Quantcast

LoanPerformance Mines Unstructured Data For Collections

Increasingly, credit-card lenders are using "unstructured data" to help guide collection and default management decisions, and now the trend is making some inroads in the mortgage arena as well.

LoanPerformance here, in conjunction with Intelligent Results, recently launched ScoreText, a product that allows lenders to turn textual information gleaned from conversations between borrowers and collection counselors into data that can be used to estimate the risk of default or foreclosure. While default management has been the prime use of text scoring, the data can be used for a host of other purposes, such as customer-relationship management, portfolio retention and marketing campaigns.

Richard Harmon, senior vice president for scoring analytics and services at LoanPerformance, said that most scoring systems that analyze borrower behavior rely upon structured numeric data, such as credit scores and payment information.

ScoreText allows lenders to augment these traditional systems with information and observations generated by other forms of interaction with the borrower, including free-form text and collectors' notes as well as e-mails from the borrower.

These "mixed data analytics" can help lenders refine collection and default management decision making, Mr. Harmon said in a recent interview. While traditional, numerical data models have been improved, those improvements yield "significantly diminishing returns because we are basically extracting all of the signals out of the data that we can get."

That unstructured, textual data allows lenders to incorporate what the borrower is actually saying to collection counselors into the scoring methodology. It also allows the collection agent's notes, such as indications that the borrower is lying, to be incorporated into the proprietary scores as well.

"It gives you insight into what the customer is saying as well as what they are doing," he said.

The models are customized and proprietary to each institution that is using ScoreText. Loan servicers can use the scores to standardize how collection agents and managers deal with loans with a specified score at a given point of delinquency across the institution.

The ScoreText product has been in development for at least a year-and-a-half. LoanPerformance said it has several customers already using the product.

Damien Weldon, director of text mining solutions at LoanPerformance, said that ScoreText is designed as a forward-looking, predictive tool that augments existing numeric data models for analyzing likely behavior. The model essentially helps lenders predict, for instance, the likelihood that a delinquency will proceed to foreclosure within a certain timeframe.

"This whole approach is very much grounded in the view that people pay off loans, loans don't pay off themselves," he said.

For instance, if a borrower has a roommate and rental income is paying all or some of the mortgage, that may affect the performance of the loan. Words such as "bankruptcy" or "fired" also factor into ScoreText's model, as do less traditional factors such as whether or not the borrower owns a motorcycle. ScoreText helps lenders leverage information gleaned from call centers and other customer contact points to understand those customers better, he said.

"It gives them a much better read on what's going to happen with that account," Mr. Weldon said.

The technology also can be used to pick up signals that a borrower may be planning to move, obtain a home-equity loan or make other life changes. Those factors can be harnessed for customer-relationship management purposes, he said.

Copyright 2005 SourceMedia, Inc. All Rights Reserved. http://www.sourcemedia.com http://www.mortgageservicingnews.com