Keep up with all of the essential KM news with a FREE subscription to KMWorld magazine. Find out more and subscribe today!

Predictive analytics: an asset to retail banking worldwide

 Predictive analytics is a dimension of business intelligence that allows organizations to assess both risks and opportunities. In retail banking, that process translates into questions such as: Which customers are likely to default on loans? Which are likely to be profitable, long-term customers? Getting the right answers is important because it has a direct effect on the bottom line.

 To answer those questions, historic data is used to construct a model that correlates the characteristics of a group of customers with their financial behavior. From a large set of measurements, key influencing factors are identified. The same information is then collected about other customers or prospects and matched against a profile that correlates with the target behavior. Banks can make decisions about issuing loans or marketing new products based on expected consumer behavior. IDC (idc.com) predicts that over the next four years, predictive analytics will grow at a healthy 8 percent per year.

 In general, the issues of concern to overseas banks are the same as those for domestic banks, although the economic environments may differ. For example, in some countries, the "credit economy" is a relatively new phenomenon. Assessing risk may be more difficult and at the same time, more important. In rapidly growing overseas retail banking markets, knowledge about customers may not have kept pace with growth. The need to comply with the Basel II Accord as of Jan. 1, 2007, is another driver for predictive analytics.

 One of the earliest predictive analytics techniques in retail banking was the use of scorecards, which were developed in the 1950s by Fair Isaac (www.fairisaac.com). Points were awarded for various criteria such as history in paying off loans, employment and credit usage. A single number derived from that data was used as an indicator of credit-worthiness.

 That indicator survives today in the form of the FICO score, widely used by financial institutions to help make decisions about consumers. The models used by Fair Isaac and other software producers are far more sophisticated now, but they still apply the same principles. The models first look at historic data, isolate key criteria and then project future behavior.

 Among Fair Isaac's customers is Caixa Catalunya (caixacatalunya.es), a large savings bank in Spain. The bank, which has nearly 1,000 branches and more than 3 million customers, wanted to convert from a manual risk management system to an automated one. Caixa Catalunya selected a set of products that included a customer management system and Application ScoreWare software.

 The new system allows the bank to assess risks based on customer application data and behavior. Decisions about whether to extend credit are made more quickly and consistently. Caixa Catalunya cites a substantial reduction in loan delinquency rates for existing as well as new customers.

 Predictive analytic methods help financial institutions meet requirements of the Basel II Accord by demonstrating improved risk management. Such procedures lead to a lower risk weighting, a factor that is used to calculate capital reserve requirements. When a bank has a lower capital reserve requirement, more capital is freed up for profit-making activities such as making loans.

 The underlying mathematics for calculating risk are complex, yet the users of those decision support systems are operations staff rather than statisticians.

 "Our philosophy is to put the power into the hands of people who make the financial decisions," says Rahul Asthana, director of product marketing for the enterprise decision management group at Fair Isaac. "We have done the analysis of the factors that define risk, so a non-statistician can immediately understand and explain the reason for a decision, such as why a loan was denied."

 Predictive analytics can create significant efficiencies in marketing. When a Dutch bank wanted to get the best response rate possible to a marketing campaign, it used Chordiant's (chordiant.com) analytics to develop a scoring model that indicated the response rate based on selecting different customers for the campaign. A pilot study was used to test and validate the model.

 The value of a model is illustrated by looking at how many customers need to be selected in order to get a given level of responders. Using Chordiant's model, selecting the right customers meant that just over half of the total candidate customers had to be contacted in order to reach 75% of those who will respond. A better return rate means lower marketing costs. In addition, use of the proper selection criteria means that customers will not be deluged with unwanted offers.

 Since only those factors that can legitimately be used to evaluate customers are used in the computer models, discrimination is eliminated.

 "Factors such as race and age may not be included for the purpose of denying an individual a loan or a credit card," notes Rob Walker, director of product marketing for decisioning at Chordiant. "We can also make finer distinctions than before," he adds, "looking for the exceptions in a group that might otherwise not be a good risk." Although decision making via analytics is sometimes seen as mechanical, Walker maintains that such systems actually are more subtle than a human-based evaluation, since people cannot take in—let alone combine—all the factors that play a role.

