President at Sunwest Bank.
Technology has dramatically changed how the financial services industry operates. This has been consistent over many decades; however, recently the pace of change has become exceptionally fast. The fintech market has deployed these technologies to disrupt the broader industry by enhancing the customer experience and changing the traditional customer acquisition model.
The next evolution of fintech will focus on the back end and middleware software that powers the financial services industries. What few market participants realize is the shiny front-end customer experience of these new fintechs and neobanks are still often powered by traditional banking systems. There has been little innovation in the back end of banking over the past decade, and it is hampering the ability for fintechs to grow. However, that is changing.
Machine learning (ML) in the near term and artificial intelligence (AI) in the medium term will be transformative for the fintech and financial services industries. These industries, specifically banking, possess vast amounts of customer, business, transaction and financial information. These technologies can help financial services companies in two primary ways: more effective customer management/retention and more efficient back-office operations.
Benefits Of ML/AI
Financial services companies can leverage ML/AI to understand their customer and business lines more effectively. For example, many firms are using machine learning to power financial product recommendation engines or customer engagement prompts to relationship management teams. It combines personal data, including how someone uses credit, their scoring and balances and then suggests suitable products that will fit the individuals’ needs.
MORE FOR YOU
Banks are beginning to use ML/AI to create predictive analytics surrounding customer behavior, buying preference, and outlier fraud detection for card and transaction management. Improved fraud detection provides opportunity for financial services companies offering credit cards and virtual payment options to use AI-powered algorithms to spot stolen card activity. It can combine device data, IP addresses, physical location, and behavior patterns and compare all that information against a baseline of a “regular” single-card user. These are technologies that historically only the large card issuers had access to; however, ML/AI companies are enabling tech-forward banks and fintechs to compete with best-of-breed technology.
AI enables customer service at scale through automated systems, with the system learning over time. By reducing interruptions to the customer’s journey through automated intelligent service, brands can lower their acquisition costs and other customer-centric metrics. For help desk inquiries, AI-powered tools can intelligently route requests based on content. They can skip a ticket “gatekeeper” and instead move requests to the right department or use automation to provide immediate answers.
Clean Data Is The Foundation For ML/AI
It is easy to talk about these technologies but can be hard to implement them. Successful and effective ML/AI requires structured data housed in a database before any analytics can begin. Think about it like learning the alphabet before beginning to read.
Fintechs are talking about ML/AI, but the financial services industry must solve the data problem before it can even begin discussing ML/AI. Those early adopters that solve the data problem and then learn from that underlying data will be light years ahead. However, there are multiple steps to successful adoption, and the most important is accurate and clean data. This problem has plagued the financial services industry for decades. There is an old saying as it relates to data, “garbage in, garbage out.” The results of ML/AI are no different. If you train your models on bad data, your artificial intelligence is going to create bad or unintended results. It is like studying for a test with the wrong notes.
Therefore, the first step in any ML/AI strategy is cleansing and normalizing all of your data sets to ensure your models will be fundamentally sound. One of the most important data normalization decisions a financial services company will make is how to define an aggregate relationship or household. The second step is ensuring your database has all of the relevant data to properly train your models and identify the trends, outcomes or results you are looking for. This requires your team to fully understand the respective outcome or job that the AI model is producing and the data it requires. Today, there are incredibly robust data sets across multiple industries, economic variables and human behavior, but they require subject matter experts (SMEs) to identify the proper data sets for the ML/AI processes you are running. The earlier you involve SMEs, the more time and money you will save.
Once you have identified and normalized your data sets, maintaining data integrity is key. This means you need defined data policies and procedures to ensure that both people and systems inside of your organization operate with the same data rules. In addition to this, you will want systematic quality control (QC) processes and some level of human oversight. Another great use case of AI is the systemization of QCing data and documents; machines do this far more quickly and effectively than humans.
In closing, the successful deployment of ML/AI tools can be a true differentiator for financial services companies. This technology allows our industry to both improve and automate back-office processes and engineer more targeted engagement with our customers. In short, it can lower operating costs, enhance cross sell and increase customer acquisition. However, a successful ML/AI strategy requires a thoughtful strategy and clean data. AI is the next phase of technology competition in the financial services industry, and it is paramount that traditional financial services companies, fintechs and the technology companies that support them are focused on developing an AI strategy to compete in the evolving landscape.