HOW PREDICTIVE ANALYTICS DRIVES PROFITABILITY IN ASSET FINANCE By Janet Orrick, Analytic Scientist at International Decision Systems EXECUTIvE SUMMARY In today s ever-changing business world, asset finance organizations face many critical, high-risk decisions every day. In order to grow profitably and outpace the competition, they must increasingly make those decisions with rapid precision. As such, more and more financial organizations are relying on the use of predictive analytics to speed the decision-making process without increasing risk exposure. Predictive analytics is the mathematical use of historical data to predict future behaviors and outcomes. It can help asset finance organizations make smarter decisions, such as: Approving good customers who might otherwise have been declined, resulting in incremental revenue. Declining risky customers who may previously have been approved, resulting in lower average loss rates. Estimating the amount of loss a customer will accrue if it defaults on the loan/lease agreement. Identifying opportunity to cross-sell or up-sell products to the customers who are most likely to respond and pay for these products. Providing an accurate estimate of residual value at the time of the booking to determine optimal pricing and term positioning. This white paper provides an overview of the science of analytics and demonstrates how it can improve decision making throughout the asset finance industry. It includes a compelling case study that describes how analytics helped one company increase its revenue and decrease its loss exposure dollars. WHAT S PAST IS PROLOGUE. Antonio in Shakespeare s The Tempest
PREDICTIvE ANALYTICS PAST, PRESENT, AND FUTURE Predictive analytics uses historical data patterns to predict future behavior. Companies use analytics to help them fi nd solutions to key questions that can signifi cantly infl uence future company performance. For example, the goal of an online retailer is to gain a greater share of the consumer s wallet by making real-time product recommendations while the consumer is shopping at their site. To do this, retailers utilize analytical strategies based on predictive models to identify what products a consumer is most likely to buy and when he or she is most likely to buy it. This is a form of predictive analytics that takes what little information the online retailer knows about the shopper and compares that information to like characteristics of others who have purchased similar products in the past. As the consumer s purchase frequency increases on that site, the online retailer s predictive models become more accurate, presenting the consumer with more products that he or she are likely to buy. Predictive analytics can be applied in a similar way for asset fi nance organizations. The goal of the asset fi nance company is to maximize revenues while reducing loss exposure. By identifying characteristics of profi table customer groups, new customers with similar characteristics can be targeted for future offers. Conversely, identifying characteristics of non-profi table customers, such as those with high delinquency, enables asset fi nance organizations to exclude or not approve like customers from future offers. The use of predictive analytics also serves to give companies a broader perspective when making important credit decisions. In the past, for example, many fi nancing organizations concerned with risk exposure have declined potential customers simply because they may have been a day or more delinquent on other trade line payments. By approving fi nancing only for those customers who have paid all their bills on time, they focused solely on delinquency and not profi tability leaving potential revenue on the table. Predictive analytics would have identifi ed these moderate risk customers with high revenue so that the company could maximize profi t. Predictive analytics also helps asset fi nance organizations look at their data in other meaningful ways. It can identify factors that may have contributed to why something happened (wrong or right) and point to actions that will help lower risk and increase profi tability going forward. For example, when underwriting an asset or set of assets, predictive analytics can help an asset fi nance organization with questions such as: How will a customer perform in the future? When booking a contract, what is the likelihood that the customer will pay? What is the probability that the customer will buy or return the asset at lease end? What is the potential for cross-selling to the customer? The use of analytics helps asset fi nance organizations with these types of what if scenarios for a specifi c account or type of assets. The accuracy of these predictions is based in the quality of the data. By studying the past, predictive analytics can help asset fi nance organizations sharpen their focus on important business criteria and realize more value in the future. IDENTIFYINg DECISION POINTS The fi rst step in predictive analytics is to determine decision points or issues that infl uence profi tability. By identifying decision points, historical performance data can be analyzed to predict future outcomes along the way. In the asset fi nance industry, the origination and portfolio management processes each have their own unique set of decision points that may include: ORIgINATIONS DECISION POINTS Is this customer credit worthy? Will this customer accept or reject the offer? What is the highest price this customer will pay? Will this customer pay on time? Will this customer default on payment? How much exposure can the organization sustain? How much profi t will this customer generate over the lifetime of the relationship? Is this customer likely to revolve? What is the potential outcome at the end-of-lease term?
