Reducing Fraud and Risk With Advanced Data Analytics
Tony Zarrella
11/01/2005
Bad debt, fraud losses and revenue leakage are serious problems for the telecom industry. Today, service providers are using sophisticated analytic technologies to stop these losses, which have risen dramatically in recent years. Without advanced analytics, they are likely to get worse.
Consolidation has left most providers with multiple, incompatible systems and inconsistently applied policies—“holes” where fraud and bad debt can penetrate. Competition, particularly in the wireless environment, has forced providers to make trade-offs between accelerated customer acquisitions and exposure to risk and fraud. In addition, the use of micropayments to purchase ring tones, music, video clips or even small retail purchases could be a significant new source of revenue—but also a significant source of risk.
However, with the expanded use of analytics, through more precise insight into customer behavior, providers can identify acceptable risk prospects in populations they would generally avoid and develop accurate, risk-based pricing to protect profit margins. This new information could also allow more effective up-selling within the customer base, increasing usage without increasing risk.
It’s All About the Data
Today’s customer acquisition systems can access a wide range of data sources, depending on a provider’s business model and processes. During the acquisition process, the system should be able to make real-time queries not only to the provider’s internal data sources (pre-approval lists, customer master files, in-house negative files and the like), but to external sources as well (consumer and commercial credit bureaus, industry fraud/negative lists, opt-in industry exchanges, “do not call” lists , regulatory agencies).
When information from traditional credit sources is scant or nonexistent, the customer acquisition solution should be able to access debit bureaus as well. This information shows how well applicants have handled their bank accounts—another important piece of consumer payment histories.
Understanding the Data
Once the applicant has been approved, risks for non-payment and fraud are equally as serious. Advanced analytic solutions capture and analyze data from every transaction (calls, payments, customer service events, etc.). For example, consider two subscribers who are late paying their bills. Their monthly activity, depicted in the chart below, shows a similar volume of calls—but their risk for early-life bad debt may be altogether different.
For subscriber 1, detailed transaction information shows high-value call volume spiking shortly after activation and remaining high. This type of usage can be combined with other calling data and application information to create a pattern indicative of fraud. A good risk management system will provide an early alert, enabling the provider to stop the fraud and halt losses.
For subscriber 2, an initially high volume of costly calls drops off, and perhaps transactional detail also shows that the subscriber made a billing inquiry to customer service. The overall pattern is indicative of a customer who may be experiencing “sticker shock” from the bill. Given an early alert, the provider can act promptly to reduce the risk of nonpayment by contacting the customer to negotiate a payment schedule or assist in the choice of a more affordable plan.
Using Neural Networks
Advanced analytics such as neural networks—a type of predictive model that is good at recognizing subtle, hidden and emerging patterns within complex data—are becoming more popular. To tease out these patterns, neural networks examine hundreds of variables simultaneously and look at relationships between data.
While neural networks are usually employed along with rules, they are far more flexible and precise than rules alone. Instead of applying the same rule to everybody, neural networks look at the complete picture of the individual subscriber created by all variables and data interrelationships, evaluate risk based on that unique picture and assign a single risk score that takes everything into account.
In addition, neural networks can detect subtle changes in subscriber behavior, which might indicate a shift in the level of risk. For example, a customer in good standing who begins making less costly calls and avoiding roaming charges may be experiencing financial difficulties. This information can be taken into account immediately. If the subscriber is also carrying a balance and becomes one cycle delinquent, a higher risk score will be generated. Based on the score, rules could automatically assign a contact treatment and account spending limit modification in line with the higher risk pattern.
Neural networks can be used in conjunction with data from multiple providers. Consortium data increases analytic precision by providing models with a more comprehensive set of data to analyze. A broad range of data is especially important in telecom, where fraud schemes tend to migrate from provider to provider, region to region and country to country as they are identified and shut down. Systems that aggregate and analyze consortium data use data from all providers in the consortium to protect each member.
Neural networks become more accurate over time by continually improving the quality of the data they are analyzing, which occurs through a process called dynamic profiling.
The term “profile” in this case refers to mathematical equations that condense a huge amount of data—perhaps a trillion bytes or more—into about a thousand very potent and meaningful numbers. The data can be analyzed against new data from a current application or call, in a real-time environment.
With each call made, each bill paid, each inquiry via phone or Internet, the analytics feed new data and analytic results back into the profiles, making them more accurate. The next time the neural network uses the profile, it is working with a more complete picture and is therefore able to make even better predictions.
