Listen to the article
Your audio file will ready in a few seconds...
Just a few years ago, taking out cash before an event or dinner was not a strange sight. However, as digital transactions and e-payments become widespread, paying for your expenses with coins and bills aren’t as common anymore.
The world of electronic payment processing is growing fast and adapting itself to the shortcomings of everyday life. With growth, of course, come new opportunities, ideas and challenges. Electronic payment systems, and financial technology in general, have begun using artificial intelligence tools to provide practical solutions for everyday necessities.
Also, most retailers and financial institutions are now using artificial intelligence, with a specific emphasis in machine learning, to tackle difficult issues such as cybersecurity and document digitization. Such tools have also been used to speed up payment processing, especially as large companies begin to handle large volumes of data.
How is machine learning incorporated into payment processing?
A Brief Introduction to AI
When American computer scientists John McCarthy and Marvin Minsky, along with a group of accomplished academics, organized the world’s first conference on artificial intelligence in 1956, their idea was to explore different ways a machine could think like a human. Now, “thinking” can be a very broad term. Their main idea was to create technology capable of abstracting thought, solving problems and improving on its own.
According to McCarthy, “every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it”. This, however, is not that simple. Sixty years after the first meeting, the area of artificial intelligence has branched into a vast number of areas. Machine learning, natural language processing, speech, expert systems, robotics and vision are some of these ramifications.
Despite such growth, the ability of emulating the way humans think and learn has been largely left untouched (this is known as General AI). Yes, programs have managed to beat people in both Go (AlphaGo, 2016) and chess (Deep Blue, 1998). Yet those are two very specific tasks with a high level of complexity. If you asked AlphaGo to make your bed or play with your dog, the program would have no idea what you are asking.
Machine Learning and Deep Learning in Finance
The great majority of advancements in AI focus on specific tasks (also known as Narrow AI), such as playing chess or recognizing a human face. In the past two decades, however, a large number of industries have begun using a specific subset of artificial intelligence to provide better and more efficient services: machine learning.
Machine learning seems to be pretty self explanatory. Yet there are some caveats. As computer scientist Tom Mitchel puts it, machine learning is the ability a computer program has to learn from experience E with respect to some task P. Said in simpler terms, an algorithm can learn how to distinguish between two objects.
Currently, one of the most used techniques in finance is deep learning, which is a subset of machine learning. It has gathered attention in recent years for performing unsupervised learning, and having a strong capability of generalization and processing big data. These models use neural networks, which are described by cognitive scientist Melanie Mitchel as elements that simulate neurons in the human brain.
Different deep learning applications have permeated to fields such as medicine, neuroscience, physics, to name a few. Deep learning has developed a large number of different models, most of them based on neural networks, to tackle different issues within finance and banking.
Innovating the Industry
According to Jian Huang from the United International College in China, the use of machine learning algorithms (specifically deep learning) can be divided into two major areas: banking and credit risk and financial investment. In a study performed by Haung, he found that most deep learning models within finance and banking domains focused on stock market prediction.
So, how does the payments industry fit into this? Machine learning algorithms can detect highly probable transactions from being approved, while still reducing the amount of false positives and further reducing the costs. It can also help reduce complexity and make sense out of emerging fraud patterns and their correlations.
An artificial intelligence system, such as the one used by payment processors like VISA or Mastercard, have the ability of learning user behaviour and understanding patterns. Such a system uses a rule-based logic to derive insights of which variables may lead to fraud. When there is sign of irregular activity, the payment processing service contacts the bank to let them know what’s going on.
Historical data, that is, the information gathered from hundreds of thousands of transactions, can also help the algorithm make quick adjustments to its logic without any human interference.
In 2017, one of the largest banks in Singapore launched a virtual assistant that guided users through the digital experiences of finance. At that time, the use of chat bots in service platforms was not new, yet their application to payments was a game-changer. By providing a human-like experience for people who need support or details on their accounts, the system reduces human error and speeds up information.
Nowadays, chatbots and other forms of NLP or natural language processing (such as voice processing services) are a common practice. You can create an entire shopping experience within a voice channel, just as if you were speaking to a sales representative.
Finally, machine learning has been incredibly helpful in supporting data driven decisions, as quantitative methods are now used by key game placers. The more data you have, the greater opportunities machine learning models have of achieving great operational and strategic efficiencies. These can help individuals, small businesses and even large corporations.