Analytic Learning Loops Propel Bankcard Growth Turn your acquisitions and originations processes into an analytic learning powerhouse for boosting card revenues, margins and profits Number 51 June 2011 In dynamic markets, there s a direct relationship between profitable portfolio growth and a bank s ability to quickly understand and adapt to changing consumer behavior. Economic volatility, new regulations and competitive forces all make learning speed and yield critical to success. Faster analytic learning is within the grasp of banks of all sizes. An analytic learning loop can be created around existing systems and deployed in just months. Analytic learning loops which can be added to existing systems enable banks to make decisions based on how consumers are behaving in their markets today. By accelerating feedback about market performance, they reveal variances between what was expected and what happened. Rapid analytic learning helps banks target the right card products to consumers most likely to respond and use the card in a manner that generates profits. It enables them to adjust acquisitions and originations decision strategies to boost results while campaigns are still ongoing. Card issuers that cultivate analytic learning will be tough competitors. They ll become extremely adept at bringing consumers in targeted markets innovative products that serve their needs and stimulate usage and loyalty. This white paper: Discusses why traditional methods are retarding profitable portfolio growth. Explains the fundamentals of analytic learning loops. Describes flexible implementation options. www.fico.com Make every decision count TM
»Growth» in Unsettled Times How do banks grow their card businesses in markets that seem to be in a permanent state of transformation? The challenge of growth amidst change is especially clear in post-crisis markets. Banks coming out of retrenchment are faced not only with new regulations, but with consumers who have become careful about taking on new cards and savvy about how they use them. Card issuers are locked in battle over the same population of low-risk, high-value customers. Banks need to keep every one of these customers they re able to book and incent them to use their cards more often in ways that are most profitable for the issuer. They also need to expand their prospecting pool beyond these obvious targets without increasing risk exposure. Even markets that avoided the worst shocks of the crisis, including those where growth never stopped, are increasingly competitive. Banks in these markets need an early warning system to alert them to the effects of new competitors and technologies, and changing consumer attitudes and preferences. Nor is any market impervious to future waves of global economic turbulence, even if that prospect seems remote at present.»» Fixed Views of Fluid Markets Traditional approaches to bankcard acquisitions and originations, based on analyzing historical behavioral data, provide a fixed view of customer risk and reward potential. In today s markets, however, customer behavior is changing, so this fixed view is less helpful for understanding potential risk and reward. The situation is something like this familiar scene in films and television shows: A guard watching a live video monitor is fooled when a segment of video is copied and looped in place of the live feed. The monitor shows the scene unchanged, even though there s actually a lot of activity going on. For bankcard issuers, the result of relying on a fixed view of customer behavior is that decisions are never as accurate or effective as they should be. Growth is never as strong as it could be. Growth retardants Stale data. Some of the data used to build predictive models and segmentation analytics may already be more than a year old at deployment. Customer behavior patterns, more variable than they used to be, can quickly move out of alignment with existing models and strategies. Loosely fitted offers. The limited range of analytics typically used for acquisitions doesn t provide a precise enough way to determine which products will be the best fit for customers before making an offer. Lack of visibility into changes between prescreening and originations. New customer decisions usually don t take into account any factors that have changed since the initial promotion went out. Has risk risen or fallen? Is there a different product that is now a better fit? Slow feedback from originations. It usually takes many months before acquisitions receives feedback from originations. During that time, there s no visibility into whether a campaign is successfully attracting customers who drive profitability. No inkling of whether these customers are showing early signs of using products in the way the bank intended. By the time the feedback arrives, it s too late; the world has changed. www.fico.com page 2
