Citizens Bank is seeing measurable productivity gains from generative AI across its engineering organization, with top-performing teams seeing the biggest returns. 

“While AI adoption is expanding across our engineering organization, our early results are promising, indicating productivity gains around 23% to 24%, with top adopters seeing over 30%,” Michael Ruttledge, chief information officer and head of enterprise technology and security, told FinAi News.

The $218 billion bank measures AI’s impact not by the percentage of code it generates, but by noting how it strategically enhances developer productivity, accelerates speed to market and improves engineering outcomes, he said. 

Citizens defines top adopters using a combination of adoption, contribution, and productivity indicators, rather than a single metric like token consumption, he said.

The bank had 17,398 employees at the end of 2025, up almost 0.7% year over year, according to its 10K filings. Ruttledge declined to say what percentage of coders are top performers. 

“As adoption grows and teams become more proficient, we expect these gains to increase through yearend, especially in early development stages where AI’s impact is greatest,” he said. 

AI is concentrated on lower-complexity, repetitive work, he said. 

AI is used for “generating boilerplate [reusable] code, assisting in testing and speeding up documentation, freeing engineers to focus on high-value work like architecture, design, and tackling complex challenges,” Ruttledge said. 

The bank’s AI use spans the development lifecycle, including architecture support, but within strict oversight, he said. 

“Every AI-assisted activity follows the same governance, security and architectural oversight as traditional development, with mandatory human review before code is released to production,” Ruttledge said. 

Picking a model 

Rather than building a large language model, Citizens hosts and governs existing models within its own cloud environment, he said. 

That governance-first approach extends to how the bank selects models to deploy, Ruttledge said, adding that it uses Github Copilot and Anthropic’s Claude for software development. 

“Citizens evaluates large language models based on their alignment with our rigorous security, governance and engineering standards, not brand preference,” Ruttledge said. 

Developers don’t typically interact directly with consumer AI products, he said, adding that they access models including those from Anthropic and OpenAI through enterprise platforms. 

“Our developers primarily access advanced large language models through secure enterprise tools like GitHub Copilot and our internal AI platform, rather than interacting directly with standalone products such as ChatGPT or Claude,” Ruttledge said. That approach improves the security of Citizens’ data. 

He added that the model-agnostic stance lets teams pick the best tool for each task.  

“This strategy enables us to harness rapid innovation in the LLM space without locking into a single vendor,” Ruttledge said. 

Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas.