Citigroup

Citigroup entered its AI transformation under direct regulatory pressure. In 2020, the Federal Reserve identified deficiencies in the bank's data quality management — the result of what CEO Jane Fraser acknowledged as "decades of underinvestment" in data infrastructure. Those deficiencies culminated in $135.6 million in regulatory penalties levied jointly by the Federal Reserve and the Office of the Comptroller of the Currency in 2024. Simultaneously, Citi operated thousands of legacy applications across its global commercial and corporate banking operations, creating fragmented data pipelines that slowed software development velocity and complicated compliance reporting. The status quo carried both direct regulatory costs and significant operational drag that made large-scale modernization non-negotiable.

Citi's response was a coordinated AI and infrastructure transformation executed across multiple tracks. The bank equipped 30,000 developers with AI-powered coding tools — built on Google Cloud's Vertex AI platform as the enterprise AI foundation — integrating automated code review capabilities directly into the development workflow. Alongside the developer tooling, Citi deployed two internal chatbot assistants to support routine operational tasks across business lines. To drive enterprise-wide adoption, the bank recruited Tim Ryan, a former PwC senior partner, to lead technology and business enablement, and hired Dipendra Malhotra from Morgan Stanley to head wealth technology. On the customer-facing side, Citi introduced "Agent Assist," a generative AI tool for U.S. Personal Banking customer service, initially piloted in credit cards. Technology and communications investment reached $2.4 billion in Q1 2025 as the buildout accelerated.

The developer AI tool completed approximately 220,000 automated code reviews, measurably accelerating throughput across Citi's engineering organization. Legacy modernization showed compounding progress at scale:

• 220,000 automated code reviews completed by the developer AI tool
• 130 applications retired or replaced in Q1 2025 alone
• 2,000+ legacy apps decommissioned over the prior three years
• $2.4B technology and communications investment in Q1 2025, up from $2.2B in Q1 2024

CEO Fraser noted that "many of the efforts are now impacting how we run the bank better and more efficiently," indicating that infrastructure gains are beginning to translate into measurable operational performance improvements.

• Regulatory compliance failures can generate the organizational urgency — and budget authority — required to execute large-scale infrastructure modernization that deferred maintenance programs cannot achieve.
• Deploying AI coding tools at true enterprise scale (30,000 developers) compounds legacy decommissioning velocity beyond what manual migration programs can achieve alone.
• Data quality remediation must precede broad AI adoption; clean, governed data pipelines are a prerequisite for reliable AI outputs in regulated industries like banking.
• Leadership composition matters: recruiting senior AI executives from peer institutions accelerated both governance frameworks and cultural adoption across the enterprise.