»»»» Decisions Based on What Consumers Are Doing Now For faster, stronger and more sustainable growth in dynamic markets, banks need acquisitions and originations decisioning to reflect what s going on in card markets now. The way to achieve this without organizational and infrastructure upheaval is to thread an analytic learning loop through existing processes. Figure 1: Analytic learning loop concept ACQUISITION An analytic learning loop provides constant feedback on the performance of decision strategies in operations and early warnings of changing market and economic factors affecting customer behavior. It s a shared resource that enables acquisitions and originations to learn from each other s results and work toward common profit and loss (P&L) objectives for driving portfolio growth and profitability. Offers While analytic learning loops don t eliminate uncertainty, they do help banks become more effective at managing customer decisions in the midst of it. By providing a systematic approach to learning from recent results, they improve the ability of bankcard issuers to: Design innovative products customers need, based not on their past behavior, but on how they re responding to and using bankcards right now. Analytic Learning Hub Deliver more effective pre-approved lead lists. Richer customer profiles enable more granular and insightful segmentation of prospect populations. Higher-performing lead lists boost the performance of marketing campaigns and the productivity at call centers and branch offices. New Accounts ORIGINATION Improve response rates and early-life account performance by pinpointing individuals who are not only most likely to accept a specific offer, but also to use the product in a manner that is profitable for the bank (e.g., high usage and low balance for a product designed for transactors). Boost ROI on marketing campaigns by spotting variances between what was expected and what actually happened, determining sources of variance, and adjusting profiles, segmentation, offer strategies and approval decisioning to improve results while campaigns are still going on. It s also important to point out that an analytic learning loop provides only the mechanism to accelerate learning the degree of success achieved depends on how banks use it. The biggest winners will be companies that employ their analytic learning loops to: 1. Conduct designed experiments, whose results can be accurately analyzed and causes of variations understood. A well-designed series of experiments tests carefully chosen challenger www.fico.com page 3
»» decision strategies against the current champion. There s no need to test every option or even all the best options because enough is being learned to extrapolate some outcomes. 2. Probe beyond the edges of business as usual (BAU). A well-designed series of experiments will include a certain proportion of challengers aimed at producing controlled variation. Banks that test only decision strategies that are close to their BAU limit the amount of learning they can gain from live production data. By pushing the design of some challengers outside of these bounds, they introduce variation into the production data, thereby expanding what they can learn from it. Learning from live markets Here is an example of an analytic learning loop in action: Product: Introduction of a new card product for high-volume transactors. Objectives: Increase interchange revenue. Increase fee revenue. Decrease attrition in targeted populations. Strategy: Segmentation: 10 consumer profiles of transactors based on a wide range of predictive behavioral characteristics (card usage and balance growth patterns, etc.). Overlay of economic impact analysis and geographical analysis enabled further refinement (e.g., changes in price sensitivity in deteriorating economic conditions). Treatments: Targeted packages of offers, messaging and packaging tailored to the target segments and incentives (rewards program), limits and penalties aimed at encouraging high usage and low balances. Simulation: Using action-effect modeling and simulation, the bank projects the behavior (response, acceptance, activation, transaction volume and frequency, balances and delinquency) of consumers in each profile and the impact of that behavior on P&L metrics (revenue, cost, profit). Test and learn: The strategy is deployed into production as a challenger for champion/challenger testing, and prescreened offers are mailed out. Now the bank waits to see if the actual behavior of these segments aligns with the simulated behavior. Data from originations systems, as well as outcome data from other operations systems, are automatically fed back to an analytic learning hub used by both acquisitions and originations. Performance metrics tracking activation, usage and payments are analyzed at intervals following time of deployment (T + x days). At T+15, response and activation behavior patterns are emerging, and at T+30, the bank can do some preliminary assessment of early-life behavior. At T+60, delinquency can begin to be discerned. The bank doesn t have to wait for all of the data to come in to determine which strategies are working and which aren t. As shown in the graphics below, by comparing early actual results against simulated results, analysts can extrapolate longer-term outcomes. In this way, the bank can take quicker action, accelerating test-andlearn cycles. continued on next page RESPONSE T+15 Profile 3 is performing as expected; continuing on-track performance can be extrapolated TRANSACTIONS T+30 Profile 3 is performing differently than expected; change strategy sooner rather than later DAYS DAYS www.fico.com page 4
