Custom Scoring Gains Ground
Increasingly, credit grantors are mixing and matching external and internal data, developing credit-scoring models that fit their customer base.
By Joanne Y. Cleaver
Why should George P. Monastra pay $35 for a fresh credit report on a current customer? He doesn't think he should. So the vice president of risk management for the office equipment division of leasing firm DeLage Landen Financial Services, of Wayne, Pa., is investing in custom-built credit scoring models that extract key payment history factors from the company's internal databases.
Taking archived data on current customers that includes slightly outdated commercial credit reports, Monastra and his colleagues can see how good a credit risk a customer is for the specific application under consideration. After all, it's cold comfort to know that the customer always pays utility bills on time but treats leasing obligations cavalierly. We look at 60,000 applications a year, so there's a lot of savings to continually building on a mini-model of that customer, based on our own data, he explains.
That's precisely the mentality behind a growing preference for custom and semi-custom scoring models. An increasing number of firms are investing more heavily in proprietary models to not only minimize credit risk but also provide a critical element to database-driven marketing efforts. At the same time, companies that serve up off-the-shelf models report their corporate customers are mixing and matching scores to better reflect their own customer bases. We do see a trend toward more proprietary types of models, says Chuck Byce, a director with San Rafael, Calif.-based Fair, Isaac's global custom analytics group. Customers, he says, recognize that proprietary models are the most powerful because each lender likes to think that he or she is in a certain niche, and proprietary models are the best tools to hone in on that particular population. People are beginning to look at the models as tied into the overall decision-making process, for marketing and other purposes.
The lending criteria that lenders want to use these days is not the applicant's history in paying back debt overall, but how likely the applicant is to pay back a particular type of debt promptly and in full. Lending institutions are internalizing the credit decisioning process relative to their lending and underwriting criteria, coupled with the additional information they know about that consumer, says Michael J. Mazzola, president, CreditXpert Inc., Towson, Md.
For instance, a consumer who has a rock-solid history paying back a jumbo mortgage, but is occasionally sloppy about getting credit card bills paid on time, is a better risk for a mortgage than a consumer who is occasionally late paying all sorts of bills. Custom-built scoring models give lenders a chance to pinpoint what their successful customers look like, says Herb W. Siedschlag, chief executive of Merit Credit Systems, Glendora, Calif. The firm specializes in building models for credit card issuers, student loans, and tenant screening.
Lenders are enthusiastically culling out their most profitable customers and developing profiles based on them, then reverse-engineering credit scoring models to detect applicants who fit that most-desired profile. We're not throwing away scorecards, emphasizes Peter J. Margaros, vice president, marketing, for scoring software firm Monetrics, of Beverly, Mass. But scorecards aren't able to weigh an element relevant to that particular buying decision.
Founded in 1997, Monetrics is just now unveiling its first software package, which is geared to auto lenders. The software constantly reformulates credit-scoring models, comparing the auto loan history of an applicant with real-time models of the current best-performing borrowers. The system can be accessed through an Internet browser, which means that judgment calls made in disparate locations should serve up the same credit recommendations.
Lenders with their own internal scoring models are quick to use them as a marketing tool. So far, only lenders with large customer bases and generous information technology budgets have been able to afford custom-built models. Siedschlag estimates that a minimum of 75,000 records is required to develop statistically valid models. Ordering up an initial proprietary model from Fair, Isaac is likely to cost a hefty $80,000 or so, but the cost drops in half - or more - for each subsequent model.
High-octane models are high-maintenance, too. Unlike off-the-shelf credit scoring models, which are maintained by the firms that create and market them, proprietary models spawn their own maintenance schedule. As a customer base shifts, custom models must be constantly tested, redefined, and refined.
Firms with custom-built scoring do want to see a return on investment right away. They're rarely disappointed, says Fair, Isaac's Byce. Typically, firms experience a 20% reduction in loan losses immediately after implementing a proprietary model. As cost-effective as completely custom models may be, medium- sized companies still may not be able to afford them. Instead, they are experimenting with pooled models that layer a variety of off-the-shelf models to come up with a semi-custom model, reports Fair, Isaac's Byce. Such models cost $10,000 to $30,000 a year.
Used consistently, proprietary scores can help keep customer acquisition costs down, points out Charles S. Chung, vice president of information intelligence for Experian, based in Orange, Calif. Some Experian customers are now paying as little as $50 to acquire a bankcard customer - a third of the prior expense - because the scoring models they use shut out delinquency-prone customers. ROI math, Siedschlag explains, can be as simple as figuring out how many bad accounts were avoided, compared to the cost of the model that screened them out: If the model would not have approved 10% of the currently delinquent $1 million worth of accounts, then its ROI would be $100,000.
Discovering a niche that is responsive, profitable, and pays on time - one others have dismissed - is where the big dollars are in terms of return on investment in the scoring model. For example, leveraging a custom model to market to people with spotty records can yield big results because they're often not inundated with offers, explains David R. Kelly, principal with Sigma Analytics and Consulting Inc. of Atlanta. He sees clients expanding their scoring models with in-house and demographic data to develop profiles of desirable customers. If your FICO score is a 620, probably you're not being marketed to, but you might be a low risk [for some loans], and you might be receptive because you're not getting 20 pitches a month, he says.