Banks are increasingly relying on generative AI to streamline operations, enhance customer service, and drive decision-making. However, the effectiveness of these models hinges on the quality of the data they consume. Without robust data governance, even the most sophisticated AI can produce flawed outcomes. To address this, financial institutions are exploring a novel concept: a data nutrition label that provides transparency and accountability for the data feeding AI systems.
The Growing Importance of Data Quality in Banking AI
Data quality has long been a concern for financial institutions, particularly in the context of AI models. Regulatory guidance such as SR 11-7, issued in 2011, has emphasized the need for sound model risk management. Yet, high-profile failures, like JPMorganChase’s $6 billion trading loss in 2012 partly due to faulty risk data, underscore the stakes. Generative AI amplifies these risks, as it can absorb and propagate errors from unstructured or poorly curated data.
As Ned Carroll, head of data and automation at PNC, notes, “AI, absent well-organized, well-curated data, is nothing more than interesting math on a whiteboard.” This applies equally to traditional machine learning and generative AI. Unlike deterministic models that rely on structured data and clear ground truths, generative AI operates probabilistically, making data quality even more critical.
Introducing the Data Nutrition Label
The Financial Services Sector Coordinating Council (FSSCC) has proposed a data nutrition label framework to help banks assess and communicate the quality of data used in AI models. The label aims to answer a fundamental question: how do you attribute “goodness” to a data supply chain? This goodness can vary based on factors like input accuracy, timeliness, and relevance to the specific use case.
For instance, data used for marketing offers may not require the same rigor as data used for loan approvals, where risk exposure is higher. The nutrition label provides a standardized way to convey these quality levels, both internally and with external data vendors.
How the Label Works in Practice
In this framework, data owners attest to the quality of their datasets. When banks rely on third-party providers like credit bureaus or market data firms, they can request these vendors to certify the accuracy and timeliness of their data. This creates clear accountability, which is essential for effective data management.
Carroll emphasizes that without accountability, managing data quality becomes nearly impossible. The label also helps set expectations for data scientists and modelers, fostering a culture of responsibility. As he puts it, “It establishes the behavior expected of data scientists and modelers.”
Challenges and Limitations
While the data nutrition label is a promising step, it is not without challenges. Ian Schnoor, executive director of the Financial Modeling Institute, supports the idea of transparency but warns that labels can be misinterpreted. He draws an analogy to food nutrition labels, which sometimes list zero calories for cooking sprays—technically true for a brief spray but misleading if used liberally.
Similarly, data labels may be subject to interpretation or manipulation. Schnoor notes, “It’s not a foolproof be-all, end-all, but it’s a good start to transparency and protecting investors and the integrity of deals.”
Looking Ahead: The Future of Data Governance in AI
As generative AI becomes more embedded in banking, the demand for robust data management practices will only intensify. The data nutrition label offers a practical framework to address this need, promoting transparency and accountability across the data supply chain. While it may not solve every issue, it represents a meaningful step toward ensuring that AI models are built on a foundation of reliable, high-quality data.
For banks, adopting such standards is not just about compliance—it’s about safeguarding their proprietary knowledge and maintaining trust in an increasingly AI-driven world.