»» continued from previous page Profiles 2, 5 and 7 are performing exactly as simulated. Acquisitions steps on the gas pedal for those profiles: Additional mailings go out to populations fitting those profiles. Additional incentives (special bonus point gathering offers) are applied to booked customers from those profiles to further stimulate behavior profitable for the bank. Profiles, 1, 3, 6 and 9 accepted the card quickly and activated, but have low usage. It can be inferred that these customers are using the card as a backup. Acquisitions modifies the strategy to apply a different treatment to those profiles: Additional incentives (extra rewards based on transaction volume) are offered to stimulate usage. Messaging on the mailings is reworked to emphasize the advantages of using this card over those of competitors. Profile 10 responded at very low rates to the card offer. The bank modifies the strategy to see if it can boost activation rates: Activation incentives (extra rewards for all transactions within 30 days of card approval) are added to the offer. Messaging on the mailings is reworked to create a greater sense of urgency. Call center and branch staff receive supplemental training on how to facilitate and encourage activation. Profiles 4 and 8 had almost no response. Acquisitions immediately stops mailing to this profile and uses the learning to develop a product that better meets the needs of these consumers. The learning goes on, with strategy modifications deployed as one or more new challengers prove their worth against the current champion. At T+90, performance metrics are showing that profiles 2 and 7 have significantly lower bad rates than had been simulated. The economic impact analysis had predicted that transactors would be less likely to change their behavior in economic downturns, so the bank feels confident that it can adjust its decision criteria for these profiles without increasing its risk exposure. Acquisitions and originations use this information in a coordinated way to make these changes: Offer more competitive pricing. Increase initial line assignments. Ease prescreening and underwriting score cutoffs. This treatment is not applied to all consumers in these profiles, however. The economic impact analytics also predicted that populations in certain geographical markets would be more likely to have higher bad rates during an economic downturn. A new split is added to the decision strategy so that consumers in these markets continue to receive the more cautious treatment. www.fico.com page 5
»» Fundamentals of Analytic Learning Loops Figure 2: Traditional approach to acquisitions and originations The fundamental enabler of an analytic learning loop is an analytic learning hub through which from data analysis and learning from production testing flow. This hub will generally be composed of analytic data marts (continually refreshed from internal and external data sources), a variety of analytics and a repository for strategies. Decision to Acquire New Customers Prospect Targeting Existing Customers Bank Footprint Prospect Database Decision Strategies for Acquisitions 1. Direct Mail 2. Prescreen of One 3. Test and Learn Credit Offers Taken to Market Acquisition Strategy Tracking As shown in Figure 3, the learning loop is then threaded through both acquisitions and originations processes. This threading occurs through shared access to the hub as well as automated data feeds from operational systems, bringing decisioning results and account performance outcomes back to the hub. Decision to Accept/Reject Credit Applications Account Origination Direct Mail Prescreen of One Decision Strategies for Booked Accounts Test and Learn Application 1. Accept/Reject 2. Initial Credit Line 3. Test and Learn Strategy Tracking Figure 3: Analytic learning loop threaded through both processes Existing Customers Bank Footprint Marketing Analytics Analytic Learning Hub Decision Strategies for Acquisitions and Analytic Data Mart Credit Offers Taken to Market Acquisition Performance Tracking Direct Mail 1. Risk Score 2. Economic Impact Model 3. Segmentation Model 4. Action-Effect Model 1. Direct Mail 2. Prescreen of One 3. Test and Learn 4. Accept/Reject 5. Initial Credit Line 6. Test and Learn 1. Tracking 2. Simulation 3. Learning Prescreen of One Booked Accounts Test and Learn Management System Performance Tracking www.fico.com page 6
