Mitch Bonilla is deploying one of the most exotic, sophisticated kinds of technology available today to tackle one of humankind's oldest tasks: bill collecting.
Bonilla and his San Diego-based company, ContiAsset Receivables Management, are using neural network software to better understand the people who fall behind in their bills-and to help identify which ones will ultimately pay up.
Neural networks are computer programs that actually "learn" from their experience performing a particular task, and then get better at it, like IBM's Deep Blue, by identifying significant patterns and aberrations. The programs are used for a wide variety of complicated jobs, from managing power plants and understanding traffic jams to doing medical diagnosis and detecting fraud in financial transactions.
Bonilla just wants a little help figuring out which of his delinquent credit card customers will ultimately pay their bills in full, which ones will only pay 50 cents on the dollar, and which ones will tell him to get lost. The neural network his company has designed takes data about customers from several different sources and searches for patterns that will predict how various groups of delinquent debtors are likely to behave. Bonilla would, of course, much rather call on the full-pay customers first. "We're in the developing stage now," he admits. "It's a little fragmented, but even in fragmented pieces, it's working pretty well."
ContiAsset is one of a handful of companies nationwide pushing into a new frontier in financial services: studying the demographics and psychographics of delinquent debtors so that creditors can tailor their collection approach. The company is looking for signals about payment behavior based on its customers' current situation-their present income and how they're dealing with other creditors.
Other financial services companies are trying to understand whether the circumstances that caused people to fall behind on their bills can be used to predict if they will ultimately pay those bills. Do divorced people need different incentives than those who have been widowed? Do kids fresh out of college who max out their cards with furniture and bar bills respond differently than people who were unemployed for nine months and used credit to buy milk and bread? Who struggles to pay, and why?
Tulsa-based Consumer Financial Services, which owns $15 billion in charged-off credit, has been the giant within the group of companies that buy delinquent debt from credit card companies for a substantial discount, then work to collect that debt. CFS, which stumbled this fall amid allegations of irregularities involving its financing, is nonetheless a pioneer in studying delinquent debtors. This summer, the company went so far as to hire a full-time clinical psychologist to help it learn more about its customers, luring John Bachman from private practice in Menlo Park, California.
Demographic analysis has been common for years, if not decades, among companies that offer credit-using sophisticated formulas to figure out who to offer credit cards to. But even the credit card companies haven't expended much energy trying to understand why people stop paying their bills, and how best to persuade them to resume paying.
"What we would like," says Wayne Learned, operations manager of CFS, "is to be able to say about a customer, 'This is a WASP from Minnesota who subscribes to these three magazines, attends this kind of church, has this socioeconomic status,'" and got into financial trouble because of, say, a divorce. "So when we contact him, our computer [will] say, 'This is who he is and this is the available employee who is the closest match, in terms of demographics and psychological profile.'"
Two things are surprising about these efforts. One is that they reveal how little financial services companies often know about their customers. The other is how revealing, and how powerful, even a little bit more information will be.
"We don't know where this is going to end up," says Learned. "We're just dreaming this stuff."
More than ever, this is the season of consumer debt, and not only because December is the month Americans traditionally buy things they pay for come spring. In 1997, there were 1.3 million personal bankruptcy filings-an all-time record-up 20 percent from 1996 filings, which were 29 percent above 1995. Those three years alone account for more bankruptcies than were filed during the entire 1970s. At the end of July 1998, Americans owed $1.266 trillion in nonmortgage debt-a number that has ballooned 66 percent since the roaring 1980s and now equals roughly $5,000 in debt for every man, woman, and child in the nation.
Of course, most people pay their bills. But the ones who can't keep up are a huge business in the United States. Credit card companies alone charge off $22 billion a year in unpaid bills-$60 million a day. The 1.6 million families who voluntarily sought counseling for financial problems in 1997 with affiliates of the National Foundation for Consumer Credit, a nonprofit association of credit counseling agencies, together claimed delinquent debts totaling $35 billion-and they make up less than 2 percent of all U.S. households. (The percent of bank credit cards with past-due balances has ping-ponged up and down over the last decade between 2.5 percent and 5.6 percent, but the size of the total debt, and thus the mass of the delinquent debt, has grown steadily.)
Despite the amount of money involved, almost nothing is known about these delinquent debtors. People who pay their bills on time are not lenders' profitable customers, while people who go bankrupt turn the profit equation inside out. It is the people in the middle, those who fall behind but are determined to make good, who can generate the best return for credit companies. "That one sector should be of interest to creditors," says Mac McCarty, an analyst for Trajecta, an Austin, Texas, consulting company that develops tools for assessing risk in financial portfolios.
Trajecta has begun to work with some clients, including credit issuers and financial services clients, to understand delinquent debtors better. "With the market saturated, with banks having issued all the credit cards they can issue, people are saying, 'Hey, let's look at the back end of this process, let's learn to collect better,'" says Matt Harris, vice president of analytical services at Trajecta.
