Citi mandates AI prompt training for most employees

A Citi sign outside a bank location.
Victor J. Blue/Bloomberg

  • Key insight: Citi is putting most of its employees through prompt training in the hopes of improving productivity.  
  • What's at stake: Poor prompting risks degraded competitiveness and slower operational outcomes.  
  •  Forward look: Prepare for expanded agentic AI rollouts and governance decisions across financial firms.

Source: Bullets generated by AI with editorial review.

Citi is kicking off a new AI training program that will teach employees how to write better prompts for the bank's generative AI programs.

In an internal memo shared with American Banker, Tim Ryan, Citi's head of technology and business enablement, and Anand Selva, the bank's chief operating officer, stressed the importance of well-written prompts.

"Just as the right question in a client pitch can reveal clarity and create advantage, a well-crafted prompt can accelerate your work, surface insights and amplify your impact," Ryan and Selva said. The memo was sent to 175,000 employees in 80 locations on Tuesday morning.

So far this year, Citi employees have entered more than 6.5 million prompts in Citi AI tools, "turning work that once took hours into tasks done in minutes," the memo said. "The scale isn't just impressive; it marks the beginning of a new way of working."

The new mandatory training is called, "Asking Smart Questions - Prompting like a Pro." Employees have 60 days to complete it. It was built using a platform that tailors the learning experience to the user's existing knowledge. Experts can complete it in under 10 minutes, while beginners take about 30 minutes, the bank said.

"As colleagues learn to use AI, there is a natural process of testing how to write prompts to produce the best results," Peter Fox, head of learning at Citi, told American Banker. "This training is about teaching our colleagues the possibilities of great prompting versus basic prompting."

Citi is not alone in seeing better prompts as a key to more productive use of generative AI.

"The output of AI is going to be only as good as the prompt and the system prompt," said Mike Gaultieri, vice president and principal analyst at Forrester Research, in a recent interview about AI workslop.

"That output could be very wordy or it could be too succinct," he said. "And a lot of this has to do with the prompt and the system prompt – what gets added to the prompt the user is using. Generative AI "is still very, very immature in terms of people knowing how to use this stuff."

Alex Sion, who is currently financial services lead at Blend and formerly was general manager of mobile banking at JPMorganChase and managing director of Citi Ventures, seconded the need for smart prompting.

"There's an old saying: 'Judge a person by their questions rather than by their answers,'" Sion told American Banker. "Generative AI makes that maxim more relevant than ever. When you're interacting with a system that can tap into vast stores of human knowledge, the difference between an average and an exceptional outcome lies in the precision, creativity and structure of the questions." The barrier isn't the machine's capability — it's the human's ability to engage with it thoughtfully, he said.

In addition to prompting, "you have to focus on improving and truly educating teams on the underlying data sources and the AI workflows built around them," said Brad Leimer, executive leader of AI and fintech partnerships at Darrery Capital, who will be a guest on the next American Banker podcast.

More than 175,000 employees have access to Citi's generative AI tools, Citi Stylus Workspaces and Citi Assist. Stylus Workspaces offers document intelligence and a virtual conversation assistant tool for summarizing and comparing large documents, draft presentations and speaker notes and Q&As, research and brainstorming, Fox said. Citi Assist is a desktop AI assistant that helps employees navigate Citi policies and procedures in compliance, risk, HR and finance. The bank started rolling these out last December.

Last week, Citi upgraded Stylus Workspaces with agentic AI, artificial intelligence capable of making autonomous decisions and taking actions with minimal human oversight. This has been given to 5,000 users of Citi Stylus Workspaces. Employees can use the system for research, to extract insights from datasets and to streamline workflows. For instance, with one prompt, Citi Stylus Workspaces with agentic AI can identify the top five branded card businesses in the U.S., explain their strategic goals, and translate the answers into Spanish.

In July, Citi began rolling out agentic AI to its 40,000 developers, who are using it to automate simple tasks like software patch and upgrades. The bank is using Cognition's Devin agent. Developers also use GitHub Copilot for real-time code suggestions, debugging help and code review automation.

"We're giving our people smarter, faster and more connected tools so they can focus less on manual tasks and more on the big ideas that drive our business forward," said Chief Technology Officer David Griffiths of the agentic AI rollout.

Citi's new AI training push comes a few days after Accenture laid off 11,000 people as part of a "business optimization program" that is refocusing the firm on AI.

"We are exiting on a compressed timeline, people where reskilling, based on our experience, is not a viable path for the skills we need," CEO Julie Sweet said in the company's second quarter earnings call.

Sion pointed out that technical skills can be taught, but an AI-first mindset is harder to instill.

"Successful AI users tend to share certain traits: curiosity, creativity and critical reasoning," Sion said. "They approach problems with a 'life-hacking' mentality — looking for better, smarter and more efficient ways of doing things."

For reprint and licensing requests for this article, click here.
Artificial intelligence Citigroup Technology
MORE FROM AMERICAN BANKER