PORTFOLIO MANAgEMENT DECISION POINTS What are the chances this customer will default in the next six months? Is there cross-sell potential? Where should the collection department focus their limited resources? Does a delinquent customer deserve a second chance? Where is the best place to focus marketing dollars? Is the residual valuation accurate? Once a decision point is identifi ed, a unique target variable is created and applied to fi nd a solution. For example, an organization may fi nd characteristics of delinquent customers who are 60 or more days past due within an 18-month period. Then, they can apply the same characteristics to new accounts to identify the potential for customers to go 60 or more days past due in that same period in the future. Those accounts with a high potential for delinquency are declined an offer. Each target variable is like a lever you can turn up or down based upon key business objectives. These targets could include companies that have potential to generate high revenue or likelihood of end-of-lease renewal. The ultimate goal is to use those levers to identify the optimal profi t solution. USINg STATISTICAL MODELINg Historically, underwriters created a set of rules to determine if a customer was credit worthy. These rules were often based on speculations and subject to human interpretation. Statistical modeling, on the other hand, takes raw attributes (variables) and formulates an equation. The equation produces an outcome, or a score, that predicts where a specifi c customer can be expected to rank relative to their peers. This is an outcome of predictive analytics. A model creates a score value that ranks customers based on a key objective and predicts the likelihood a customer will fall into a specifi c category. For example, when looking at a potential customer, their past success (e.g. payment history) can predict their future behavior. The objective of a predictive model is to identify which variables are most predictive of future outcomes and the weighting of those variables to best fi t historical and future performance. The combinations of variables that can be used in the analysis are endless, so it is often challenging to determine where to apply a cut-off and reduce the risk exposure. For example, does a small company in business for more than 20 years have more potential to be successful than a large company that has grown rapidly over the past three years? Without statistical modeling, the difference between these two customers can seem indistinguishable. Several types of modeling techniques are available. The most explainable approach to modeling produces a scorecard model that groups the predictive elements and assigns points to each group and weights to each attribute. The outcome to this equation is a number, or score. A scorecard format allows the underwriter to view the variables that drive the resulting score. See Figure 1 for an example scorecard, which includes the following predictive attributes: Payment History The customer has made at least 8 payments. Payment Method The customer has used credit for each payment. Number of Trades The primary owner currently has 15 trades on the bureau. Based on the above attributes, the customer score card would refl ect a fi nal point score of 71.4. ScOREcARD VARIABLE FIELD ID WEIGHT RANGE POINTS PAYMENT HISTORY Pay. Hist 40% 1 to 5 20 6 to 20 75 21 to 30 100 PAYMENT METHOD Pay. Meth 40% Debit 45 credit 100 check 20 cash 5 NUMBER OF TRADES Ext. Trade 20% 1 to 5 3 6 to 20 7 21 to 30 25 31 to 100 50 101+ 100 customer ScORE 71.4 FIgURE 1: SAMPLE SCORECARD The model equation would be (75 * 0.4) + (100 * 0.4) + (7 * 0.2) = 71.4 score value Many types of statistical models are available including neural networks, regression, decision trees, and Bayesian or Artifi cial Intelligence. A statistics modeling analyst will defi ne the best modeling approach based on the business key objectives or decision point questions that need to be answered.