Among the early users of such techniques are a few of the credit card issuers themselves. Banks and card issuers have traditionally written off their delinquent accounts after six months, and sold them at a steep discount-a dime on the dollar, or less-at the time of write-off. For some banks, that's changing. They still write the debts off, as required by banking regulations, but they are choosing some of the delinquent accounts to work themselves. Lenders are reluctant to discuss how they handle delinquent customers, for reasons of competition and customer sensitivity.
ContiAsset's Bonilla, who buys the bad debt from banks, attributes his idea to do more analytical work on delinquency to the credit card companies themselves. "I know of at least three [credit card companies] who are identifying the accounts most likely to pay and sending them to their own collection agencies. They are identifying the accounts least likely to pay, and those are the ones they are going to market with."
And what does determine whether someone digs out of debt or goes bankrupt? Does the original source of the debt matter? Does it matter if some one-time event-like a temporary layoff-causes someone to fall behind? If someone has never been delinquent before, is there anything in his history that helps predict whether he'll catch up?
"The short answer is, nobody's got the data," says Jesse Snyder, managing editor of Collections & Credit Risk magazine. The application of the kind of demographic, statistical, and algorithmic energy that is common elsewhere-from understanding how often to send someone a mail-order catalog to deciding whether to offer credit in the first place-is only now being applied to delinquent debtors.
"I was talking to a credit-card company official the other day," says Leigh Smith, of SMR Research in Hackettstown, New Jersey, which does research on consumer credit. "And he said to me, 'The difference between our most profitable customer and the delinquent customer is not that big, demographically. Age, income, debt outstanding, they are all very similar.'"
Of course, there is all the difference in the world between the best customer and the charge-off customer: It's the difference between the person who has the wherewithal to pay his bills and the person who doesn't. The question is whether that difference can be quantified.
"We call it 'The Grit Index,'" says Suzanne Boas, president of Consumer Credit Counseling Service of Atlanta (CCSA). "What kind of determination do they have? There is a different internal motivator for people who are successful at repaying their debt." If no one has refined the Grit Index the way Fair, Isaac handles credit scoring, there is a wide scattering of data, plenty of impressionistic material, and a whole world of experienced marketing people who are now trying to understand delinquent debtors. The National Foundation for Consumer Credit gathers a snapshot of data about everyone who seeks credit advice from its affiliates each year.
All credit-counselling clients are delinquent debtors. But they are a self-selecting group, those who have taken the initiative to see if they can find a way out of their thicket of bills.
Of that group, 52 percent reported in 1997 that the reason for their delinquency was simply "poor money management." Another 40 percent say a "life event" propelled them into debt: about 23 percent attributed their financial problems to unemployment or reduced income, 10 percent to a change in marital status-divorce or separation-and 7 percent to medical problems. Those numbers have remained stable within a percentage point or two for the past five years.
On the other hand, in the experience of collection company CFS, 80 percent of their customers say they have fallen into financial trouble because of a life event, double the counseling agency's average, according to Learned. There are other differences between the self-selectors and the rest. The average debt of someone seeking help from a counseling service is more than $20,000. At ContiAsset, the average debt is only about $1,500; at CFS it's $3,700.
Part of the difference between self-selectors and the average debtor is where they come from. Counseling services essentially take all comers. In 1997, almost 11 percent of NFCC's customers reported annual incomes of more than $50,000, and one out of six identified their work as professional, technical, or managerial. But most counseling clients skew toward the lower end of the income scale; the average household income for those seeking counseling clients in 1997 was $28,661, more than 15 percent below the national average household income. Although charged-off credit card debt, too, comes from all socioeconomic levels, most of the debt that companies like CFS buys is A-rated credit card debt-that is, the credit card holders who have stopped paying their bills were, at one time, considered the best risk among consumer borrowers.
West Capital Financial Services, a San Diego company, works much the way CFS does, buying high-quality, charged-off credit-card debt, then trying to work with debtors politely to get all or part of it paid. "Our experience shows us that of the people we deal with, only a small percentage will tell us to take a hike-10 to 15 percent, perhaps," says West Capital president and chairman Carl Gregory.
West Capital, too, knows what it doesn't know, and it is using the work of a Ph.D. statistician, hired as a consultant last spring, to refine the way it approaches customers. The statistician is working through data from West Capital's 3 million customers, looking for patterns among triggering events, among demographic data, and doing pure statistical analysis, to discover connections that the company might not otherwise think of.
Gregory says these efforts should improve not only his collectors' efficiency, but their effectiveness, as well. "As a result of all that'sgoing on, we believe we will get more money from certain accounts and require fe wer people to collect from others." Gregory, like others in the collection arena, wouldn't be specific about what kinds of connections between demographic data and likelihood of payment his company is discovering. But only a few months into the analytical effort, he says, "We use the results with decisions about who to call, and it is becoming more suggestive of what approach [with a customer] would be most successful."
