How Citi deployed AI agents to 4,000 employees at scale

Most enterprises treat AI adoption as a pilot program — small teams test tools, present results, and struggle to scale beyond departmental silos. Citi took a fundamentally different approach: building an internal workforce of 4,000 AI-enabled employees across technology, operations, risk, and customer support.

With over 70% of its 182,000 global workforce now using approved AI tools in production, Citi's rollout offers critical lessons for enterprises moving from experimentation to operational deployment. The bank's strategy prioritizes people over technology, distributed ownership over centralized control.

People-First Deployment Strategy

Rather than starting with tool selection, Citi focused on human infrastructure through its AI Champions and AI Accelerators programs. The bank recruited volunteers from across the organization to serve as local support resources, not formal trainers.

This peer-driven model addresses a fundamental adoption challenge: tools fail not because they lack features, but because users don't understand when or how to deploy them effectively. By embedding AI expertise within existing teams, Citi eliminated the gap between experimentation and routine workflow integration.

Key components of the people-first approach include:

• Volunteer recruitment — employees self-selected into champion roles rather than being assigned
• Local support networks — champions provided team-level guidance and troubleshooting
• Distributed expertise — reduced dependency on centralized AI specialists
• Cross-functional participation — non-technical roles actively engaged in AI implementation

Training and Credentialing Framework

Citi implemented a badge-based training system where employees earn internal credentials by completing courses or demonstrating practical AI implementation in their daily work. While badges don't translate to promotions or pay increases, they create organizational visibility and peer credibility.

This gamification approach accelerated adoption faster than traditional top-down mandates. The training framework includes:

• Practical demonstrations — showing real workflow improvements using AI tools
• Peer recognition — badges create social proof within teams
• Incremental learning — modular courses allow flexible skill development
• Use case documentation — employees share successful AI applications across departments

Risk Management and Governance

In regulated environments like banking, trust often matters more than speed. Citi limited employees to firm-approved tools with explicit guardrails around data usage and output handling.

This constraint slowed some experiments but enabled broader access by making managers comfortable with distributed AI deployment. The governance model balances innovation with compliance requirements essential for financial services.

Production Use Cases and Applications

Citi's AI deployment focuses on everyday efficiency gains rather than transformational moonshots. Current applications span document summarization, internal communication drafting, data analysis, and software development assistance.

The scale advantage becomes clear when small efficiency gains multiply across thousands of employees. Rather than asking whether AI could transform the business, Citi asked where it could remove friction from existing processes.

Infrastructure-First Mindset

By treating AI as infrastructure rather than innovation, Citi made progress easier to measure and reduced pressure for dramatic results. This framing shift proves critical for enterprise adoption — focusing on practical workflow improvements over theoretical transformation potential.

The bank's approach challenges the assumption that AI adoption must start at the executive level. While senior leadership provided support, momentum came from employees who volunteered time to learn and teach others.

Scaling Challenges and Solutions

Peer-led adoption depends on sustained interest, and team adoption rates vary significantly. Some groups benefit more than others, creating potential inequality in AI access and expertise.

Citi addresses these challenges through:

• Champion rotation — preventing knowledge concentration in specific individuals
• Regular content updates — keeping training materials current as tools evolve
• Cross-team knowledge sharing — documenting and distributing successful use cases
• Feedback loops — continuous improvement based on user experience data

Lessons for Enterprise AI Deployment

Citi's model sidesteps common production challenges by distributing ownership across teams while maintaining centralized governance. This hybrid approach addresses talent gaps and unclear ownership that plague many enterprise AI initiatives.

The cultural signal proves equally important — encouraging non-technical employees to participate demonstrates that AI becomes standard business infrastructure, similar to spreadsheets or presentation software in previous decades.

Bottom Line

Scale doesn't come from buying more tools but from helping people feel confident using existing ones. Citi's experiment shows that bottom-up energy often determines whether new technology sticks in large organizations.

For enterprises wondering why AI progress feels slow, the answer lies less in strategy documents and more in how work actually gets done, one team at a time.