CASE STUDY The following case study focuses on an independent, small-ticket asset finance organization (referred to as Company ABC) and shows how predictive analytics can dramatically impact its loss exposure and revenue projections. Background Company ABC was generating an adequate volume of business but was concerned with their rising loss rate at 6% or approximately $1.1 million annually. The company was pleased with its current approval rate of 70% and wished to maintain this volume of approved accounts into the future. They currently used a manual approval selection strategy that included a list of recommendations for approval, but each underwriter was free to approve or decline an application based on their own judgment. Each deal took an average of three hours to review. Company ABC s objectives were to speed the review process, reduce the number of delinquencies, and continue to maintain the 70% volume of approved accounts. To improve this process, a predictive model was created specifically for Company ABC s business and applied to the selection criteria. This model was designed to predict customers who were likely not to pay their bill within 60 days of the due date over an 18-month period. While increased loss rates were a concern, Company ABC knew there were additional cost benefits for improving their manual approval process. By adding an automated process, Company ABC found it could target a low-risk group for automatic approval, target a high-risk group for automatic decline, and manually review the remaining applicants using the current process for approval. This enhancement was chosen to save processing time, increase approval response time, and maintain a 70% approval rate. By speeding up their approval process, they could also decrease their cost per application and quicken their time to market. Results from Comparison Testing To determine if a predictive model could improve upon the company s existing process, the population was ranked by score value and split by rank into five equal segments each containing 20% of the population. The average loss rate was calculated within each segment. Before applying the model to the data, it is important to note the average overall loss rate was 6% as represented by the gray horizontal line in Figure 2 below. This means that if no model was applied, the average loss rate, assuming a random selection, would be 6% in every group. When applying a predictive risk model, the wider the spread between the lowest and highest segment loss rates, the better the model is able to classify the characteristics of bad customers. The dark blue line represents the segmentation of the predictive model with an average loss rate for each segment ranging from 1.7% to 15%. In the highest risk segment, the model identified 152% more bad customers (difference between 6% and 15%). LOSS RATE LOSS RATE (60+ DAYS PAST DUE WITHIN 18 MONTHS) 16.0% 14.0% 12.0% 10.0% 8.0% 6.0% 4.0% 2.0% 1.7% 0.0% 1 Low Model 2 Manual 2.9% 3.7% 6.5% Based on these findings, Company ABC created a strategy to automate the approve/decline decision given the following rules: Maintain the current approval rate of 70% or higher Auto approve applications that fall within segments with less than a 4% loss rate Auto decline applications that fall within the highest loss rate segment (20% of population with 15% bad rate) Continue to manually approve all other applications 15.0% 5 High Business Outcomes Given these criteria, the models would decline the top 20% of applications in an automatic approval decision, 60% of the population would automatically be approved, and the remaining 20% of the population would be reviewed using the manual process. This increase in automatic approvals results for each model is displayed in Figure 3 below. 3 RISK SEGMENT Figure 2: Loss rate for each model by population segment MANUAL SEGMENT % OF POP LOSS RATE SEGMENT % OF POP LOSS RATE AUTO APPROVED 0% 0.0% 1,2,3 60% 2.8% Manual REVIEW ALL 100% 6.0% 4 20% 6.5% AUTO Declined 0% 0.0% 5 20% 15.0% 4 MODEL Figure 3: Selection results for model strategy
As Figure 3 shows, the predictive model identifies 60% of the population with an average loss rate of 2.8% or less for automatic approval. The model also reduces manual application review to 20% of the population with an average loss rate of between 2.8% and 6.5%. The worst risk segment, with an average loss rate of 15% or more, would be automatically declined. Auto-Review Savings Using the above predictive model, Company ABC would now process 80% of its applications automatically. With 11,000 applications processed in this period, this improvement alone could generate huge savings for the company as outlined in Figure 4 below using the following key assumptions: Cost Assumptions Average cost per man hour is $50 Average time to automatically approve an application is 30 minutes (or $25 per application) Average time to manually review an application is three hours* (or $150 per application) *based on information from the 2010 ELFA Survey of Equipment Finance