 The availability of extensive amounts of customer data has made analytics both more feasible and more challenging. "Earlier in the history of predictive analytics," says Walker, "we had to wait while data was collected before the process could begin. Now all banks have a centralized location where customer data is stored." However, the abundance of data from multiple products such as credit cards and mortgages, as well as channels such as call centers and online services, can make the analytical process quite complex.

 The Absa Group (www.absa.co.za), a bank in South Africa, chose SAS (www.SAS.com) for customer intelligence and uses it to match product marketing with customers' needs. However, the bank also used predictive analytics to reduce the risk of armed robbery. By examining the characteristics of banks that were most likely to be robbed, Absa was able to adjust its security procedures and sharply reduce the number of armed robberies, by 41 percent over two years, even though the number increased in the country as a whole. A model was developed that combined data mining and spatial mapping to help predict which banks would be at risk. In addition, the analyses were used to plan the location and type of new branches to be built.

 SAS, which specializes in business intelligence solutions built around its flagship product, the SAS 9 Enterprise BI Server, has a family of analytic products that help financial institutions manage risk and maximize customer value. Those include SAS Credit Risk Management, SAS Cross-Sell and Up-Sell, and Customer Retention Solutions. The Customer Retention Solutions allow companies to determine which customers are leaving and the reasons for doing so. Then, an alerting system is designed to identify customers at risk for leaving so that proactive steps can be taken to retain them.

 A new offering, SAS Forecast Server, adds the dimension of time to its predictive analytics. "Organizations can now look at causal relationships over time," says Charles Chase, product marketing manager for forecasting and econometrics at SAS. "For example, how does product pricing affect customer behavior during a series of defined intervals." SAS Forecast Server automatically builds dynamic regressions and assigns the most accurate modeling methodology. "As with our other products," Chase continues, "SAS Forecast Server does not require any program expertise on the part of the user—it's geared toward the business user."

 Sometimes, rapid change in a bank's customer base means that it has to play catch-up in getting to know its customers. Disbank TURKEY (www.disbank.com.tr) tripled its retail customer base over a three-year period after making a strategic decision to expand its business in that area, and wanted an increased understanding of the new group. Disbank hoped to strengthen its relationship with those new customers by engaging them through multiple products and services. The bank chose predictive analytics from SPSS (www.spss.com) to help understand and predict customer behavior.

 The first step in the process was to use predictive analytics to collect and analyze a wide variety of customer data, ranging from demographics, product purchases and choice of channel. A series of customer data analyses were conducted and a model developed to predict the likelihood of a given customer to purchase a product. The model predicted a very high rate of sales for an initial campaign—more than 93 percent. When the first campaign was completed, the rate did not quite match the expectation, but was an impressive 86 percent. As a result, Disbank reported a return on its investment within just three months of deployment of SPSS predictive analytics.

 With a long history in statistics, SPSS offers a platform that enables organizations to analyze their enterprise data and predict future behavior. The insights gained can be used in operational processes such as CRM, risk analysis and underwriting. SPSS text mining solutions allow organizations to include both structured and unstructured data in the same model. The heart of the platform is an enterprise data mining workbench used to create applications that detect patterns in data.

 Developing the business rules is a vital component of predictive analytics. "We always sit down with the organization and ask what they already know about typical patterns in their business," says Marcel Holsheimer, VP of product marketing at SPSS. "A lot of knowledge is held by the experts, and it should be incorporated into the automated system.

" Real-time predictions give organizations a significant edge, according to Holsheimer. "SPSS systems can be set up to scan data on a daily basis to look for a customer that fits the crossover or retention risk pattern," he observes. "There are indicators in the transactions, complaints and other data that is collected. Automatic alerts are sent to customer-facing personnel such as account managers, so they can take timely action." Call center representatives can check on those indicators while speaking with a customer and pinpoint opportunities in real time.

Judith Lamont is a research analyst with Zentek Corp., e-mail jlamont@sprintmail.com.

KMWorld Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues