Marrying up the law of entropy and the fact that the number of digital payments being made across the globe is increasing dramatically, by default means that fraud management processes need to be constantly refined.
Entropy as a scientific principle concerns the loss of energy from a system and describes how an ordered system moves towards disorder. The key point in understanding entropy is that it cannot be stopped, and to maintain a desired level of order, energy or performance, more of the same must be added into the system. A simple example is when you wear your coat on a cold day. When you take your coat off, entropy is the process that explains the loss of warmth, which can only be countered by putting it on again.
In the payment industry this is fully supported by the fact that fraud has reached the highest levels on record, affecting more organizations than ever. The scale of the problem was revealed in last year’s PWC Global Economic Crime and Fraud Survey. Nearly half (49%) of the 7,228 businesses across 123 territories that were interviewed reported that they had experienced fraud and economic crime over a two-year period.
Today, fraud management consists of several manual processes — models and rules performance monitoring, fraud pattern discovery and fraud alert management, to name a few. While these manual processes may be manageable at first, as the number of payment types and channels increase, it can rapidly become untenable to add more and more staff to manage and monitor all processes. Managing fraud can become very expensive, which is why efficient management processes are so important.
The fact that entropy exists, and remains a factor that cannot be stopped, means that all aspects of the business need to be monitored. You may be most interested in product development or working with clients, but if you do not watch the other parts of the business, such as accounts payable or accounting, entropy will eventually cause problems. Management regarding entropy aims at small corrections to keep projects or departments on track, rather than letting those areas run in isolation, until there is a much larger breakdown or problem.
There is a lot of information on how machine learning is helping to understand human behavior and more specifically, false/positive detection. However, there is little available research on how this relates to the end-to-end process within fraud management.
This draws you back to the fact that the industry is focused, and rightly so, on detecting fraud, but is not focused on evaluating the impact to the whole end-to-end process. Clearly the interdependencies on these two activity streams are significant, so the question is why both factors aren’t being considered by the fraud prevention suppliers.
While things naturally move to disorder over time, we can position ourselves to create stability. There are two types of stability: active and passive. Consider an airplane, which, if designed well, should be able to fly without intervention — this is passive stability. Conversely, a fighter jet requires active stability. With active stability, you are applying energy to a system, in order to bring about some advantage (keeping the plane from crashing). The plane can’t fly for more than a few seconds, without having to adjust its wings and these adjustments happen so quickly that it’s controlled by software in modern airplanes.
Autopilot ML, in this analogy, is the fighter pilot for fraud defense. Reacting quickly to fraud pattern changes by creating new machine learning models to stop the threats; while continually optimizing the fraud detection strategy, so that it is equipped to counter the newest and most damaging threats, while maintaining high acceptance and low insult rates.
In summary, machine learning is having a huge impact on the entropy of fraud detection, it is helping to maintain order, providing the system with passive stability. However, as stated above, without constant refinement and active stability, effectiveness is likely to decay. This rate of entropic decay needs to be measured, understood and more importantly learned from and acted upon. In terms of the latter, it is the efficiency of the changes that are critical, essentially providing stability optimization.
Are the fraud prevention products and services you are currently deploying maintaining stability?
