From AI curiosity to AI capability: A data driven path for financial institutions 

How important is data in determining the success of an AI program?

Across every stage of AI maturity, one factor consistently determines success: data.

AI systems are only as effective as the data that powers them. If data is siloed, inconsistent, inaccessible or poorly governed, the resulting models will be limited at best and risky at worst. For many institutions, the biggest obstacle is not a lack of AI ambition; it is fragmented data environments. Core systems, subledgers, document repositories and business applications often do not connect cleanly. That makes it difficult to build a reliable picture of the customer or generate meaningful insights.

High-performing institutions treat data as a strategic asset rather than a byproduct of operations. That shift is crucial. They invest in:

• Improving data quality and consistency
• Centralizing and integrating key data sources
• Establishing governance and traceability
• Enabling broader, secure access to trusted data

Better data enables stronger fraud detection, better anti-money laundering (AML) outcomes, more effective personalization and smarter product recommendations. It also supports the kind of natural-language access to information that many institutions now want from AI tools.

What foundational capabilities are needed to support safe, transparent and scalable AI adoption?

Many AI initiatives struggle to produce return on investment (ROI). Often, the problem is not the technology itself, but rather the absence of clear production criteria and value metrics. A pilot should not just be an experiment for experimentation’s sake. Institutions need to define in advance what success looks like, what performance thresholds must be met and how value will be measured. Whether the goal is reducing processing time or improving fraud detection accuracy, clear metrics are essential to guide decision-making and justify further investment.

Beyond this, moving from pilot to production is not just a technical exercise. It requires institutions to think about how models will be monitored, retrained and governed over time. This is especially important because AI changes the environment around it. As employees and customers interact with AI-enabled systems, behaviors shift. That can change the underlying data and create drift in model performance. Institutions therefore need mechanisms for version control, observability and re-training. Leading institutions do not treat model deployment as the finish line. Instead, they treat it as the beginning of an ongoing feedback loop.

Finally, even the best AI solution will fail to deliver value if employees do not trust it, understand it or use it consistently. That is why change management, AI literacy and internal support structures matter so much. It is important to think carefully about the human side of AI adoption, because people need training and teams need clarity on when and how to use AI. Leaders need to communicate why the technology matters and how it fits into the institution’s broader strategy.

In many cases, that “last mile” of adoption is what determines whether an AI investment creates a real return.

What are some practical frameworks for effective AI governance?

As AI adoption expands, so too does the importance of governance. For financial institutions, innovation cannot come at the expense of risk management and regulatory compliance. It is important to note that many existing regulatory frameworks already apply to AI. If a model influences decision-making, it falls within established model risk management expectations, including documentation, independent validation and ongoing monitoring. Key considerations include:

• Model governance: Ensuring transparency, validation and performance monitoring
• Data privacy: Managing personally identifiable information through techniques such as de-identification or synthetic data
• Third-party risk: Understanding and governing AI embedded within vendor solutions
• Fair lending and bias: Ensuring decisions remain explainable and equitable

The practical challenge is that governance often lags behind innovation. Many institutions have oversight structures built for a smaller or less complex operating model. As AI-use expands, that mismatch can become a source of examiner concern. Good governance begins with a simple but essential question: Do you know what AI is already being used within your institution? That includes not just internally developed models, but also AI embedded in vendor solutions and core platforms. Institutions should maintain a comprehensive inventory of AI use cases – both internally developed and vendor-provided – to support audit and regulatory readiness.