Inside Citizens’ plan to reimagine itself with AI

• Key insight: Citizens is reimagining core operations around generative and agentic AI. 
• Supporting data: The bank has moved 750 applications to the cloud, eliminated 25 data centers and cut infrastructure costs 10–15%. 
• Forward look: Expect Citizens’ agentic AI pilots to expand into fraud, complaints and quality control next year.

Source: Bullets generated by AI with editorial review

CEO Bruce Van Saun said in July that Citizens Financial Group is reimagining the bank with AI in mind. “The objective is to redesign how we serve customers and run the bank, taking advantage of new technologies like [generative] AI and agentic AI,” Van Saun said during a second-quarter analyst call.

At the center of much of the work, the details of which are still being worked out, is Chief Information Officer Michael Ruttledge, who recently gave American Banker an exclusive look behind the curtain.

“There’s still a lot of proof of concepts happening right now, and a lot of test-and-learn,” Ruttledge said during an interview at a Citizens office in New York City. “But the types of use cases we’ve seen are where there’s manual processes today that we think we can automate.”

Ruttledge’s team will execute on 47 initiatives, some using agentic AI, others simply using business process reengineering to make processes more efficient, to reduce operating expenses and improve revenue.

Ruttledge shared details about some of the areas where gen AI and agentic AI are starting to make a difference.

AI-assisted software development

About half of Citizens’ software engineers, 800 people, are using Github Copilot, and some of them are using Github Copilot coding agent, which generates code for them.

The bank is also working with a few fintechs to use generative AI to rewrite its older code. For example, they rewrote a legacy mortgage system to move it to the cloud in six weeks.

“That was millions of lines of code, they just completely rewrote it, and now it’s in the cloud,” Ruttledge said, explaining that the process has significantly reduced the bank’s software-as-a-service costs.

While there’s still a human reviewing the code, gen AI saves time and helps with documentation and testing, leading to a 10% to 15% productivity uplift for engineers, Ruttledge said.

The next step — having AI automatically generate pieces of code — will eventually lead to five times greater productivity, he said.

However, Ruttledge doesn’t see software development jobs going away.

“Coding is still only a fraction of the overall development cycle,” he said. “You’ve still got to write the business user stories, you’ve still got to understand the business function. You’ve still got to do the design, then you’ve got to write the code, build it, test it and integrate it with all the other platforms in the ecosystem. You’re still going to need engineers to do that.”

As AI is used to automatically generating code, “there are some engineers who will love that, and there are some engineers who won’t love that because they enjoyed actually doing the coding part themselves,” Ruttledge acknowledged.

“Are they going to feel like they’re losing something because now they’ve got to pivot to be more designers or supervisors?” he wondered.

Citizens has taken a number of steps to manage the impact of AI adoption on software engineers.

It has “released” some employees who work for vendors such as Infosys, Accenture, IBM and Cognizant, Ruttledge said. Some legacy coders who weren’t ready to move to agile development have also been let go.

The tech staff has also been upskilled through three-week “academies” focusing on software engineering, data engineering, AI engineering and IT architecture.

The bank also runs an AI Academy for employees throughout the company, where they learn how to use Microsoft Copilot.

Uses of AI at call centers and in underwriting

Citizens Bank uses AI in its contact centers to make suggestions to service representatives during calls. “If someone calls in and says, I want a credit card that has low interest, that gives me good rewards, it will listen to that and put on the screen a suggestion, they should use the Summit Reserve card,” he said. This implementation of AI is now used by more than half of Citizens’ customer service representatives.

Within Citizens’ gen AI portal, employees can use OpenAI, AWS Bedrock, Anthropic or another LLM provider. “This gives us the freedom to adopt different ones where we think someone’s got a leap on the market and we want to get there,” Ruttledge said.

His team built an AI-based lending onboarding platform. It’s been rolled out for student lending, and components of it are being used for home equity lending and will be used for credit cards.

Now, when a group at Citizens wants to use a new tool, such as a new type of credit score or cash flow underwriting, it will be easier to deploy.

