Citi to move risk models, workforce and customer apps to Google Cloud

Citigroup has begun moving some software applications from its data centers to Google Cloud and is experimenting with Google’s AI technology through a multiyear agreement.

The terms of the agreement were not disclosed.

The bank plans to start with enterprise analytics, high performance computing for its markets business, desktop solutions for employees including customer service agents, and customer-facing apps. 

The move is part of a broader modernization effort that Citi CEO Jane Fraser mentioned in several earnings calls. It’s also part of an industry-wide shift toward cloud computing that has some regulators worried.

“We’re focused on simplifying our operational model, modernizing our infrastructure risk and controls and all of that reduces risk as we go,” Fraser said in the third-quarter earnings call. “We’re well on the way in executing the transformation plans.” 

Citi is the latest large bank to announce a major move to cloud computing. In 2020, Capital One closed all its data centers and moved everything to Amazon Web Services. In 2021, JPMorgan Chase said it would use a cloud-based core banking system from Thought Machine for its retail bank, and Wells Fargo said it would migrate several applications to Microsoft Azure and Google Cloud. The next year, KeyBank said it would put primary applications in Google Cloud and U.S. Bank planned to move most applications to Microsoft Azure. 

“Big banks need to modernize their platforms and architectures with support for advanced data analytics and AI,” said Sumeet Chabria, CEO of consultancy ThoughtLinks Group and former chief operating officer of global tech and ops at Bank of America. “Cloud platforms provide optimal flexibility and scalability for the data-intensive workloads required to fully leverage AI. They support quick deployment of applications and offer an elastic cost structure with the ability to scale up and down as needed.” Most banks will keep their most sensitive data on premise, he said. 

Citi positions its cloud initiative as a way to support its global infrastructure.

“Citi does business in 90-plus countries,” Balaji Kumar, the bank’s CIO of global technology infrastructure, said in an interview. “For us to be able to do business and provide customer-facing products and services within the marketplace, we need partners that can also work at the same global scale that Citi does, understand local compliance and regulatory laws, understand the security first, controls first mindset, and also be able to deploy infrastructure that will allow us to go to market fast.”

Citi will use Google’s Vertex AI, a platform that lets users create, train and deploy artificial intelligence models, to develop and roll out generative AI models across the enterprise. Vertex AI can work with several large language models, including Google’s Gemini. 

Citi also plans to use Google’s Agent Assist, a voice-activated AI assistant, to help customer-service agents navigate knowledge bases and provide relevant answers as quickly as possible. 

“We have quite a few exciting use cases, specifically for Agent Assist, that we are working on that we will bring to market pretty soon here,” Kumar said. 

Customer-service representatives using Agent Assist could onboard new customers more quickly or find human resources policies more easily, according to Rohit Bhat, general manager and managing director, Google Cloud for financial services. 

At the other end of the spectrum, Citi plans to use Google Cloud in its markets business, for risk calculations and other tasks that require high-performance computing.

High-performance computing technology Google Cloud supports includes Nvidia graphics processor units and Google’s own tensor processing units, chips designed to accelerate machine learning workloads. 

In addition, Citi and Google are developing a virtualized environment for employees. 

“We are working heavily with Google to integrate and see, how does a virtualized desktop environment work within the Google Cloud platform, and how do we enable access globally?” Kumar said. “Once you have the access to a desktop virtualized environment where you have your applications, then you go into use cases based on personas.” One persona might be a customer-service agent helping a customer deal with credit card fraud, he said. 

“How do you look at that use case and layer an AI engine on top of it, and then say, what are some of the solutions that we can bring to life for that agent to help that customer?” Kumar said. 

In a separate project, Citi’s developers continue to use GitHub Copilot to speed up their work. 

The Treasury Department came out with a report last year that raised many concerns about the increasing use of cloud computing in financial services. Among them were issues around security, resilience, incident disclosure and response, concentration risk (too many banks relying on a small number of vendors, like Microsoft, Google, Amazon and IBM) and imbalances of power between bank users and large tech company providers.

Bhat said Google worked on security and resilience.

“Google has spent a lot of time on control mechanisms,” Bhat said in an interview. It’s also worked to make sure the information generated by any usage of cloud is auditable and trackable. “So all their environments and any services that they use, they have full control and protections around,” he said. 

Google has localized its assets to make sure global customers can adhere to local principles and rules. It’s also helping clients avoid vendor lock in, Bhat said. For instance, it’s made some technology assets like BigQuery and Kubernetes available on cloud environments other than Google’s. 

“BigQuery runs in Amazon and Azure,” he said. “Kubernetes is available now in all those platforms as well. So we’re taking our technologies elsewhere to make sure that you have optionality.” 

It’s also developed open source versions of some of its AI technology, such as Gemini. 

Citi’s Google Cloud agreement is an extension of its standards-based hybrid cloud strategy, Kumar said. 

“Even within our on-prem data centers, we have also made sure that the focus is always on adopting standards, such as container-based standards,” he said. For example, the bank uses Kubernetes-based workloads. 

This is a best practice — banks should implement cloud-agnostic architectures even if they begin with one hyperscaler, ThoughtLinks’ Chabria said. 

“This approach reduces concentration and vendor lock-in risk and enhances resilience by allowing workloads to be shifted across providers, although more industry collaboration is needed to develop standards that improve interoperability,” he said.