M&T Bank's Approach to AI Governance and Data Control

“You keep the human in the loop, because the human has to be accountable for the final product,” M&T Bank Chief Data Officer Andrew Foster told CDO Magazine in a recent podcast. The bank’s core stance as it scales generative AI is that productivity gains are welcome, but not at the expense of trust, governance, or control.

Foster joined M&T in 2023 and began by defining where the bank stood. His team used the EDM Council’s Data Management Capability Assessment Model (DCAM) to set the baseline and build a two-year strategy, extending through October 2025. Governance, quality, and lineage became foundational elements, later reinforced through additional cloud-based tooling and a federated model that reduced reconciliation work and cut analytics ticket time by an average of 50 days.

That groundwork has shaped how the bank approached generative AI.

Building Control Conditions Before Scaling AI

As generative AI accelerated across the industry, M&T chose a sequencing that ran counter to hype. “With banks and other well-regulated institutions, you must have your compliance and governance ducks in a row,” Foster said. The bank conducted a six-month proof of concept for Microsoft Copilot before deploying the technology across the organization.

By early 2025, M&T had rolled out Microsoft 365 Copilot and Copilot Chat to approximately 17,000 employees.

Productivity use cases drove early adoption. Employees use Copilot to draft internal communications, summarize documents, prepare meeting notes, and assist with coding tasks. Foster describes these efforts as part of a broader attempt to “return time to the employee.”

Other regional banks are moving along similar lines. KeyBank CIO Amy Brady said, “I think in our industry, where trust is so important, you have lots of opportunities to earn trust – you can lose it very quickly.”

KeyBank, like M&T, is experimenting primarily in middle- and back-office functions while training large segments of its workforce on generative AI foundations. Fifth Third Bank has emphasized culture, empathy, and governance as it builds AI capability.

The pattern across these institutions reinforces Foster’s point: the early value of AI lies in internal efficiencies, not automation of customer decisions.

Applying AI Where Expertise Already Exists

M&T is advancing targeted AI deployments in commercial credit and customer data.

The bank is integrating Rich Data Co.’s AI decisioning platform through vendor nCino, marking RDC’s first U.S. bank deal. The system provides early warning indicators, automated risk signals, and portfolio-level insights for commercial lending teams. It augments credit officers.

M&T is also adopting Amperity’s customer data cloud to unify interactions across channels and deliver more relevant communications. “Amperity will enable us to further consolidate customer interactions into a unified view of the customer,” said Kim Nupp, Director of Customer 360 Management.

Internally, M&T continues to expand its Data Academy, which trained more than 1,000 employees in 2024. These programs support Foster’s view that talent determines whether AI tools translate into long-term value. “The intersection point is talent,” he said. “If you are talking about a new form of AI enablement, you need a leader who will drive that forward.”

Other banks echo the emphasis on skills. KeyBank required employees across operations, technology, and risk to complete generative AI training by the end of 2024.

A defining feature of M&T’s approach is its methodical sequencing: assess data foundations, formalize governance, deploy internal AI tools, and layer in domain-specific applications only where mature oversight exists.

Foster described that discipline: “Everything we’re looking at is: How can we return time to the employee?” He evaluates tools based on their ability to help employees digest increasing volumes of information and focus on the decisions that matter.

Yet M&T is deliberately holding back from customer-facing AI. “Our focus is not going to the customer-facing side, because you need to learn and mature the technology,” Foster said. That stance places the bank among the cautious cohort of regional institutions, even as others pilot conversational agents externally.

Industry spending underscores the stakes. A World Economic Forum and Accenture report estimated financial services firms spent $35 billion on AI in 2023, with investments expected to reach $97 billion by 2027.

Foster’s response is to stay aligned with M&T’s identity as a community bank and to scale only where governance and talent support the effort.