New software is allowing bank marketers to use sophisticated statistical techniques to fine-tune their campaigns while cutting costs and increasing success rates, moving predictive modeling, once the domain of expert statisticians, even closer to becoming an everyday tool for bank marketing.
Barclays Bank, the UK's third- largest bank, is using predictive analytics software Analytical Framework to speed up development of its marketing models and to improve customer targeting. The bank is using the product, created by San Francisco-based KXEN, for its consumer loan products, and is planning a rollout across the entire retail product line.
Matthew Common, head of CRM support at Barclays Bank's Customer Insight unit, says the software has cut model development time from more than two months to less than four weeks. "Another benefit of the software is that it has helped us improve the targeting of direct marketing campaigns by developing propensity scores for our customers," he says.
Barclays was concerned about the volume of direct mail that customers receive, particularly from financial services companies, notes KXEN's vp of marketing Joerg Rathenberg, who says the bank wanted to get away from a mass targeting base.
Common says the bank seeks "to constantly improve our targeting to ensure we only contact customers when we believe we have a relevant offer for them." This would also improve the efficiency of direct-marketing.
Barclays' has found statistical models can cut direct-mail volumes by as much as 70 percent while campaign-response rates can rise twofold or even threefold. Before installing the software on an AIX Unix server, a Common-led team of statisticians had been using older-generation software tools to manually develop models based on the variables stored in the data warehouse.
His team runs hundreds of campaigns per month, of which about 20 contain models. "The manual modeling techniques that we used prior to KXEN were time-consuming and resulted in models that were complicated to implement," he says. "Using traditional techniques to build models risks missing those variables that are most relevant for a specific model, as it's too time-consuming to review all possible combinations."
A key reason for selecting KXEN was its ability to import customer data from a much wider range of sources than those of other vendors. According to Rathenberg, KXEN can handle 500 to 2,000 data variables, while rival systems can handle only 30 before producing unstable models. "I've not seen anything comparable with KXEN in its ability to handle lots of variables and its speed of model-building," says Carol Baroudi, co-founder of research firm Baroudi-Bloor of Arlington, MA.
Barclays typically uses up to 2,000 customer variables to build a model, drawing them from its 24-node, eight-terabyte NCR Teradata system, which, according to TowerGroup analyst Guillermo Kopp, is one of the largest data warehouses in the world.
"Our data warehouse has the ability to store a huge amount of data, and also allows the analyst to manipulate and aggregate this data very quickly," says Common. "We have a very powerful toolkit to improve our understanding of customer behavior."
KXEN's product is based on a statistical learning theory developed by mathematician Vladimir Vapnik. "Because of this statistical approach, KXEN does not need to use the traditional data mart operational structure to produce results. It can go directly to the data, at the source," says Baroudi. "If a bank marketing executive needs an answer to an important business question, KXEN will provide that answer in 24 hours, whereas traditional, non-statistical techniques could take weeks or even months. If you can get useful data within 24 hours, then you could create some very interesting products, and you could be more confident about what types of people would apply for them."
KXEN doesn't require model developers to have training in statistics. "You still need a good knowledge of data and of statistical techniques to get the full benefit of the system, but you don't need to be a statistician, which means that a wider group of people within the bank can use KXEN to design models," says Common.
Kopp says the real benefits will come when banks use KXEN not just for marketing but also for improving CRM and customer service. But Common admits that Barclays is only currently using KXEN to make its marketing programs more efficient. "We don't currently have plans to investigate customer satisfaction using KXEN, but that's not to say we won't do so in the future," he says.
