The Key Components of a Gen-AI Program with Steve Holden, SVP & Head of Analytics at Fannie Mae

Introduction

In November of the previous year, Steve Holden was tasked with leading the implementation of a generative AI program at Fannie Mae. To build a successful program in an organization like Fannie Mae, which operates in a highly regulated financial landscape, he identified three guiding principles: balance, transparency, and humility. These principles have become foundational to the approach being taken in the establishment of the generative AI program.

Balance

Holden emphasizes the importance of balance in their approach to generative AI. He acknowledges the risks associated with moving too fast or too slowly, especially in a heavily regulated environment. The key is to make progress without compromising on responsibility. Companies that rushed into AI initiatives often faced significant challenges, making it crucial to find a middle ground. Throughout the development of the AI program, this principle of balance recurs repeatedly.

Transparency

Another cornerstone of Holden's approach is transparency. He actively shares insights through weekly blogs and bi-weekly knowledge-sharing sessions where engineers present their work. This ensures that information flows freely across the organization, allowing for shared learning and decision-making. Advisory groups and councils are utilized to maintain transparency and to disseminate knowledge systematically throughout the divisions.

Humility

Perhaps the most critical of his guiding principles is humility. The generative AI landscape is dynamic; developments occur rapidly, and decisions made today may become outdated tomorrow. Holden advocates for a "test and learn" strategy, encouraging an adaptable mindset. He references a concept introduced at an AWS conference by Jonathan Allen, distinguishing between one-way and two-way doors in decision-making processes. Early-stage decisions should be reversible to maintain flexibility as new information emerges.

Key Considerations for Speed

Holden outlines several factors that influence the pace at which Fannie Mae can innovate with AI, beginning with governance. Aligning the governance process is vital since multiple individuals manage various risk aspects within the institution. Effective governance requires prioritization due to the volume of ideas needing review. He articulates four critical criteria for determining risk and alignment:

1. Risk Level - Starting with low-risk use cases is advisable, focusing initially on internal productivity.
2. Uniqueness - New ideas should not duplicate efforts already underway in the organization to foster efficiency.
3. Tech Stack Alignment - New initiatives should align with existing technology to minimize disruption and enhance compatibility.
4. Clear Business Use Case - Teams must articulate the expected business outcomes clearly for any proposed initiative.

Through this structured approach, Holden aims to strike the right balance while ensuring transparency and fostering a culture of humility in Fannie Mae’s generative AI program.

Keywords

• Generative AI
• Balance
• Transparency
• Humility
• Governance
• Risk Level
• Business Use Case
• Tech Stack Alignment

FAQ

1. What are the guiding principles of the generative AI program at Fannie Mae?
The guiding principles are balance, transparency, and humility.

2. Why is balance important in AI implementation?
Balance is crucial to avoid moving too quickly or too slowly, especially in a regulated environment like finance.

3. How does Steve Holden promote transparency within the organization?
He shares insights through weekly blogs, bi-weekly knowledge-sharing sessions, and utilizes advisory groups for systematic knowledge dissemination.

4. What does humility mean in the context of Fannie Mae's generative AI initiatives?
Humility refers to the acknowledgment that the AI landscape is constantly changing and encourages a "test and learn" strategy for flexibility in decision-making.

5. What criteria does Fannie Mae consider when prioritizing AI use cases?
The criteria include risk level, uniqueness, tech stack alignment, and clear business use cases.