The Result: More Profitable Decisions
Fraud, risk and revenue assurance systems generally deliver analytic results in the form of scores indicating risk level. These scores can be used by customer acquisition systems, which may also incorporate their own analytics, to automatically gain customer insight and assign decisions, treatments and offers. Customer acquisition systems should be able to handle up to 90 percent of new account decisions without human intervention.
Account management systems use analytics to automate collections, retention, marketing and other interactions with customers. Using a variety of standard or custom models, they make ongoing, statistically derived assessments of risk and revenue on a customer and account level. These assessments allow providers to precisely segment their customer base and assign treatments to expand usage without increasing risk.
These systems provide “what if” analyses and testing to compare an existing strategy against a proposed improved one. Tests take place within the real market environment, but on a limited scale, running the strategies side by side against randomly sampled segments that represent a small percentage of the portfolio. Using this approach, providers can measure results before rolling policy changes out to larger populations.
In addition, powerful statistical comparisons and visualization aids enable providers to rapidly test far more alternatives than would otherwise be possible, refine the best performers and compare results in out-of-market tests. This involves the use of comprehensive models to look at all of the factors that go into a particular provider’s decision strategy—objectives, resource constraints, possible actions and customer reactions, even market uncertainties. Such models produce strategic recommendations, which the provider’s subject matter experts can subsequently explore and refine.
Benefits of Analytics Decisioning
Applying data and analytics in real time to telecom decisioning substantially reduces losses. For example, a major global wireless and wireline service provider was experiencing a sharp increase in early-life bad debt for residential long distance accounts. The provider implemented an automated system that predicts payment risks throughout the life of both new and existing accounts. The system lets the provider use flexible rules to combine a predictive score with account-level characteristics in a process that automatically assigns and performs appropriate treatments. The results were a 95 percent reduction in average debt per fraud account and a drop in average early-life bad debt accrued before detection from $125 to $7.
Maturing, changing markets mean telecom providers must look harder for customers. More precise risk analysis, coupled with new data sources such as debit bureaus, let them safely increase market penetration, selecting subprime and thin-credit prospects who represent relatively low risk. Better customer insight also enables wireless providers to more accurately assign deposit requirements, maximizing adds while protecting profit margins.
Analytics-based risk management helps telecom service providers achieve higher recovery rates at lower cost. One reason is that a system incorporating neural networks refers far fewer accounts to investigators. And when it does, all information about the risky subscriber is aggregated and thereby easier to access. Referred accounts can be prioritized by score, with the riskiest cases routed to the most skillful collectors. Reason codes accompanying scores can be used for more detailed triage as well as skills-based queuing.
Consider the example of a business person suddenly making calls to Taiwan. The telecom service provider might have had a rule that said all sudden changes in international calling should be referred to an analyst—but in this case, the referral would have been accompanied by a score indicating low risk, typically the most frequent designation. As a result, the analyst would be able to place an appropriate “customer service” type of call to the subscriber to confirm everything was all right or, more commonly, defer any contact until further transaction data clarified the risk outlook. If contacted, the legitimate subscriber, instead of being annoyed, would understand that the service provider was actually looking out for their best interests.
On the other hand, had the referral been accompanied by a high risk score, a less frequent but higher dollar risk scenario, the call to the customer would have been of an investigative nature.
The analyst’s inquiry would have been focused, by the reason codes, on specific problems.
Helping to keep good customers satisfied is one way advanced risk management can contribute to improved profitability. In addition, some providers are finding that improving collections efficiency enables their staff to spend more time providing superior service and cross-selling or up-selling. Transactional data can even be used to identify lifestyle changes and events, which might provide opportunities for offering a more profitable service mix.
The Move Toward Unified Risk Management
Today analytics-based decisioning solutions can reduce risk across the entire telecom customer lifecycle. These systems can be used separately or together to achieve a consistent level of risk management enterprise-wide.
Consensus is developing in the telecom industry on the need to move toward more unified fraud, risk and revenue assurance management. It’s a daunting challenge, however, particularly given that rapid acquisitions and mergers have left most providers with more of a hodgepodge of disparate systems and methods than they had a decade ago.
However, by providing a centralized, data-driven, objective mechanism for assessing risk—a mechanism separate from but available to all applications—these solutions will improve decisioning across telecom enterprises, inevitably changing the economics of service delivery.
Tony Zarrella, vice president of telecom risk analytics at Fair Isaac Corp., is responsible for leading the company’s overall planning, direction and integration of fraud management, risk management, network assurance and revenue assurance solutions for telecommunications.
Email this article
Add a comment
Printer version
Order reprints
RSS Feed
Bookmark article