Activity Origination Cost/Appl manual model Automatic $25 $ $222,700 Manual Rev $150 $1,668,450 $332,250 TOTAL $11,123 $1,668,450 $554,950 Improvement $ $1,113,500 Figure 4: Savings using automatic selection process The manual review process costs over $1.6 million per 11,000 applicants. Given that 80% of the population would flow through the automated process using the predictive model strategy, the estimated savings from the model is more than $1.1 million or a decrease of 67%. Reduced Loss Rates In addition to time and dollar savings from an automated approval selection process, the right predictive model provides further advantage by reducing loss rates. By identifying customers with a greater potential to go delinquent, a predictive risk model declines the worst 30% of the population and approves the best 70% of the population with a greater likelihood to pay. As shown in Figure 5 below, this results model produces a lower average loss rate of 3.2% for Company ABC, significantly lower than 6% rate seen with the previous manual process. Loss Assumptions Average loss is $2,500 per bad account (using estimated 10% of $25,000 average asset value) Worst 30% will be declined (based on a 70% approval rate) loss $ / Bad manual model top 70% LOSS RATE 6.0% 3.2% BAD COUNT 467 251 BAD TOTAL $2,500 $1,167,915 $626,409 Improvement 0% 46% Figure 5: Loss savings gained by model Figure 5 shows that when manually reviewing all its applications, Company ABC experienced a total expected loss of more than one million dollars. By using a predictive risk model to identify and remove the highest risk customers, however, the company can lower its loss by 46% and significantly increase its revenues. By combining both the savings from automated decisions and the savings from decreased customer loss rates, Company ABC stands to gain more than $1.6 million dollars by incorporating predictive analytics into its business process (see Figure 6 below). manual Figure 6: COMBINED savings (ADDING FIGURES 4 AND 5) model total COST $2,836,365 $1,181,359 COMBINED SAVINGS $1,655,006 In addition, applying predictive analytics to additional key objectives such as reduced fraud accounts, higher response, and increased lifetime value could grow profit margins even further.
CONCLUSION In the asset finance industry, historical tracking is necessary to conduct performance and trend reporting and adjust corporate strategies as companies move forward. Similarly, predictive analytics turns historical information into a forward-looking science that can greatly influence current decisions and forecast future results. In today s business climate, it is more important than ever for asset finance organizations to use every financial tool at their disposal to gain a competitive edge. As such, predictive analytics should be considered a critical device for companies to automate business processes, improve decision making, boost profitability, and combat risk. With the use of predictive analytics, asset finance organizations can develop a consistent and repeatable decisioning process that will yield them a high rate of return now and in the future. Janet Orrick, Principal Analytic Scientist at International Decision Systems, brings with her more than 15 years of strategy management and predictive analytics, applying statistical modeling to develop innovative solutions to business problems. Prior to joining IDS, Janet held an assistant vice president position at HSBC working with the Modeling and Analytic Solution group where she developed risk, response, and revenue models for the auto, mortgage, and retail credit card divisions. Janet found solutions for a variety of legislation issues including Regulation B, Basel II, and Dodd-Frank. Prior to HSBC purchasing Metris Companies, Janet led the originations modeling group at Metris where she earned the Award of Excellence for her efforts with Hispanic Marketing. Prior to this she managed the list rental and origination marketing efforts at Fingerhut Corporation. Janet Orrick holds a Master s of Business Administration (MBA) from University of Phoenix with a Bachelor of Science in Math and Business from Concordia University, Moorhead, MN. International Decision Systems is the leading provider of software and solutions for the asset finance market. For over 30 years, the company has offered integrated solutions from origination through disposition and asset management, supported by an ongoing research and development effort unrivaled in the market space. Customers include approximately 50% of the largest leasing companies in the United States, and an increasing number of the largest global players. In addition to offering powerful, flexible software, International Decision Systems provides the industry experience and expertise to assure successful, complete solutions. 2012 International Decision Systems 1500 IDS Center 80 South 8th Street Minneapolis, MN 55402 UNITED STATES OF AMERICA Tel: +1.612.851.3200 / Fax: +1.612.851.3207 / www.idsgrp.com UNITED KINGDOM / INDIA / SINGAPORE / AUSTRALIA