Perhaps most surprising is that West Capital's statistician hasn't discovered any previous research of significance on delinquent debtors. "It's pretty astounding that there's all this debt out there, but there is not a lot of data," says Gregory.
West Capital is also working on a partnership with an outside data firm to see what value there is in matching its files against commercially available data on households and individuals. Although other kinds of companies have long used demographic data to better understand their clients (and although financial services firms use such information to solicit credit customers), using such data to understand delinquent debtors is still in its infancy. Only now are financial institutions learning to use data-mining techniques that are commonplace in other businesses.
Most collection companies, including ContiAsset, CFS, and West Capital, do use analytical models to score the debt they are buying before they buy it; that, in part, is how they decide how much to pay for it. But those models rely on the scant information found in the delinquent debt file itself: Is there a current address, current phone number, social security number? when was the last payment? how big is the debt, and how old?
"From the banks," says Learned of CFS, "we get almost nothing in the way of information. We know where [the debtors] live, perhaps a bit of collection history. We can usually deduce their gender."
Still, over time, Learned says, CFS has learned a lot about its customers. Only 60 percent of the credit files it buys, for instance, have accurate contact information. About 7 percent of the people CFS inherits go bankrupt.
"Another small group," says Learned, "perhaps 10 percent or so, never intended to pay the money back the moment they spent it. They are kind of a casual crook-someone, say, a college student, who gets offered a card and says, 'I'm going to max this out. What can they do to me?'"
The largest group, perhaps 80 percent, says Learned, "are people who have gone through some life-changing event, something that many times is beyond their control. But in the long-term, they want to pay." Those are the defaulters Learned wants to understand better.
Veteran debt collectors know intuitively that more information will change the collection process. "If we know that the debt we are buying is all VISA gold cards instead of regular VISA cards, that makes a difference," says Gary Wood, who buys consumer debt for Collins Financial Services in Austin. "Gold cards are typically only offered to people who meet some demographic and credit requirements, in excess of the basic card. So instead of collecting 10 percent of the debt, we might expect to collect 15 percent. It's a flag. If we knew that half the accounts were people with a college education, that would be helpful to us. If we could find out that the person owns his own home, that's a big bit of information to get hold of. Homeowners tend to be more responsible in taking care of their debt."
Although Trajecta won't talk about the specifics of the work it is doing, the modeling company has come to some of the same conclusions as Wood. "We know some things do matter," says Harris. "Things that indicate basic stability in life typically indicate a person will be okay on their debt." So snippets like how long a person has been at their address, how long they have had their checking account at the same bank, whether they own their home, how many times they've moved-if available, can help predict collection success or failure.
Trajecta has discovered that delinquents do look different in one area of their credit files: "Delinquents," says Mac McCarty, "tend to have fewer active credit lines"-than either good customers or those going bankrupt. "The average number of accounts opened by delinquent debtors is roughly half that of good or bankrupt account holders."
And Harris says an emerging area of study involves relating what people actually bought with their credit cards to whether they'll pay their bills. "If you've got someone using credit cards for gambling, that's an indicator that you're dealing with a real dangerous person."
When psychologist John Bachman arrived at CFS from California last summer, his starting place was a file of letters from CFS customers-500 thank-you notes from people happy with their "collection experience."
Bachman is using that batch of happy customers to begin his research into how CFS clients got into debt and what motivated them to get out. "My ultimate goal is to develop an algorithm from which I can predict payment behavior based on history," he says. Bachman's preliminary work already sounds different than the kind of scoring credit card companies do, although it betrays a faint echo of a much earlier era, when bankers knew their customers personally.
"As a psychologist," says Bachman, "I'm interested in looking at personality characteristics as underlying motivation." So he is thinking about psychological categories such as "competitive consumers," people who buy to maintain a lifestyle similar to their peers and whose overspending comes from an aggressive, competitive personality. Another classification he calls "compulsive overspenders," people who feel empty inside, hungry, never satiated, who buy in pursuit of feeling full. And then there are "codependent spenders," people who buy things to take care of others, to make them dependent.
Bachman isn't sure how such categories will be useful in collection. "The question is, if we get a customer who has these kinds of personality issues, what kind of account officer can we get to work with this person?"
West Capital's Carl Gregory makes the same point, but from a different perspective. Although debt collection is an ancient trade, it is still in many ways a cottage industry, with thousands of small practitioners. The new focus on understanding delinquents simply reinvents and technologizes a process common a generation ago.
"It wasn't that long ago," says Gregory, "that if you wanted to borrow a thousand bucks, you had to walk into a bank, sit down across from a guy at a desk, show him your W-2 statement, maybe your taxes, and he would say yes or no." Often, the yes or no was based as much on the banker's knowledge of your character-his intuitive understanding of your demographics, psychographics, and Grit Index-as on the papers you brought with you.
Today, says Gregory, "when you want $1,000, all you have to do is open your mail."