Citi’s Next Chapter in AI: From Traditional Models to Generative Intelligence

Jon Lofthouse from Citi explains how they are moving from long-standing, “traditional” AI into a new phase powered by generative AI and large language models. Lofthouse says AI isn’t new to Citi, but these newer models are flexible enough to be used across both internal productivity and client-facing services, opening up far more use cases than before.

Lofthouse breaks Citi’s approach into horizontal and vertical layers. Horizontally, the goal is to “give AI to everyone” inside Citi which includes an internal assistant that lets employees ask natural-language questions like “How many vacation days do I have?” or “What’s our policy on expenses?” without digging through portals. This is built on a RAG setup, which lets staff upload documents, ask questions, and get summaries which is also now gaining an “agent mode,” so it can start performing chained tasks rather than just returning answers. In technology teams, coding assistants help developers write and refactor code, while an internal tool for code reviews and quality checks.

Vertically, Lofthouse highlights client servicing, especially in the North American consumer business. Generative AI is used to understand customer intent and route them into the right path in Citi’s IVR system, instead of relying only on rigid menus. After calls, AI generates summaries, reducing manual note-taking and improving records. The same stack supports frontline agents by surfacing faster, more relevant responses during live interactions. Overall, Citi positions generative AI and hyperscaler LLMs as core capabilities being embedded into everyday tools, developer workflows and customer journeys, rather than a standalone experiment.