Beyond the AI Black Box: why financial institutions are refocusing on data foundations

While the industry conversation continues to accelerate around AI, GenAI and agentic solutions, one thing became very clear throughout ABA’s panel session and the interactive roundtable discussions that followed: before financial institutions can scale AI confidently, they must first address their data foundations and data governance capabilities.

With the room primarily filled with Risk and Compliance leaders, conversations were not theoretical or aspirational based on unspecified future innovation. Instead, they were pragmatic, with tactical discussions centered around trust, accountability, explainability, operational risk, and how organizations can realistically begin their governance journey. Very quickly, the focus shifted away from AI strategy and toward a more fundamental set of questions:

• Can we explain where our data comes from? 
• Who owns it? 
• How has it been transformed along the way? 
• Which data actually matters most?
• And how do we get started without trying to govern everything at once?

During the initial panel discussion, we explored an important reframing of governance. Too often, organizations define data governance in terms of the existence of policies, committees or controls. 

In practice, governance is much more operational than that. At its core, data governance is about building confidence in the decisions that organizations make using data. Data management does not have an end-state, which in turn requires data governance to be comprehensive, consistent, and continuous.

The key challenge is not necessarily that data might be ‘wrong.’ The larger issue is often the absence of clear ownership, visibility and accountability across the data lifecycle. Teams consume data without fully understanding how it was created, transformed, filtered or enriched before reaching them.

Historically, many banks could manage the complexities of data governance through traditional reporting and manual oversight processes. But AI and advanced analytics fundamentally change that equation. Each additional layer of abstraction – models, aggregations, AI outputs and automated decisioning – increases the distance between decision-makers and the underlying raw data.

As a result, financial institutions increasingly rely on output without fully understanding the inputs.
That is where real risk begins to emerge.

One of the strongest themes that surfaced during the roundtable discussions was the realization that organizations cannot govern all data equally. Several groups initially described their challenge broadly as a ‘data quality problem,’ but as conversations evolved, the underlying issues became clearer:

• inconsistent ownership
• fragmented business processes
• lack of lineage transparency
• siloed reporting logic
• and limited understanding of which data elements truly drive critical decisions.

This led naturally into discussions around Critical Data Elements (CDEs) – the subset of data that directly impacts regulatory reporting, financial reporting, risk management and strategic business decisions. The most productive discussions were not centered on enterprise-wide transformation programs or large-scale tooling implementations. Instead, participants focused on practical starting points:

• identifying one meaningful business problem
• clarifying accountability
• improving visibility into lineage and transformations
• and establishing targeted governance around the data that matters most. 

Another recurring theme was the importance of stakeholder alignment. Effective governance requires the participation of multiple business lines and enterprise functions – it cannot sit solely within technology or centralized data teams. There must be coordinated ownership across business teams, Risk, Compliance, Finance, Operations, Audit and Technology.

The conversations around data fluency and literacy also highlighted the importance of building a strong data culture. Financial institutions cannot transform governance through policy documentation or annual training exercises alone. Sustainable progress happens when organizations embed shared language, accountability and the understanding of data directly into operational processes and day-to-day decision-making. 

Building a stronger data culture is often what enables governance programs to move from theoretical frameworks into operational reality.

Ultimately, the session reinforced a broader industry reality: the financial institutions most likely to succeed with scaling AI and controlling its risks will not necessarily be the ones with the most sophisticated models. They will be the organizations with the strongest confidence in the data underpinning those models.

At Capco, we partner with financial institutions across the maturity spectrum, from designing and implementing scalable data governance capabilities, to helping financial institutions respond to regulatory findings and remediate governance gaps. This includes helping clients explore how agentic solutions can support and enhance modern data management capabilities, governance operations and decision transparency. If these challenges resonate with your organization, we would welcome the opportunity to continue the conversation.