How Financial Institutions Detect and Prevent Fraud with Machine Learning
Machine learning is a topic that I have been spending a lot of time with over the past couple of years. There is a lot of momentum behind it in the industry. As is the case with so many 'buzzy' topics, you have a lot of different definitions of what is machine learning. The way that we at Aite Group define machine learning is that it’s about baselining good user behaviors, and then using iterative optimizing analytics to detect anomalous activity.
Machine leaning is great from a fraud use case perspective as it helps financial institutions keep pace with the evolution of the fraud and the changing patterns of the fraud. This is as opposed to the legacy approach, which was rules-based and looked for bad behavior. When you have that approach, you tend to get a lot of false positives and you also miss a lot of things because you're not looking for the new and emergent patterns of fraud.
Operationalizing machine learning is definitely one of the more challenging aspects for the banks that I've talked to. It's about harnessing the data in the bank. That is not an easy task when you have a number of business units that have grown up in silos and have data in disparate places. If you have banks that have grown by acquisition, then the silos may have entirely different data structures and different data definitions.
As you look at the journey toward machine learning, planning for the data harnessing and the data wrangling is a really important piece of the process. Large institutions, such as top 100 global banks, have a combination of internal bespoke data scientists that are dedicated to operationalizing machine learning as well as partnerships with vendor partners. Large regional banks are increasingly looking toward working with vendors that can help them on their machine learning journey. Finally, the smaller financial institutions, community banks and credit unions are dependent on their core partners to help them operationalize machine learning.