When designing a strategy for detecting and preventing fraud, everyone always comes to the same conclusion—there is no silver bullet. There are simply too many variables, and too much change in technology, customer behavior and fraudsters’ tactics for any one solution to work effectively and sustainably for every organization, no matter how sophisticated.
Consequently, experienced fraud management executives are constantly experimenting and evaluating new data sources, scores, models, algorithms and technologies for that competitive edge. They observe customers’ behavior, survey their preferences and maintain a working knowledge of fraudsters’ evolving tactics.
The goal is the same for everyone—minimize fraud losses while effectively balancing customers’ experiences and operational expenses. But the exact recipe each organization lands on—the mix of processes, people and products—varies widely and changes constantly.
Fraud Models – Five keys to finding the right fraud score
Many different providers—whether associations, processors, switches or analytic firms—have begun to offer unique fraud scoring models, targeting different products, channels and customer segments. There are also many fraud platforms that allow organizations to build and deploy their own internal models.
So, which fraud scores will be most effective for your organization? There really isn’t a secret formula, but there are some basic principles, gleaned from years of experience working with industry leaders, to incorporate into your organization.
First, you should know that leveraging multiple fraud scores is a perfectly fine practice. Each vendor has different techniques and algorithms to produce their fraud score. Every technique has its own advantages and disadvantages. While it is important to understand the underlying technology driving the different fraud scores, what is more important is the performance and effectiveness of each fraud score, and whether it solves your business challenges.
Second, you should not underestimate the power of consortium data. The sources, quality and quantity of data is a critical component in developing robust models. Be mindful of startup vendors with a minimal client base touting consortium models. A good consortium should be representative of the industry it is representing.
Third, some fraud scores are now “mandatory.” What this means is that a provider (scheme/association or processor/switch) may be requiring the use of their fraud score, but don’t be afraid to question and quantify the effectiveness of the fraud score.
Fourth, measure the effectiveness of fraud scores. You can measure model performance effectiveness in a dozen different ways. What is important is that you are using the same approach and methodology across all fraud scores. Never apply performance metrics you have received from one vendor across all other vendors, as they are all likely using different ways to measure performance. A simple metric like value detection rate can be measured in several different ways. Find a common suite of performance metrics you can measure against all fraud scoring models.
Lastly, don’t forget about cost and benefit. Understanding the cost should always be part of your evaluation of fraud scores. The benefit is equally important, as it is in any performance comparison. Do your fraud scores overlap in some areas? Can one model be utilized for part of your portfolio and another model for the other portfolios?
As fraud continues to evolve, so should fraud technology and scoring models. There are now a number of different providers out in the market that provide effective scoring models. As such, you shouldn’t rely solely on one provider; rather, leverage as many as possible. And lastly, ensure a fair model comparison has been completed utilizing a common suite of key performance.
Drew Manuel is a senior director within the Fraud, Security and Compliance unit of FICO Advisors. He has over 24 years’ experience in the fraud industry and is regularly called upon to do fraud model/score reviews by clients around the world.
Enjoyed this blog? Why not read this one too.