By John Hintze
Citi Wealth describes its AI-generated avatar Citi Sky as an “always-on AI-powered member of the Citi Wealth team.” In a video demonstration, the avatar discusses with a human client the options to renew a maturing CD and submits the client’s choice to be executed. It then confirms the college fund for the client’s daughter is on track, noting that she must be excited about her college acceptance.
The client asks about other considerations, the avatar notes a recent rate cut that could provide mortgage refinancing opportunities and offers to schedule an appointment with the customer’s advisor.
Citi Sky applies generative AI, which creates new and original content, and in this case enables the avatar to interact directly with wealth-department customers.
“Over time, the new capability will create a more intuitive, responsive and personalized wealth experience — elevating the Citi Wealth client journey while empowering advisors,” Citi says, noting the rollout in phases to Citigold clients.
As one of the largest national banks, Citi’s customer-facing generative AI agent is leading the banking pack, according to the results of an ABA survey published in March. Surveying 250 banks with assets between $75 million and $600 billion, ABA found that early use cases are narrow, internal and efficiency-focused, prioritizing low risk applications that support human judgment. It also found that most banks remain focused on traditional AI — mainly machine learning.
“Generative AI, by contrast, remains earlier stage and tightly controlled, with significantly lower confidence in near term benefits due to data security, accuracy and regulatory concerns,” according to the report.
Nevertheless, the results of the survey point to the conclusion that “the advantage will accrue to banks that build internal understanding and governance capabilities through controlled use.”
United Community Bank has taken an active if cautious approach to implementing AI. The community bank operating across six Southern states is focused now on using the technology to augment its current capabilities, says Abraham Cox, chief consumer and small business banking officer at the $28 billion-asset bank.
Bankers in its wealth management business, like the rest of the bank, are now using Microsoft Copilot internally to develop presentations, take notes, and otherwise make their communications more efficient, whether internally or with clients, as well as improve the speed and quality of generating insights into customers.
“That allows our private bankers, our advisors, to be able to spend more time with customers on more value-added types of duties and responsibilities,” Cox said.
Cox says the bank does not anticipate deploying AI involved in wealth-customer interactions or decision-making anytime soon. Its CIO, however, is leading efforts to identify potential internal projects and third-party tools that would apply AI to reduce fraud risk, administrative work and improve the customer experience. Facilitating employees’ access to and understanding of human-resource policies is one consideration. On the wealth front, for example, Cox adds, AI could enable bankers to more efficiently review and apply the bank’s mortgage guidelines and policies.
“In a controlled, internal fashion, can we use AI to digest all this information?” Cox asks. “Then once tested, AI would enable bankers to understand quickly what the policy is for loans offered to wealth customers.”
South State Bank has given all employees within the bank access to Copilot and some to Copilot Premium. The latter unlocks AI functionality inside Microsoft Office, such as specialized AI assistants that, for example, gather, summarize and analyze complex data.
As the technology becomes more widespread and newsworthy, examiners are digging deeper into things such as contracts, validation techniques and more.
“We’re working on giving everybody a baseline understanding of the tools” within a certain timeline, said George King, EVP of the $68 billion Florida-based company.
The institution recently started taking a more granular approach to determine how copilot can be applied to specific job functions, by pairing bankers in those roles with IT professionals to customize existing AI tools.
“So, if you’re a relationship manager, what can we do with the existing tools to create efficiency specific to your job function and improve our client experience?” he asks.
South State is evaluating third-party options that employ AI, such as a tool it recently acquired independently from its existing wealth-services vendors that evaluates and summarizes trust and estate documents and provides a user-friendly presentation for clients. Such presentations could have previously taken financial planners 10 or more hours to create.
“We would not have had the capacity to devote that level of resources to a client,” he said.
Cox said that his bank’s core wealth vendors have yet to deploy AI capabilities in their platforms. And while the bank has yet to adopt wealth-specific AI tools, he said, it is looking at third-party offerings, such as Jump’s AI assistant that facilitates note-taking and preparing for meetings.
More choices may soon arrive. Anthropic, for example, announced in May that it was issuing 10 “ready-to-run” agent templates covering what it calls the most time-consuming work in financial services, including building pitchbooks, screening KYC and closing books at month’s end.
The largest banks are already leaps ahead in terms of AI assistants. BNY’s Eliza platform, launched in 2023, has deployed 140 autonomous AI agents, which it refers to as digital employees, to aid its bankers across a variety of tasks. It is currently building agents to analyze complex trust documents, and while such applications are not new, says Alvina Lo, head of advice, planning and fiduciary services at BNY Wealth, the agents feed the information back into BNY’s proprietary wealth management system.
“That will help tremendously with work flow,” Lo says, adding that the bank is also considering AI agents to streamline trust administration reviews, account onboarding, distributions and wealth-account terminations.
BNY is considering agentic AI to help trust officers to determine whether the distributions are in line with trust agreements and with balance-sheet needs of beneficiaries, and to recommend issues to consider for specific clients. Lo adds the bank may opt for third-party tools in well-tread areas such as estate planning. For more specific functions, particularly in complex areas such as trust, BNY employees often point the way.
“Some of the best ideas have come from junior employees doing the analytical work who often understand the technology better,” Lo says.
Given AI’s tendency to “hallucinate,” Lo emphasizes that a trust officer must review all decisions generated by AI agents before they are acted upon.
Banks must also consider compliance risk. Ryan Miller, senior counsel of innovation policy at ABA, says regulators typically examine AI through several risk lenses, including model risk, third-party risk, data privacy and security, and the impact of AI models on protected characteristics such as race and gender.
Miller says banks are “well served” by documenting their governance programs, including policies and procedures, formal agendas, minutes for key enterprise meetings and employee training materials/logs. They should also keep records of validation techniques and decisions, and their respective audit trails showing the work.
“Regulators want to see that banks are thinking about the risks presented by traditional AI use and are taking steps to institute appropriate controls,” Miller says. “As the technology becomes more widespread and newsworthy, examiners are digging deeper into things such as contracts, validation techniques and more.”
He adds regulators are anticipated to provide guidance specific to newer forms of AI such as generative and agentic in the near future, and a lot of the same principles will apply.
In terms of next steps, King says, South State Bank is exploring how to pull information across its distinct technology platforms to one place, to further improve efficiency and the client experience. He added that he would like advisors to be able to prepare for client meetings by using AI to summarize information about the family from the bank’s CRM system, topics of recent email correspondence and any public records, as well as suggest topics to address.
“Banks have a tremendous amount of data, but we haven’t been great utilizing it,” King says. “AI should give us a lot more insight into our client base, but that’s going to require accessing a broader range of information.”
Cox sees AI ultimately providing wealth advisors with a better understanding of customer data and customer events.
“It could help us understand money in motion, things that are changing and provide prompt notification for bankers and wealth advisors to better help their clients,” he says.
Contributing editor John Hintze is a financial journalist who writes frequently for the ABA Banking Journal.
