Citi's Services unit – a group that provides cash management, payments, trade finance, and asset custody for 19,000 corporate clients in 100 countries – has had a performance uptick the company is chalking up partly to AI.
"Services, our crown jewel, had an exceptional first quarter," CEO Jane Fraser said during the company's most recent earnings call. "The combination of client-driven growth and fees underpinned a 17% increase in revenues. Cross-border transactions were up 12%. Deposits grew by 16%, and assets under custody and administration were up over 20%."
AI is helping the business division achieve speed and efficiency, according to Shahmir Khaliq, Citi's head of Services.
"Our clients expect best execution, but they also expect the cheapest execution," Khaliq told American Banker. "They also expect us to drive down the cost of the service year over year. And that's really the agenda and the mission we're on."
AI is part of this agenda.
"It's not the only agenda," Khaliq said. "We're modernizing our platform as well. We're building out new client-facing protocols as well. So all of that should drive more efficiency, but it should also drive greater client interaction, client simplicity, client experience, which should drive new business into the bank, and therefore greater job opportunities across the board as well."
Citi's is using next-generation AI models to help people arrive at decisions faster and digitize client-facing and internal processes, he said.
Citi has developed a set of virtual assistants for Citi Services employees and customers. Most are based on Google Gemini, but some also use technology from Llama and Anthropic. (Bank executives declined to say specifically which models underpin each assistant.)
It's a message being heard across the finance ecosystem, as investors and analysts push banks and other companies to show performance results for the AI investments they're making.
"In the near term, I think the efficiency story is justified, but it's more likely to show up first as capacity creation than immediate cost takeout," said Devin Ryan, managing director and head of financial technology research at Citizens Bank. In other words, banks can use workflow copilots to reduce cycle times, cut errors and increase throughput without adding headcount.
"Over time, if AI capabilities advance and more systems become automated and truly agentic, I would expect to see a more direct impact on headcount and expense," he said.
Four uses for AI within the Services unit
One example of Citi Service's AI adoption is Agent Assist, which has been given to the unit's 1,200 customer-service agents, who receive more than three million queries a year.
"If a client comes in with a query, that agent finds the data, extracts the data, and puts in place an actual response to a client query based on what the client is asking for: Where is my payment? Has my payment settled? Has my security been serviced? Have I received the dividend payment into my account?" Khaliq said.
Agent Assist helps customer-service agents find the information they need quickly from a simple chat interface, where previously they had to go into multiple systems to look for answers. The bank says it's delivering more accurate answers than humans on their own.
Another virtual assistant, called Sales Assist, helps sales teams interact with clients. It analyzes clients' requests and provides background information, helping salespeople have a more intelligent conversation with the client when they're trying to close a deal or sell more products. An RFP response tool analyzes client information and requests for proposals and suggests responses.
A few customers are piloting a new chatbot on the Citi Direct banking portal called Client Assist. It lets customers "talk" to their data, for instance, to ask for the status of a payment or for an account balance.
Another virtual assistant, Ops Assist, helps with document processing and operational workflows. It automates much of the work of onboarding a new client. It scans documents, extracts needed elements and feeds that data into the bank's other software programs.
"Previously, when a client would deliver us a deck of documents — signature cards, account opening forms — it took a few hours," Khaliq said. "That time has been reduced by something like 80%, to 10 to 15 minutes."
This has also improved the client experience, he said.
"We deal with sophisticated, large multinational banks, broker-dealers and investors," Khaliq said. "They see Citi as a far more responsive, engaged, better execution-oriented counterparty. It's enabling our clients to get more and better and timely access to their information so they can make smarter decisions."
These use cases were chosen after a careful analysis of processes throughout the organization, according to Khaliq.
"We hand-picked those processes where we felt we could make the biggest difference," he said. "Take away a manual process and reduce the number of errors that a human makes, or take away a client-facing process which had significant turnaround time because it was a complex query. Now some of those tasks have become much simpler, so the human in the loop can review the data for context, review it for accuracy, review it with actual intelligence, as opposed to the manual labor of extracting it from multiple systems and putting it together and then writing out a response."
Not a road to layoffs, bank leaders insist
This accelerated use of AI doesn't necessitate headcount reductions, Khaliq said, because the Services unit has been growing. Transaction volumes, settlements, payment processing have all risen in the past six years.
"That means job opportunities, career development opportunities for people," Khaliq said. "So while we've gone through technology as a big change driver across the globe, technology has also given rise to new job opportunities as well and allowed people to progress their careers and continue to build momentum with new products, new services as well."
There are still humans in the loop and that will continue, according to Khaliq. "Not every model is 100% efficient, so you will need to have some oversight and some constant review of the model," he said.
"Depending on new business, depending on how efficient we get, the team sizes could go up or down," Khaliq said. "But overall, our agenda is growth. Our agenda is to do more with our clients over the next five to 10 years."
Selling AI to employees
Khaliq is trying to get his entire department to use these tools.
To keep people enthusiastic and to alleviate fears of AI related layoffs, "we over-communicate and make sure that our employees understand why, and they understand the outcome as well," he said.
Some jobs are changing. "A person who was previously constructing or responding to the queries by actually constructing the answers can now go into reviewing some of the answers," Khaliq said. "So that allows us to do more with a larger set of clients."
Some roles that Khaliq calls "manual" will change, he said. "But will the human entirely be taken out of the loop? Not at this particular point of time."
The employees who succeed will be those who put AI to use in their day-to-day work, he said.
More banks are likely to start drawing a line between their AI initiatives and performance numbers.
"Do investors care? Increasingly yes, but they're still looking for proof points and measurable outcomes, not demos," Citizens Bank's Ryan said. "We're starting to see more credible disclosure, but it's early — and the technology capabilities are evolving rapidly. Over time, investors will reward banks that can tie AI initiatives to hard key performance indicators and then translate those gains into either structurally better efficiency or revenue upside, through higher throughput, better client experience, and the ability to scale volumes without commensurate cost."
