Credit scoring and other behavioral models are standard in credit card issuing. Now, scoring is becoming more widespread in consumer and even small business loans. The implications loom large.
Along with pre-approved offers and magnetic stripes, credit scoring is ubiquitous in the credit card business. There probably isn't a bank card issuer that doesn't assign scores to potential cardholders according to risk. In fact, issuers couldn't survive without it.
That's not been true in other areas of credit-granting, where bankers have relied upon credit reports, personal histories and old-fashioned intuition. But that situation is changing quickly. Scoring developers have created models designed for other areas of consumer credit, and they are gaining currency among bankers.
The ramifications of widespread scoring use could be huge. Banks could use credit scoring to remove human decision-making, reduce costs and segment accounts to sell other bank products. By automatically approving or denying loans by score, some human decisions could be removed altogether from the credit-granting process.
Combining scorecards with a bank's database promises massive gains in targeting consumer product needs. And neural networks, a technology that mimics human thought patterns, holds the promise that scoring could be taken to a new level. Already, scoring is being used to estimate revenues customers may bring in and to create scorecards for mortgage lending.
But bankers aren't necessarily cheering: The prospect of taking humans out of the application process doesn't sit well with most banks. Nor are databases and neural networks close to being integrated with scoring.
Still, credit scoring's move into the mainstream of credit operations appears inevitable. "If you can use these kind of scores to predict who will pay you and who is going to take advantage of the offer, you're going to get the maximum return," says W. Douglas Meredith, senior vice president for Liberty National Bank in Louisville.
Building a scorecard entails taking information on a body of loans - credit card or car loans, for example - to predict future behavior. A scorecard predicts future behavior by plugging in credit bureau and other data. There is no "right" score: Some banks will find a credit score of 500 fine, while others will avoid the applicant at all costs.
"There's nothing magic about credit scoring," says a senior vice president at a medium-sized Mid-western bank. "It's just giving a number to things such as stability factors."
If it isn't magic, it certainly isn't new, either. Scorecards have been available since the 1950s. As new scorecards developed, techniques evolved, integrating more credit bureau and bank data through computers.
The price has come down as well. Scoring is inexpensive enough for most banks. Credit bureaus, with immediate access to credit histories, sell credit and behavioral scores to banks, or credit modeling operations such as Fair, Isaac and Co. or The MDS Group build custom scorecards. Custom scorecards cost about $20,000, while scores from credit bureaus are as low as 10 cents per score for large users.
As with other consumer credit areas, credit cards have been scoring pioneers. Many banks use scoring at every stage of an account. Prospective applicants are scored to see if they're desirable customers, then scored to see if they'll respond to a mailing. Cardholders are scored for profitability, risk and attrition. When an account goes bad, the collections department scores the account for the likelihood that it will become a charge-off, scores it to predict how much money the bank can collect and whether the cardholder will go into bankruptcy.
Yet "most people think of scoring in terms of acquiring accounts," says Steve Darsie, senior vice president, CCN Division, at Atlanta-based scoring developer MDS.
Enter Revenue Scoring
But bankers are expanding their horizons beyond credit scoring. Revenue scoring is becoming one of the most sought-after scoring products, says Drew Shurmantine, manager of credit scoring services for credit bureau TRW Inc. Such models predict the amount of revenue a customer will generate if approved for a loan or a credit card. "We're having a lot of discussions with people," Shurmantine says.
In addition, scorecards have been built for products that haven't used scoring, such as home mortgages, which have become increasingly competitive and risky. Not only does scoring evaluate risk, but it can help a bank comply with fair lending laws by providing empirical reasons for rejecting a loan.
The ability to build scorecards for products that don't have much credit history has helped spawn one of the scoring industry's hottest bank products, scoring for small business loans. While not new, it's a segment that only now seems to be taking off. Banks demand greater information on small business applications and more measures of applicants' risk. At the same time, banks are trying to cut the time to approve small business loans, which are often $100,000 or less. Scoring enables a bank to get a quick read on a business, saving hours of having to pore over financial statements.
