CREDIT & COLLECTIONS scoring software can reduce these cycles through automated development tools, prebuilt end-to-end workflows, model management and deployment directly into data warehouse appliances. One large South American bank using modern credit scoring software reduced modeling dataset development time from 75 days to under one hour, and total model deployment time from 18 months to three months for a customer base of more than 100 million accounts.
Credit scoring data: what’s new? Using unstructured external data from social networking platforms and modeling that data using text analytics can predict character (one of the Five C’s). Analyzing the customer’s number of connections, memberships in professional organizations and related data points may provide improved predictions about creditworthiness as well as potential for fraud. Unstructured data can be analyzed by modern text mining software, with the results incorporated into credit scoring models. An often-cited McKinsey study identified external data sources from telecom providers, retailers and wholesale suppliers as valuable proxies for judging a customer’s ability and willingness to repay and for confirming identity and income. McKinsey also highlighted the value of the financial services providers’ own, often unused, internal data. Their report stated that “many large banks still do not take into account information as simple as the balances and transaction patterns of their own customers’ checking and savings accounts when assessing creditworthiness.”2
Credit scoring modeling: what’s new? Interest in machine learning, a computeintensive method of data analysis that automates model building, has risen recently for credit scoring and other areas of financial services. Part of the interest comes from the availability of increased and cost-effective computing power along with large amounts of available memory. Industry research is focused on the use of neural networks, decision trees and random forest techniques for credit scoring. Today, most production users of credit scoring software use logistic regression
techniques because they deliver excellent results and are well-understood. Since credit scoring models are typically “regulated” models, banks and other lenders must be able to explain, validate efficacy and defend usage to regulatory authorities. Machine learning models can become so complex that modeling staff cannot explain “why the model does what it does” to regulators. Modern credit scoring software provides a wide choice of predictive modeling techniques, including machine learning, so that firms can identify the techniques with the highest “lift” (predictive power) and then determine what technique should be used in production scoring.
Real-time decisioning The use of credit scoring models began decades ago when computing was almost exclusively batch processing. Today, banks want to accelerate the opportunity for profitable lending revenue through realtime credit authorization decisioning for many types of loans, including credit cards, personal and automobile loans. Modern credit scoring software can integrate with real-time decision management tools to provide this capability. An eastern European bank interested in growing its small-amount credit business by focusing on lower-income segments combined modern credit scoring, real-time decisioning and marketing automation software to increase loans by 25 percent and achieve ROI of over 175 percent.
Collections It may seem strange to view collections as a competitive activity. However, all borrowers have a personal payments hierarchy for determining which bills to pay first in times of financial stress. Collections can be thought of as competing for “share of wallet,” similar to the sale of new financial products and services. A best practice for collections is to start as early as possible in the delinquency cycle, given the likelihood of recovery will decrease over time. Collections scoring models can help identify the likelihood for delinquent customers to “cure” across various communications methods. These methods could include contact center phone calls, automated IVR, emails, SMS messages or “watchful waiting” (self-cure) by the next
billing cycle. Banks have to balance the use of these communications methods with capacity constraints and costs. Mathematical optimization, which seeks to maximize or minimize an objective while considering real-world constraints, is an excellent method to obtain the “best” answer to the collections problem. All collections customers, and their propensity to cure using various communications channels, can be analyzed together with capacity and cost constraints to arrive at the “best” or optimal communication channel for each customer. A large Asia Pacific bank used modern optimization software to significantly increase the cure rate (reducing outstanding delinquent debt), delivering a 300 percent ROI and a payback on their software investment in less than six months.
Wrapping up Modern credit scoring software lets banks speed up the model development and deployment cycle, leading to faster credit quality assessment; the end result of which is increased revenue from lending and reduced credit risk. New types and sources of data can be incorporated into the scoring process, and new modeling techniques are also available. Credit decisioning can now be real-time, and collections can increase while minimizing costs. Implemented together, these new capabilities can modernize a financial institution’s more traditional credit and collection processes to compete with the flashy new tactics from FinTechs. As 2016 begins, consider how a modernized credit and collections program can garner more customers, more revenue and more profits. David M. Wallace is global financial services marketing manager for SAS with responsibility for defining industry strategy for banking and capital markets. He has over 30 years’ experience in the application of information technology to solve client needs, including a focus on financial services for over 20 years. Wallace holds a Bachelor of Science in economics from the University of North Carolina Wilmington and a MBA from East Carolina University. He is a member of GARP, PRMIA, and SIFMA Compliance & Legal Society, and is also the chair of BAI’s Solution Provider Executive Council. 1 University of Edinburgh Business School, Biennial Credit Risk and Credit Scoring Conference, survey of 200 delegates from 40 countries, August 2015. 2 McKinsey and Company, New Credit-Risk Models for the Unbanked, April 2013
Financial Operations | WINTER 2015 | www.financialoperations.ca