With the likes of Apple and Intel making some high profile acquisitions over the last few weeks, it’s pretty safe to say that machine learning is big news right now. But what does it mean for financial risk analysis and decision-making?
Risk analysis and decision-making is all about data. In fact, the smooth-running of business operations in general depends on the outcome of data assessments. By reaching insights from data faster, businesses can perform better.
Machine learning is also all about data. Instead of slavishly gathering and analyzing data to inform decisions, machine learning enables intelligent data; ‘learning’ occurs through the process and the model auto updates as a result of the learning. Pattern recognition enables automatic analytical model building. Insights from data can be arrived at quicker because the software program isn’t dependent on human intervention to update it or to tell it where to look.
It all sounds a bit ‘Skynet’ but actually it has very practical applications to help address today’s financial risk challenges. Take for example, fraud detection. Standard processes flag up potential fraud when a trigger is activated, such as a card being used in an unexpected foreign location, an unusual pattern of spend or suspiciously heavy use of a card in a single day.
Static triggers can be improved to dynamic triggers that are able to learn from a customer’s transactions, such as purchasing a plane ticket, prepaying for a hotel and buying a meal at an airport. In this example, when that customer then uses their credit card at a new location they won’t be bothered by a fraud check because the algorithm has learnt from the previous spending behaviour that led up to transactions in a foreign location.
With machine learning, a flexible environment is created that changes and adapts according not only to individual patterns of behaviour but also collective patterns. If an online site has become a ‘testbed’ of choice for identity thieves to process small-value transactions before using stolen financial information to make bigger purchases, transactions on those sites can be labelled and used for future model updates.
This can improve customer satisfaction by removing the inconvenience of a manual validation process to allow a genuine transaction, and – importantly – increase fraud detection rates, saving considerable cost.
In fact, analysis firm Oakhall has estimated that global financial services firms could save an eye watering $12 billion a year through machine learning fraud management.
To get full benefit from the predictive analytics power of machine learning, financial institutions need a fast, simple way to connect their machine learning application to their credit and lending decisioning processes. With this power harnessed, the potential for risk analysis could be huge.