By Don Keeley, SVP, Senior Director - First Citizens Bank, United States
In an industry where control, compliance, and risk management are the foundation of every process, the idea of relinquishing control to artificial intelligence (AI) feels, at best, provocative and at worst, reckless. Either way it can be terrifying to consider the consequences of just one high-profile failure. Yet as a retail banking executive overseeing technology strategy and operations in an era of rapid digital transformation, I’ve come to the realization that implementing AI is not about surrendering control; it is about establishing effective governance and maintaining control through dogged process, disciplined risk management, and well- designed reporting. Similar to managing your human employees, but at a much faster pace.
AI has the potential to transform everything from underwriting to account closure and just about everything in-between. AI has the potential to automate and speed up most customer needs from the beginning to the end of their account lifecycle. But unlike fintech startups, we don’t have the luxury of “move fast and break things.” As stewards of customer trust and institutional stability, we must implement AI in ways that are consistent, explainable, with tight guardrails to ensure we are meeting the expectations of regulators, shareholders, and most importantly our customers.
The Control Dilemma: Fear Meets Opportunity
Retail banking is highly regulated and inherently risk averse. Constant regulatory scrutiny, high- profile banking failures, and armies of lawyers always looking for the next class-action lawsuit have ingrained a culture where change is slow and calculated, auditability is expected, and consistency is king. The rise of AI challenges that culture.
AI systems, especially those based on machine learning, operate in ways that often resist traditional forms of human oversight. They learn from data rather than follow a rigid set of instructions. That learning can eventually yield astonishing accuracy, but it can also introduce bias and unpredictable outcomes at an astonishing rate.
So why take the risk?
Because the cost of inaction is now higher than the perceived risk of relinquishing some control to AI. Jump onboard or quickly be left behind. The top 3 banks (JP Morgan, Bank of America, Wells Fargo) have already made tremendous progress in AI automation and will continue to thanks to their eye-watering budgets ($2B+/yr). In addition, every vendor who provides banking technology offers AI in most if not all their new offerings. Of course, some are more advanced
and functional than others, with some being mere marketing. This customization of AI will ensure that banks of all sizes can have AI at their disposal, for a hefty price of course. The cost of inaction is quickly approaching and will soon exceed the cost of action. Those that fall too far behind will likely never catch up and become a casualty of the latest technological revolution. In addition to competition, customers will grow to expect immediate action, faster decisions, personalized interactions, and lower costs at their convenience, 24/7-365. AI adoption is not optional.
AI That Enhances Control, Not Replaces It
A new paradigm is taking hold. Instead of seeing AI as a loss of control and something to fear, we must view it as a tool that strengthens our control framework by reducing human error, highlighting anomalies, and enforcing consistency at scale. AI will not only improve the speed of decisions, transactions, and customer care, it will also improve the speed at which we can detect actions and decisions that are outside of our process and guardrails, followed by immediate action. In most cases, far faster than a human can.
For example:
As long as these systems are deployed with clearly defined boundaries, auditable logic, and rigorous human oversight via clear analytics and reporting, control won’t be lost, it will be improved.
Building the Right Foundation
Implementing AI responsibly requires more than just technology, it requires effective organizational change management, a robust governance framework, and close alignment with regulator expectations. As the old saying goes, 20% more planning upfront will save you from 80% more problems on the back-end. Thorough planning and deliberate execution are essential to a successful AI implementation.
Key considerations to factor in your planning must include:
1. Governance First
Before any model goes live, it must pass through a multi-layered governance process that has been reviewed and vetted by all key stakeholders associated with your AI implementation. Let others in your organization help you see what can go wrong before your customers or regulators do. This governance should include:
AI models should follow the same high standards as any financial reporting system; if a model can’t be explained it can’t be trusted.
2. Data Discipline
A common misconception of AI is you buy the technology, define your model, and turn it on. A critical component and necessary prerequisite of an effective model is high-quality data with clear data definitions. Your model is only as good as the data it learns from. This means ensuring your data is accurate, consistent, and well defined. If your organization has not already invested in data lineage tracking, cleansing, and secure storage, your plan must include adequate time for these activities. Your ability to effectively control your results starts with effectively managing your data.
3. Effective Human Oversight
AI works great where the equation follows a common pattern with normal inputs expecting a standard result. But as with any model, there are exceptions and outliers. When the equation falls outside your standard scenario and model guardrails, AI doesn’t get the final say. In addition to human review and oversight of all results, your process must require a human review and decision where the parameters are outside of the model norm. This human review not only ensures the right decision is made, but it may also be an opportunity to improve your model and in turn improving your control.
4. Change Management and Culture
A frequently underestimated aspect of AI implementation is organizational change management. Most focus on the technology and the customer but managing the effects AI will have on your organization are equally important. In a vacuum of information, rumor and misinformation quickly takes hold. AI will most certainly change the way the organization operates, what is prioritized, and yes, how it is staffed, but as with the advent of automated assembly lines and ATMs (Automated Teller Machines), the ways and places humans will work shall adapt.
Progress should not be feared but embraced. A critical component of controlling your AI is effectively controlling the change to your organization.
Conclusion
The future of banking is clear; AI is real and necessary to stay competitive. Jump on board or quickly be left behind. Banks must learn the balancing act of controlling their data and decisions while relinquishing decision making to non-humans. While this should evoke concern and night sweats for banking leaders, the path to safely and effectively giving up control to AI is quickly becoming safer and more clear. Not only does AI aid servicing and decision making, but it can also help you manage and control those actions and decisions. Leveraging this rapidly evolving technology will help you to stay relevant, meet stakeholder, regulator, and customer expectations, and improve your product and servicing offerings. Do not underestimate or shortcut the foundational work required for AI success. Clearly defined and managed governance, effective data management, human oversight, and organizational change management are the foundation of your success. You would not build a house on a weak and brittle foundation, don’t build your AI on one for the same reason.
