Sounding the Alarm: Using Predictive Analytics to Make Commercial Alerts More Effective

Recognizing that information has value is the stuff Nobel Prizes are made of. In fact, in 2001, George Akerlof, Michael Spence and Joseph Stiglitz earned the Nobel Memorial Prize in Economic Sciences for their work in the area of "information asymmetry," which is the study of decisions in transactions where one party has more or better information than the other.

The theories of information asymmetry are relevant to a range of industries, from insurance to credit management. In fact, they form the basis of many modern lending and pricing structures. For instance, by allowing clients to select the level of protection that best fits their personal situation, the insurance industry can customize rates and coverage rather than using a one-size-fitsall, single-priced system. Similarly, lenders can tailor their offerings and make credit available to a wider audience (at appropriate rates based on risk level) rather than simply eliminating higher-risk prospects altogether or increasing prices across all products.

For credit managers, information economics ring especially true as one party, the borrower, is likely to have a better understanding of his or her ability to repay obligations than a lender. This requires companies extending credit to constantly monitor each borrower's financial picture and adjust terms accordingly. Because information relevant to creditworthiness is constantly in flux (as previous obligations are met and new debts incurred), what constitutes "more or better information" is always in question. Therefore, "time-to-information" impacts credit data's value; the sooner it can be obtained, the more valuable it becomes.

Advances in networked communication have made realtime data access a reality. Pertinent information about a topic as far-reaching as a natural disaster is now available and accessible as events unfold. This capability has led to the development of automated systems that trigger alerts based on relevant events and send prearranged notices in response. This offers a wealth of innovations in other areas as well, such as supply chain management, investment banking and portfolio risk management:

* If barometric pressure drops below 27, sound siren.

* If 200 pounds of nails are sold in Wichita, order more from supplier.

* If price/earnings ratio drops below 15, execute "buy" order.

* If client exceeds credit limit, hold shipment.

Each of these examples demonstrates how various events trigger corresponding alerts. For lenders this is nothing new; automated alerts have become standard procedure when monitoring borrower risk. Instant access to credit data translates into greater risk detection and improved portfolio performance.

Commercial Triggers Meet Predictive Analytics

Whether used to predict storms or ongoing risk, accurate forecasting is based on sound mathematic principles. Edward Thorp demonstrated this famously 50 years ago with the publication of Beat the Dealer, which proved that the game of black jack was beatable using predictive scoring (and wager adjustments) despite an environment of constantly changing variables. For lenders, alerts triggered by changing data are enormously valuable.

While there has long been interest in predictive analytics, until recently there have been a limited number of studies on the efficacy of event triggers (and their corresponding alerts) as applied to business monitoring. However, a recent study by Experian carefully reviewed the performance of nearly 20 million commercial events triggered over a three-month period. To ensure objectivity, the research eliminated those triggers generated by businesses with a history of delinquent payment habits, leaving 16 million "clean" triggers for analysis. This study offers insights into the power of predictive analytics, particularly as it applies to business monitoring and portfolio risk strategy.

Lenders now have the ability to closely select and adjust which triggers are received. This has the dual benefit of distinguishing between irrelevant triggers and those of great significance. Further, applying trigger analytics ultimately enhances account management strategies by creating more streamlined, effective ways of identifying future bad behavior.

It's inefficient to simply have every event trigger an alert. The same event can have widely varying levels of importance depending on what is triggered. This is one of the undervalued concepts with alerts. Setting inappropriate thresholds causes a high frequency of triggers, so much that they lose their effectiveness. In a world of information overload, too many notifications are worse than no notice at all. It's therefore imperative that triggers are filtered and validated to better reflect which triggers are indicators of risk and warrant an alert. This is especially true of alerts triggered by a change in credit score, which produces the highest volume of notifications.

The research found that score change triggers had much greater relevance at specific starting points. For example, when a business's score of 70 (considered low to medium risk) dropped by 15 points, there was only a 1.15% "bad rate" over the subsequent three months. However, for a business with a credit score of 40 (medium risk) that dropped by the same 15 points, the bad rate increased sharply to 6.51%. Therefore, an alert generated by a business dropping 15 points provided data for making wise credit decisions.

This insight demonstrates the importance of applying robust filters in any monitoring program and setting alert score thresholds according to multiple factors. This can reduce the number of alerts while also focusing attention on those alerts that warrant the greatest response.

Not All Alerts Are Created Equal

Getting alerted to a change in a borrower's payment status offers the first indication of potential issues with repayment ability. Triggers can't tell a portfolio risk manager when to act or what action to take. Invariably, it is left to individual lenders to develop sound risk strategies to determine what action to take, if any, in response to an alert.

When lenders treat all trigger alerts the same, with the belief that all events carry equal importance, they'll get notified on everything. Setting triggers to automatically filter out insignificant triggers will enable lenders to have alerts that more accurately identify future bad performance.

The research found that of the many possible events that might trigger an alert, the triggers that included a derogatory comment such as "not paying as agreed" were the most predictive of future bad payment performance. Conversely, an alert generated by a fluctuation of available credit might offer little insight. The problem for lenders is that alerts all sound the same alarm.

Identifying the underlying cause of alerts offers a more accurate indication of future events. By fully analyzing which specific events trigger alerts, and what those events ultimately signify for each borrower, the alerts themselves become much more revealing. For example, the study also found that a "derogatory trade payment trigger" (a business reporting payment of 61 to 90 days past due) indicated a 15.37% likelihood of a business loan going bad within the next three months. Notably, a "lien trigger," which often carries a harsher response, indicated a relatively lower 7.22% bad rate. These examples demonstrate the significance of studying not only the frequency of triggered events, but also their specific origins.

Of the 47 triggers that were based on bad rate and evaluated in the study, eight of the top ten were based on trade data contributed by multiple lenders as part of their monthly reporting. Again, this insight proved to be the most predictive of future behavior and highlights the importance of sharing payment behavior. "When compared across multiple client portfolios, the late payment triggers ranked the highest in predicting risk," the study concludes. "It is critical for credit grantors to share the payment behavior of their customers with others through trade contribution."

From Account Monitoring to Actionable Intelligence

Smoke detectors don't work without batteries, and fire alarms become ineffective if they go off so often that they're simply ignored. Similarly, alerts are only worthwhile if they're working properly and acted upon when they sound. The best alarms are those that not only sound, but also indicate specifically why they sounded and what should be done in response.

One of the benefits of a comprehensive trigger analysis is help in answering the question, "An alarm has sounded; what action do we take?" By understanding the relative impact and what each trigger event means, lenders are better able to build out the correct workflow to take the appropriate next steps.

While some larger lenders have already begun to assess trigger events and develop this sort of decisioning workflow internally, most businesses don't have the analytical resources to do so on their own. Now, with predictive trigger intelligence readily available, any lender can utilize this information to modify and develop more intuitive account management workflows.

Identifying which credit changes are important enough to merit notification may not win a credit manager a Nobel Prize. However, being able to accurately forecast which events should trigger an alarm (and knowing what to do when that alarm sounds) may result in something almost as goodlong-term job security. 1

Because information relevant to creditworthiness is constantly in flux, what constitutes "more or better information" is always in question.

The best alarms are those that not only sound, but also indicate specifically why they sounded and what should be done in response.

Ann Skibicki is a senior product manager at Experian. For more information, please visit, www.experian.com/b2b.