With traditional buffers falling by the wayside and consumer habits changing, companies need to update their collections strategies.

by Shrikanth Jagannathan, Yogesh Chati

In the heydays of sub-prime lending, underwriting criteria were more relaxed, as rising home prices provided a buffer against default. As housing prices rose, lending standards were lowered, and new products were designed to keep buyers from being priced out of the market. To keep the lending boom going, borrowers were offered riskier products with innovative features such as no money down, no income verification, interest-only payments, and even negative amortization loans. Reliance on automated tools, without the necessary checks and policies to govern their usage, helped create new loans and approve them in minutes. The hot securitization market, where mortgages could be pooled, rated, and sold to investors, abetted this trend. Once the loans were securitized and off the books, the loan initiators were free to go ahead and originate more loans.

Then, a convergence of several factors—a decline in housing prices, a moderate increase in mortgage rates, ARMs for sub-prime loans getting reset, incomes stagnating, decline in mortgage originations—resulted in the so-called “sub-prime mortgage crisis.” There is a growing concern over the quality of credit on the lenders’ portfolios, given the increase in late payments, defaults, and foreclosures. In a recent quarterly Federal Reserve survey of senior loan officers, more than 70 percent of U.S. respondents expected the quality of their prime, nontraditional and sub-prime residential mortgage loans, as well as of their revolving home equity loans, to deteriorate in 2008. The reasons are not hard to find.

• U.S. consumers are filing for bankruptcies at record rates. The total filings for the 2007 calendar year reached 801,840, an increase of nearly 40 percent over 2006.

• More than 1 million sub-prime ARMs are likely to reset during 2008, which could lead to an even higher number of bankruptcies.

• The average U.S. consumer is highly leveraged, and consumer debt is increasing faster than incomes. According to the Federal Reserve1, the total consumer debt stood at $2.5 trillion as of December 2007, of which $1.57 trillion was non-revolving debt.

• The ratio of household debt relative to disposable income is at a record high and is projected to be above 150 percent in 2009.
Consumer behavior is changing, as well. The average customer is now younger and more demanding. Surveys indicate that aggressive collection tactics can push customers toward filing for bankruptcy. A negotiated settlement of payment issues is preferred rather than just going by contractual agreement.

Focus on Managing Portfolios
These trends call for new ways to improve bottom lines and enhance profits. We believe that one of the top areas to focus on is improving collections strategies so that you get more “bang for the buck.” In the sections below, we lay out a framework to leverage the power of analytics to substantially increase the efficiency and effectiveness of your collections strategies.

An Analytical Framework to Drive Collections Strategies
Infosys believes this can be achieved by harnessing the power of analytics. Predictive tools can be used accurately to analyze each account and to identify the most effective strategy for that account. This will help optimize collections at each stage of the life cycle—from pre-delinquency account management to early stage and late stage collections to vendor management for recoveries after charge-off. The objective is to collect more dollars more efficiently under given business constraints, thereby reducing the cost per dollar collected.

Credit scores have long been used as one of the crucial inputs to a mortgage underwriting decision. Adoption of similar techniques has been slower on the collections front. We have seen several organizations invest large sums to enhance their operations. Although a collection organization might be primed to work at optimal levels, analytics still uncover hidden areas to optimize collections recovery.

Below, we will describe a framework that will help organizations leverage predictive analytics to gain the insight needed to develop targeted strategies for collections. Implementing such a framework can be a “force multiplier” when one considers the benefits reaped in terms of increased collections and reduced costs.

Targeted analytics
Several collection agencies still treat all cases the same way, and the amount of the outstanding balance is used to prioritize accounts. Other agencies employ simple variations of prioritizing like balance, days past due, and number of times delinquent in the recent past. Many agencies have begun using behavioral scorecards to prioritize which delinquent accounts to call.

While these approaches are still useful, with mushrooming numbers of delinquent accounts, amounting to millions of uncollected dollars, we see the need for deeper analytics that predict responses to particular actions—whether sending a reminder, making a call, or placing an account with a recovery agency.

Targeted, action-specific models can improve cure rates, reduce roll rates and charge-offs. Collections call center efficiencies also improve with reduced call volume and higher effectiveness. We have also observed that targeted actions, such as not calling self-cures (those who are likely to pay without receiving a call), improve customer satisfaction and retention levels. Action-based analytical models typically have a quick return on investment, with payback often in just a few months.

Analytics is a vast and complex area, and one needs a clear roadmap to go up the analytics curve. A collections organization should evaluate its current maturity in terms of data availability, modeling complexity, and organizational acceptance of analytics. This will enable it to determine the appropriate level of analytics that it can accept and support.

