BofA's Hari Gopalkrishnan on AI Strategy with Julie Hyman

Bank of America's AI Investment Scale

• The firm allocates $14 billion annually to technology, with over $4 billion specifically designated for AI initiatives.
• Current infrastructure supports approximately one exabyte (1,000 petabytes) of client interaction data, processed with a focus on privacy and security.

Strategic Deployment Philosophy

• The bank has shifted strategy from "proof of concepts" to "diffusion," moving from 500-person pilots to enterprise-wide rollout.
• Problem-First Approach: Solutions are driven by specific business problems rather than available AI capabilities; for example, deterministic rules were chosen over AI agents for simple payment processing tasks to ensure ROI.
• Core Pillars: AI adoption is strictly governed by three criteria: enhanced customer experience, measurable return on investment (ROI), and safety/soundness (security).

Operational Use Cases and Metrics

• Relationship Management: AI tools assist advisors by automating data retrieval (system A, B, C) to prepare meeting summaries, saving time on "toil" work.
• Adoption Numbers: Capabilities for managing client meetings have been deployed to 25,000 advisors, with plans to scale to 50,000 teammates.
• Credit Underwriting: A pilot program is underway to assist underwriters in drafting credit approval memos, expected to deploy at scale by year-end.
• Erica 2.0: A new iteration of the digital assistant is in development to integrate generative AI while maintaining strict safeguards against hallucinations for client-facing interactions.

Workforce Impact and Hiring

• Recruitment Continuity: Despite efficiency gains, the bank hired 20,000 people this year and announced a 4,000-person campus class, noting that significant work remains and new threats require human oversight.
• Productivity Logic: The goal is not headcount reduction but increasing leverage; for example, a banker covering X clients today could cover X+4 tomorrow using AI.
• Employee Sentiment: Internal listening sessions indicate high engagement and a desire to participate in design, countering general "anti-AI" narratives; resistance was minimal when framed as tools to remove barriers to serving clients.

Infrastructure and Cost Strategy

• Hybrid Model: The bank utilizes a mix of on-premises GPUs, cloud infrastructure, open-source models, and frontier models to optimize costs and performance.
• Cost Management: The firm avoids "spray and pray" spending; token costs are monitored, and the strategy focuses on "right tool for the right purpose" to mitigate risks of rising inference costs.
• Learning Loop: Internal AI applications are tested on 200,000+ teammates first to gather feedback, which informs subsequent client-facing rollouts.

Forward-Looking Statements

• Bank of America projects a steady, engineered run state for AI deployment rather than a slowdown, contingent on continued ROI validation.
• The firm expects to eliminate "toil" to allow employees to focus on high-value judgment activities, thereby reducing client backlogs rather than reducing the total workforce.
• Future investments will prioritize "Upskilling" initiatives, including training on prompt engineering and "forward design" roles, through the internal "Academy."