»» Many Ways to Implement and Expand Learning Frameworks Banks of all sizes, no matter what their current operational systems and level of analytic sophistication, can accelerate profitable growth by implementing a learning loop. The open, hub-based approach is a far less costly and disruptive way to improve visibility and coordination between acquisitions and originations than traditional one-to-one systems integration. Still, some banks face challenges in moving toward more rapid analytic learning. These may include: Mix of vendor-supplied, homegrown and acquired systems that makes sharing and reusing data and decision elements difficult. Narrow range of models providing limited into customer behavior. Aging models built from data that is out of alignment with how customers are behaving today and may behave in the future. IT resources stretched too thin to take on a major initiative. Poorly designed experiments (champion/challenger tests) that make it difficult to understand what worked and what didn t. FICO knows how to solve these problems. In fact, we recently deployed an analytic learning loop for one large regional bank in less than nine months. The engagement includes hosting and managing the company s analytic learning hub until IT is ready to take it in-house. Figure 4: Analytic learning across the customer lifecycle Customer Lifecycle Decisioning Acquisitions Adjust Strategies (and Models When Needed) Evaluate and Learn from Results Prepare Data Customer Management Develop Models Develop Strategies Collections Test/Validate Models and Strategies We are also working with this client to expand analytic learning into additional lifecycle decision areas. One of the great benefits of learning loops is that banks can put them in the area of highest priority which for most companies today is front-end growth then expand them by hooking other decision areas into the hub. They can also extend learning across lines of business, looping in credit cards, debit cards and other consumer lending products. They can scale analytic learning loops to serve acquired portfolios and new geographic or demographic markets a practical way to promote best practices, and achieve decisioning consistency and cost efficiencies across the enterprise. Run Champion/ Challenger Tests Deploy Models and Strategies Individual analytic learning loops drive higher performance locally, while also contributing to learning loops in other areas. Together, all these local loops contribute to overall portfolio performance, with everyone working toward the same P&L goals. www.fico.com page 7
»» The fastest path to rapid analytic learning While FICO can help banks implement analytic learning loops with their existing systems, we also offer a complete solution for driving higher bankcard growth. The FICO Bankcard Growth Solution is an integrated approach that works across the acquisitions and originations processes. It brings together the data, analytics and decision software applications banks need for rapid, successful learning in dynamic markets. Solution components: Analytic learning hub data mart, fed and refreshed from a variety of internal and external data sources. Advanced predictive analytics, such as the FICO 8 Score, FICO Economic Impact Service, and FICO segmentation and action-effect models. Model building, strategy simulation and optimization tools. Decision applications and services, such as FICO Precision Marketing Manager and FICO PreScore Service for acquisitions prospecting and offer generation, and FICO Origination Manager for underwriting and new account booking. FICO Precision Marketing Manager Prospect Database FICO PreScore Service New Accounts Origination Performance Tracking FICO Bankcard Growth Solution Data Mart Tracking Simulation Learning Alerts ACQUISITION Analytic Learning Hub Analytics Scoring FICO Economic Impact Service FICO Segmentation Models Action-Effect Models ORIGINATION Decisions Product/Offer Channel Type Prescreen of One Accept/Reject Initial Credit Line Test and Learn Offers FICO Precision Marketing Manager Outbound Marketing Database Performance Tracking FICO Origination Manager FICO Network Hosting Services Strategy Consulting FICO analytic and decisioning professional services consulting. Find out more about the FICO Bankcard Growth Solution. www.fico.com page 8
»Conclusion» For bankcard issuers in many markets today, business conditions are profoundly altered from the more stable conditions of the recent past. The future, in all markets, is profoundly uncertain. Things can change quickly that is the essential lesson of the financial crisis, whether endured or observed. And it means that banks struggling to re-establish their footing on stable ground may never find any. What banks can find, however, is a reliable way of learning about customer behavior at the speed of ever-changing markets. An analytic learning loop is something a bank can rely on. Find out more about analytics and learning for the banking industry: Check out FICO s Bankcard Growth page and solution brochure. Read Insights white paper #43: How Analytics Can Help Banks Navigate Financial Reform. The Insights white paper series provides briefings on research findings and product development directions from FICO. To subscribe, go to www.fico.com/. For more information US toll-free International email web +1 888 342 6336 +44 (0) 207 940 8718 info@fico.com www.fico.com FICO, PreScore and Make every decision count are trademarks or registered trademarks of Fair Isaac Corporation in the United States and in other countries. Other product and company names herein may be trademarks of their respective owners. 2011 Fair Isaac Corporation. All rights reserved. 2777WP 06/11 PDF