Breaking Complexity in Banking Risk Operations With AI - with Nathaniel Bell of Wells Fargo
🎯 Summary
Summary of AI and Business Podcast Episode: AI in Auditing Workflows with Nathaniel Bell
This episode of the AI and Business Podcast, featuring Nathaniel Bell, Data Management Director at Wells Fargo, focuses on the practical application of AI—particularly deterministic and foundational models—in reshaping traditional enterprise auditing workflows within the financial sector. The discussion bridges the gap between advanced AI concepts (like agentic AI discussed previously) and immediate, high-impact automation opportunities.
Here are the key takeaways for technology professionals:
1. Main Narrative Arc & Key Discussion Points: The conversation centers on how AI, even simple, rules-based systems, can significantly enhance the efficiency and effectiveness of auditing, moving processes from manual and reactive toward continuous monitoring. The core challenge discussed is balancing technological innovation with the inherent caution and regulatory demands of the financial industry.
2. Major Topics and Subject Areas Covered:
- AI in Auditing: Reshaping workflows, identifying and addressing risk more efficiently.
- Workflow Stages: Information gathering, control testing, and moving toward continuous auditing.
- Data Quality: The critical prerequisite for achieving continuous monitoring.
- Subjectivity vs. Objectivity: How digital transformation and codified processes reduce human subjectivity in risk assessment.
- Team Assembly: The necessary skill sets for modern audit teams navigating digital transformation.
3. Technical Concepts, Methodologies, or Frameworks Discussed:
- Deterministic AI: Highlighted as the easiest and most immediate form of AI deployment for rules-based auditing tasks.
- Generative AI (LLMs): Suggested for accelerating the initial, time-consuming information-gathering phase (e.g., querying models to describe a service or identify key risks).
- Agentic AI: Briefly referenced as a more advanced future state, contrasting with the immediate focus on simpler deployments.
- Continuous Auditing: The end goal for proactive risk management, contingent upon high-quality, connected data streams.
4. Business Implications and Strategic Insights:
- High ROI on Manual Processes: Nathaniel strongly advises targeting highly manual processes first, as they offer the biggest “pain for the buck” and are prone to human error (noting that even complex spreadsheet checks only catch about 70% of errors).
- Codifying Processes: Digital transformation, especially with AI integration, forces the codification of processes, leading to more objective risk assessments, which is a strategic advantage in regulated environments.
- Efficiency vs. Depth: Automation in data gathering frees auditors to focus on higher-value activities like storytelling, systemic risk assessment, and strategic observation rating rather than tracking individual anomalies.
5. Key Personalities and Experts Mentioned:
- Nathaniel Bell (Data Management Director, Wells Fargo): The expert guest, providing practical insights from his experience in audit and data management, including a past background in the risk-averse nuclear industry.
- Matthew Damello (Editorial Director, Emerge AI Research): The host, framing the discussion and drawing parallels to other subjective fields like transfer pricing and clinical trials.
6. Predictions, Trends, or Future-Looking Statements:
- The shift from manual/reactive auditing to continuous/proactive auditing is inevitable, driven by data connectivity.
- The role of the human auditor will evolve to focus on interpreting patterns identified by AI, validating systemic gaps, and providing subjective risk ratings based on broader context.
7. Practical Applications and Real-World Examples:
- Using LLMs to quickly summarize complex service descriptions and identify necessary controls, replacing time-consuming meetings with multiple stakeholders.
- Applying deterministic models to test controls based on established rules, immediately flagging gaps.
8. Controversies, Challenges, or Problems Highlighted:
- Data Quality Dependency: Continuous monitoring is impossible without accurate, complete, and well-controlled data flowing directly from source systems.
- Organizational Comfort: There is an emotional or comfort-based resistance to moving away from familiar, manual processes (like Excel sheets) toward continuous monitoring systems.
- Complexity for Auditors: As processes become more digital and codified, auditors must develop skills to examine the underlying code and transformation logic, increasing complexity.
9. Solutions, Recommendations, or Actionable Advice Provided:
- Start Simple: Focus initial AI deployments on deterministic, rules-based use cases within highly manual phases for immediate effectiveness gains.
- Prioritize Data Connectivity: Audit leaders must focus on building reliable data pipelines from source systems to enable any form of continuous monitoring.
- Embrace Transformation: Automating manual tasks will inherently lead to process improvement and allow teams to focus on high-level strategic risk assessment and storytelling.
10. Context About Why This Conversation Matters to the Industry: This conversation is crucial because it grounds the discussion of enterprise AI in the highly regulated and risk-averse financial services sector. It provides a roadmap for audit and compliance professionals to adopt AI incrementally, proving ROI with deterministic tools before attempting large-scale generative or agentic overhauls, thereby ensuring regulatory compliance while driving efficiency.
🏢 Companies Mentioned
💬 Key Insights
"Secondly, he challenges the common belief that organizations must move step by step from deterministic to generative to agentic AI, arguing that isolated, high ROI deployments are often the smarter path forward."
"In the end, you're going to hopefully have a team that now can spend more time on what you really want, which I think you made a great point, Matt, which is storytelling and saying, hey, we didn't find a gap in one place. This is a systemic gap in here is why."
"If I look at manual process, I already know there's likely a human element of error. I think the last time I researched this just on a complex spreadsheet, we're going to catch like 70% of the errors. That's it. That's our threshold."
"I always look at the highly, highly manual processes for a couple of reasons. I think that is usually the biggest pain for the buck."
"You can be working with numbers. You can be doing accounting. You think it can be black and white. But eventually you do need to tell a story with the data."
"eventually you do need to tell a story with the data. And that is... that's humanities, that's subjective, even when we're trying to get objective information out of it."