As the payment systems industry moves into the 21st century, several emerging technologies will be increasingly and profitably employed by survivors in the business.
Notable is artificial intelligence, which can be used to assist risk managers in their forecasts and to help card marketers to be more accurate in their targeting.
One such development has extraordinary possibilities for credit card issuers facing stiff competition and pressure on earnings: what we call fractal base geometry used for predictive modeling and data mining.
As an analytical tool, fractal base management permits credit card issuers to represent terabyte-sized data bases into mirror files far smaller than their normal size, with no loss of data. This means that queries that would have taken hours or days to compute in a priorities committee/mainframe environment can be called at the personal computer level instantaneously.
Moreover, these data equations can be used to predict more precise outcomes from irregular or chaotic variables: Predictability out of chaos.
It was Benoit Mandlebrot, a German mathematician, who first began to study chaos theory in 1875, reaching the conclusion that phenomena in nature cannot be represented by straight lines, but by more or less jagged edges.
He identified repetition in that process, though. What appeared random at first were actually patterns that repeated themselves, a "self-similar" series of events that could be repeated under various scales of magnification. Out of that analysis emerged fractal geometry, and now, fractal base modeling and data mining.
In our business, the marketing and credit processing cost per acquired account can range from $20 to more than $150, depending upon many factors: competitive product design and pricing, lists used for solicitation, timing of solicitation, among others, which drive response and approval rates.
Appropriate modeling can drive down the ultimate cost per acquired account by better targeting the applicants solicited. Far beyond marketing modeling, there are multiple potential applications for fractal base data management.
Cross/Z Corp., the Uniondale, N.Y., company that pioneered this technology, has captured the attention of some major players, such as Chemical Bank, American Express, Discover, TRW, First Data, and other nonbank commercial companies. This is the fractal technology initially used to develop the cruise missile guidance system in 1987, prompting product names like Navigator.
What's so hot about this technology and the implications for the payment systems business? Our customary statistical methods use sample sets to construct linear models. Fractal geometry, on the other hand, allows nonlinear data and whole sets to be evaluated simultaneously. Relationships among variables become equations that are manageable at the personal computer station level, independent of any mainframe interface. It's fast and easy to use.
Benefits of fractal base data management include representing massive amounts of data, at an average ratio of 1,800 to one, although at the outer limits, as high as 10,000 to one may be possible.
Also, the equation speed is unparalleled in usual analytical methods, providing the ability of data representation without loss of data integrity.
Reasonable cost for representation software: $20,000 or less for a predictive model. In-depth data mining programs can run between $100,000 and $500,000, depending upon the user's unique needs.
The opportunities for use of artificial intelligence are enormous, as neural networks have already demonstrated in the area of credit card fraud control. Account scoring models can be optimized for behavior scoring, credit bureau scoring, new accounts, and authorizations. The fractal technology perpetually revalidates the data, avoiding the degradation that occurs from traditional card risk scoring methods.
Transaction analysis at the account level, such as spending patterns by SIC code, would permit issuers better to target new products and enhancements to their customers. Data base management can then be performed in-house, without using third parties for analysis.
Market segmentation can be extremely successful, since no samples are required to conduct the analyses. Even due diligence for buyers of card portfolios could be more precise with the forecast for income yield, operating expense, delinquency, chargeoffs and attrition, key components of pro forma earnings.
While the profit optimization for the use of fractal query management seems great to issuers hungry to preserve and enhance their earnings, a few cautions are in order. First, given the depth of cardholder information this technology allows, will regulators have concern with individual privacy matters?
There should be no more concern than they already have, given the state of technology and credit bureau scoring today; possibly less concern, since no personal cardholder information is distributed with fractal queries.
Next, will the technology be user friendly to card managers not trained in geometry or mathematics, or will issuers need in-house mathematicians to implement the programs? At this stage of development, Cross/Z helps develop the model, but the end users run the programs themselves, an advantage over other third-party firms that have their hands in similar statistical projects daily.
The process of developing and implementing fractals for card issuers involves four steps. First, historical information is transformed into a consolidated "fractal-ready" data file, containing all the variables for modeling. Second, a data-map or mirror file containing all of the multidimensional variables is created, using Cross/Z "Private Eye Launch" and "Fractal Engine" products. Third, "Multi-Model" is used to build predictive models. And, finally, "Fractal Navigator" is used to select the highest-performing model.
The entire process can run from three to four weeks. Data mining projects can also be completed as quickly, although the principal component of timing is how clean the issuer's data is.
To give you a perspective of the power of this technology, a typical fractal base data management user can cycle through 40 million records in about 15 minutes. Queries can be immediately performed, with greater precision, at lower cost, at a personal computer, with no sampling error. The application of fractals to data base design can be achieved on existing hardware and software, augmented by Cross/Z enhancements.
Card issuers trying to understand how the variables in their unique portfolios interact can benefit from fractals. Linear models today provide imperfect representations. Inexpensive computers at the single user level and making data available in real time to decision makers can help provide predictability in an area of apparent chaos, fast and relatively inexpensively.
Credit card risk analysts have always studied data through approximation, commonly fitting the data into linear equations to identify solutions and probabilities.
The sampling bias that occurs over time is inevitable. Fractals discover boundaries where irregularity is predictable, can be replicated, and therefore can be more meaningful to marketers and risk managers in the areas of predictive modeling and data mining.
We do not believe it will become necessary to house internal departments of mathematicians for card issuers to benefit from the technology.
You might assign a senior risk manager to study the technology, assess the costs and benefits, test a segmented part of your portfolio, then compare the results with a similar portfolio segment where fractals were not used. The results should provide the confidence level needed to roll out a larger project over the entire data base.
Marketing of payment-system products has progressed from mass merchandising to segmentation. To the extent that marketers can more precisely segment coexisting variables in the response process they will produce far more profitable card portfolios. Operating expense is optimized, and return on assets and equity is maximized.
In the past, card managers had to wait for days for mainframe computers to analyze characteristics of target audiences; today, those waits can be far shorter and less expensive. Fractal base data management takes card issuers on a voyage into deeper data base management, with far more variable characteristics, in a stand-alone PC environment.
For now, mathematicians at Cross/Z operations in the United States, Russia, and Poland are the major driving forces behind the commercial use of proprietary software for fractal base data management. There may be others in the future, such as scoring companies, that try to get into the act as well.
