as its push to upgrade customers from narrow-band to broadband connections. Türling said that while AOL hasn't attached ROI expectations to the warehouse, which it sees as more of an infrastructure investment, the analytic applications that it supports have made a major impact by increasing the effectiveness of marketing campaigns.
For others considering similar projects, Türling suggested that companies should carefully consider scalability and maintenance needs. He wants his team's time spent on data analysis, not data management.
"There are some data warehouse systems that are highly complex to run. I prefer systems which are more intelligent, like SAND," Türling said. "Low maintenance [needs] help you run it efficiently and put the time and effort into analyzing the data."