The last decade has seen a remarkable rise in applied artificial intelligence (AI), enabling computer systems to mimic human-like performance within a defined context. From smart home devices and voice assistants to image recognition and recommendation systems, there are examples all around us. The use of applied AI and in particular the subfield of machine learning, has the potential to add “intelligence” to automation and further streamline FX operations.
FX operations are more complex than ever. This complexity can create costly manual touchpoints, such as resolving failures in trade enrichment, reference data validation, confirmation matching, and payments. These manual touch points are a barrier to STP, and account for a significant portion of total operational costs. This is compounded by increasing FX volumes, which increase operational costs as financial institutions are forced to scale up human resources to manage the additional manual operations.
Conventional forms of digitalization can be applied to improve STP rates. For example, through rules engines and workflow automation for settlement and confirmation. Vendor consolidation is another way to improve STP rates by improving data consolidation and reducing integration complexity. While the industry has made significant strides in these areas, there is still room for improvement, and this is where AI can supplement the conventional methods.
AI is an umbrella term, encompassing machine learning, deep learning, natural language processing, computer vision, and many more technologies. Although AI is not new, financial institutions have only recently started to recognize its value in revenue growth, cost reduction, and operational efficiency. Interestingly, it’s not regulations that are slowing adoption: a study by the Bank of England suggests that financial firms consider internal technology constraints (such as legacy systems and data limitations) to be a bigger barrier to the deployment of machine learning. The great news is that recent advances in cloud computing and data processing capabilities have made AI much more accessible.
Today, STP rates are relatively high for interbank trades but client trades still have some way to go. Obstacles include payment data discrepancies (particularly around third-party payments), counterparty confirmation issues, user authorization, and payments fraud. AI is being explored as a solution to increase STP rates by optimizing reference data and payment instructions validation, exception management, confirmation matching, and payment fraud detection, among other things. In a study by Accenture, securities settlement failures could be predicted with an accuracy of 83% to 97%, depending on the data samples and machine learning models used.
Machine learning can be used to build self-improving models using large quantities of historical training data to identify complex patterns, without programming explicit rules or scenarios. Optimizing STP rates through AI requires multifaceted machine learning models. Key data inputs include transactions, counterparties, currencies, settlement instructions, as well as cases of exceptions, matching failures, and payment fraud. The “trained” models can then be used on new data to identify trends, detect anomalies, or predict outcomes. The huge advantage of an AI-based approach is that the models can use newly generated data to constantly “learn” and improve themselves.
With the advent of the digital and data economy, applied AI has come of age and there are very real practical benefits that can unlock business value to lower operational costs and identify avenues for revenue growth.
To learn how our solutions can optimize your STP rates, contact us today.