And speed is one of scoring's biggest virtues. With a few keystrokes, a scorecard can give a profile on an individual's risk or revenue potential. Few banks rely on such information solely to make credit decisions, but it's clearly a precursor to reducing human involvement in granting loans.
In a new system at First National Bank of Chicago, loan applications with a certain score will be automatically approved or denied, without any input from a credit officer. First Chicago hopes to automate approximately 60% of consumer loans with scoring, says Lawrence Donoghue, vice president, community banking group, though Donoghue says the bank is not necessarily using scoring to reduce risk.
First Chicago expects loan production per loan officer to double from the current three to four loans a day. "(Our loss rates are) below the industry and have been for many years. (We want to) lower operating expenses and then build volume," Donoghue says.
The danger of using such parameters in approving or denying loans is that it fails to take into account other business ties the applicant may have with the bank. Liberty National Bank (which becomes Banc One Kentucky in November 1995) has a large auto lending business. While scoring provides a good foundation for a credit officer to decide on a car loan, Liberty National's Meredith says there are no absolutes - doing so could alienate customers.
"If an auto dealer has a strong relationship with the bank, and their general manger calls and would like to have this particular deal done because it's for the next-door neighbor, we'll take that," Meredith says. Approximately 40% of Liberty National's bad loans come from 7% of the applicants who score below the bank's cutoff score, but "we have no absolutes as far as not approving applicants below the scoring limits," Meredith says.
Two-Score Technique
Liberty uses both a bankruptcy score and credit score to evaluate applications. When a customer wants a loan that's more than the amount of the automobile, the two-score approach gives loan officers solid ground to approve the extra loan amount.
The increasing dependence on scores puts pressure on developers to build more accurate scorecards. Enter neural networks, a computer technology that can infer outcomes from dozens of seemingly unrelated pieces of data. Neural nets promise to raise scoring models' accuracy to an even higher level.
But for now, it's just a promise, many say. While neural nets have proven their worth in combating credit card fraud, the technology hasn't made its way into scoring yet. The knock on nets is that their increased ability to predict doesn't justify the investment, which can reach well over $100,000 for large, ultra-sophisticated systems.
Neural net vendors say performance problems result from software run on neural nets, not the nets themselves. In addition, banks are waiting for the other shoe to drop before installing systems: "Fundamentally, it's an issue of, 'I want to try it, but I'll wait for my competitors,'" says Alan Jost, vice president of decision systems for neural network designer HNC Inc. in San Diego.
While neural net scoring programs slowly infiltrate banks, database marketing - the use of customer files to segment customers and target products - is moving much faster. Yet it, too, hasn't been fully exploited. Combining scoring with database marketing could be a marketers' dream.
For instance, customers who fit a certain profile from database information might appear good candidates for a home-equity line of credit. But before the bank makes the pitch, it can score the prospect for risk and revenue potential, as well as the likelihood the customer will even respond to an offer. That would eliminate poor prospects and allow the bank to concentrate on the best leads.
"I see greater and greater levels of sophistication when it comes to segmenting the customer base," says Charley Schmidt, senior vice president and general manager of First Omni Bank, a Millsboro, DE-based credit card issuer with 700,000 accounts. Database marketing defines prospects, and scoring narrows that segment further, he says.
It's likely that every bank of any size will eventually use scorecards and databases in tandem. As Schmidt points out, it's already happening in credit cards, as card issuers use scoring to narrow prospect lists. No longer is the card market growing; issuers gain market share by stealing customers from each other. And usually, it's the same affluent, stable customer that lenders want. The same principle writ large raises the stakes in consumer and small business lending.
"My credit card competitors are very sophisticated in the way they target their offering," Schmidt says. "I can see the accounts being taken from me are in a very narrow range of profit and loss."