The Providence, Rhode Island-based company has done something similar with account onboarding — developing a new platform that was used first to open new small-business accounts in a day, rather than 14 days. This was then rolled out in Citizens’ private bank and then to retail customers.

Citizens has also rolled out an AI virtual assistant called CiZi. “You voice or type in and it will tell you whatever you want to know: your balance, your last statement, your fee from a credit card,” Ruttledge said. “That’s reduced calls generated from the mobile app to the contact center by about 44%.”

Agentic AI for complaints, quality control

So far, Citizens is using agentic AI mostly in software engineering, but that is beginning to change.

“We’re starting to look at it in other use cases,” Ruttledge said. For example, fraud claims processing is workflow-oriented, so AI could help get to straight-through processing claims of small dollar amounts. Citizens has rewritten its fraud platforms so that they run in the cloud and use machine learning capabilities to help with suspicious activity reporting, security checks and next year for anti-money-laundering checks. Now it’s piloting agentic AI in this area.

“We haven’t got there yet, but we’re doing some pilots,” Ruttledge said.

Citizens is also using agentic AI to handle complaints. “As complaints come into the bank, we classify them, and then we can respond back to customers. So it’s easing some of that response time,” Ruttledge said.

And it’s looking at using agentic AI for quality control. For instance, employees today review 5% of customer service call transcripts for tone and accuracy. Ruttledge’s team has been piloting the use of generative AI to automate that process for 100% of calls.

“That’s going to be a significant breakthrough,” Ruttledge said. “We’ve got hundreds of these QC-type processes that we’re looking to automate.” People who used to do this work, or still do it today, are being redeployed to other tasks, he said.

Journey to AI: Cloud migration, software upgrades

To lay the foundation for the AI reimagining, Citizens has been moving applications to the cloud. So far, it’s moved 750 applications — 90% of all the software it uses — with only 11 applications remaining on premises.

“They will move in the next couple of weeks, and then we will be 100% in the cloud,” Ruttledge said.

The bank uses Equinix as a networking hub to connect to Amazon Web Services and Microsoft Azure. Those 11 applications that it couldn’t move into the cloud environment are being moved onto co-located servers at Equinix.

As a result of moving everything off-premises, Citizens eliminated 25 data centers over the past three years. This has lowered its infrastructure costs by 10% to 15%, Ruttledge estimates. There’s also salary savings because the bank doesn’t employ data center staff anymore. It’s also saved gas and electric costs by not having to power the data center buildings.

To get to this point, about 80% of applications had to be modernized, Ruttledge said.

“We’ve upgraded our versions of Oracle databases, our operating systems, our hardware in the cloud,” he said. “We’ve probably refactored about a third of the footprint. We totally rewrote it, so even where we’ve lifted and shifted, we’ve still updated the core operating system database. So now, as we move to the cloud, we’re over 98% current, which is a huge improvement.”

The bank moved its mainframe to zCloud, Kyndryl’s system for running IBM Z workloads.

The bank is upgrading its core to FIS’ modern banking platform. “We’re on version 3.11 of that, which is probably the most up-to-date version, and all of our digital bank and deposit products are on there. And then we’re working with FIS to develop checking facilities, and then we’ll start migrating product by product off the mainframe and onto the modern core.”

Time to market is the main difference between old core and new, Ruttledge said. The code is scanned for cybersecurity issues before it goes into production, “so our security teams are no longer a blocker,” Ruttledge said.

“What used to take us more than 48 hours, we’re now able to do in 15 minutes,” he said.

Citizens’ plan to reimagine the bank via AI seems like the right approach, according to Jacqueline Rinehart, artificial intelligence and transformation officer at the AI advisory firm Limitless.

“They are not chasing pilots or bolting on tools,” Rinehart told American Banker. “It appears they are pairing AI with real structural modernization and long-term transformation. Their strategy is a strong example of smart innovation that keeps true to the principles of a banking experience that is rooted in structure, governance, security, regulation, trust and humans in the loop. It also aligns to the deliberate catch-ups of its peers.”