Profiling, Segmentation and Scorecards
The start would be to understand the kinds of customers you have and what their behaviors are. Data analysis and basic analytics are already being used to identify distinct customer segments. These segments are formed from the analysis of the customer profile and linking it with historical behavior data.

Further analysis can be done to profile these customer segments, understand their characteristics, and identify those segments that are a cause for concern.
However, just knowing your customer is not enough. The desired objectives for each segment are different and, hence, so should be the strategy employed.

For example, in the early stages of delinquency, the percent of self-cures can be 30 percent or higher. This could be due to common behavior traits such as habitually mailing the check out on or after the due date. Identification of self-curing accounts helps collections agencies to focus their efforts on accounts that need them. Traditional scoring strategies can then be used to classify the remaining delinquent accounts into high-, medium-, and low-risk categories. In the early stage, collections call centers call high-risk accounts, while low-risk accounts are not called.

Similarly, for late-stage delinquencies, behavioral scoring can help identify specific actions that will deliver the best results. For example, consider an account that has large dues, but a high likelihood of that payment happening only after a collection agency calls. One can immediately flag this account for placement with an external vendor, without a prior call or letter. Such a decision made early could result in collecting dues while the debtor still has funds available and before other lenders reach him/her.

Multi-dimensional Analysis
Segmentation and behavioral scoring are, in a sense, one-dimensional. A scorecard can identify an early stage account as high-risk, but may not determine the specific treatment that would be effective. So, all high-risk accounts would get assigned one action, say a phone call, while all low-risk accounts are assigned a “no-call” action. The collections center then misses the opportunity to call low-risk accounts that need agent assistance to pay or vice versa, leading to inefficiencies and ineffectiveness. Multi-dimensional analytics, on the other hand, provides deeper insight by allowing us to see the account from multiple angles—such as risk, propensity to pay, response to a particular action, and so on. This helps the collections agency to create specific actions to better target each particular account.

In a recent project for an Infosys client, we scored customer accounts on two dimensions—the propensity for a charge-off and the responsiveness to a collections agency call. Six segments with increasing scores were formed for each dimension. Deeper analysis allowed us to take the 36 segments and put them in eight clusters. Once we profiled the characteristics of each cluster, we were able to develop specific actions targeted at each cluster. Thus, what would have been a “one-size-fits-all” strategy using one dimension was replaced by a better targeted strategy using two dimensions.

Strategy Optimization
Multi-dimensional modeling is highly effective in evaluating trade-offs between strategies targeted at specific segments. The next level of complexity would be to create models that can select an optimal strategy from a host of
alternate strategies. Automated algorithms are deployed to determine the most successful action for each case, based on its precise circumstances and history. Such “decision models” create heuristic relationships between actions, responses, and results.

Optimization models require great amounts of trust in the model methodology and the model developers—before they can be adopted. It can lead to perceptions of adding mindless automation and removing control from “experts.” Though it can provide great returns, the investment in terms of management oversight and change management is non-trivial; hence ,strategy optimization would still be considered a technology for the future.

The Modeling Process
Analytical modeling is a data-intensive exercise. Good data collection and storage are the best preparation for any modeling project. In addition to internal account performance data, models would also require external data. The greater the extent and variety of data available, the more powerful are the models.
The data would be made up of several variables—amount due, number of days due, number of times past due, credit bureau score, income, and so on. Each variable tells us something unique about the customer. Using statistical techniques, such as logistic regression, the analyst would identify those few characteristics that influence the behavior being modeled, and carefully estimate the influence of each characteristic. The model performance would be tested on out-of-sample and/or out-of-period data before being put into production. The production data should be captured for future use.

An important part of model governance is model monitoring. Performance of statistical models typically degrades with time, due to various reasons including changes in collections portfolio and in account behavior. Therefore, models have to be periodically tested and recalibrated or remodeled, Constant model monitoring, rebuilding and periodic experimentation ensure that changes in the portfolio are captured and new predictive characteristics are included. This ongoing process helps collections call centers to test new actions, analyze new populations and enhance their strategies.

In conclusion
Leveraging analytics is a continuous process, and one needs to have a plan to enable analytics to improve business metrics. As organizations go farther up the analytics curve, they will find that rewards multiply with
minimal incremental investments.

Shrikanth Jagannathan is a manager, analytics practice of knowledge services, Infosys BPO. Yogesh Chati is a senior associate with Infosys BPO Knowledge Service.