M&T Bank Enterprise AI Reaches 15,000 Employees After Decade of Tech Rebuild

M&T Bank's enterprise AI deployment has crossed a major milestone. AI copilots are now active for more than 15,000 employees across the regional lender's operations, touching internal workflows, customer service channels, software engineering, and risk management in a rollout that spans nearly every corner of the institution. That's a huge number. And it places this effort among the largest AI adoptions in US regional banking, though it's not the only one out there.

The bank is applying the technology to analyze call-center conversations, draft internal reports, generate computer code, spot customer needs, and flag potential portfolio risks, all while weaving the capability into daily operations. That's a big shift. So the strategy has clearly matured, beginning as a cautious pilot with roughly 800 employees before it expanded organization-wide, and now it's touching nearly every corner of the institution. Don't mistake this for a flashy experiment. It's practical, grounded work.

From Lockdown to Large-Scale Rollout

M&T's path to widespread AI adoption was not immediate. Before committing to enterprise tools, the bank deliberately restricted employee access to public large language models. Chief data officer Andrew Foster explained the reasoning at the time: employees might accidentally type sensitive company data into public-facing tools that lacked governance safeguards.

That hesitation didn't last. It gave way to a structured evaluation of enterprise providers, with the bank weighing options carefully before settling on a single vendor. So they chose Microsoft Copilot. The bank ran a controlled pilot with about 800 employees first, then broadened access only after seeing how the tool performed in real-world conditions. The measured approach paid off. It's delivered measurable efficiency gains, and those gains aren't just anecdotal, they're baked into the bank's internal metrics now.

According to Foster, generative AI cuts about six minutes from each call-center conversation by condensing the chatter, and across thousands of daily interactions, that small per-call reduction adds up to big operational savings that executives can't afford to ignore. But it's a math game. So the real payoff isn't in one chat, it's in the sheer volume, where every saved minute stacks into hours, days, even weeks of reclaimed labor. Six minutes doesn't sound like much. Across a month of calls, though, those minutes become a serious chunk of the budget.

Human Oversight Remains Non-Negotiable

Software developers at M&T use GitLab tools to generate code, but the bank has drawn a firm line on accountability. Employees remain responsible for reviewing all AI-generated output.

That human-review requirement is now codified in the bank's 2026 Code of Business Conduct and Ethics. And it doesn't mess around. The policy mandates that employees use only approved AI tools, while explicitly prohibiting them from entering confidential, proprietary, customer, employee, or regulated information into unapproved systems, so the guardrails are clear from the start. Workers bear final responsibility for the accuracy and appropriateness of any AI-assisted work they submit. That's on them.

Employees remain responsible for the accuracy and appropriateness of AI-assisted work, a principle now embedded in the bank's formal ethics code.

The Long Road to AI Readiness

M&T Bank's enterprise AI ambitions rest on a foundation that took years to construct, with the real work starting back in 2018 when more than half of the bank's technology specialists were outside contractors. That's a hard truth. Now, the workforce is 80 percent in-house, and that shift didn't happen overnight. But it's the kind of change that defines what they can build next. So the tech team they've got today isn't the same one they had before.

The numbers tell the story of that transformation:

• M&T now employs about 2,000 technologists across more than 300 agile teams
• More than 1,000 technology specialists have been hired during the program
• Technology outages have dropped by over 80 percent since 2018
• Annual system upgrades have increased by 300 percent
• Technology spending exceeded $1.2 billion in 2025, nearly triple the 2017 level

Michael Wisler, who joined M&T as chief information officer in 2018 and became senior executive vice president for technology and operations in 2025, told Forbes that annual technology releases grew from about 15,000 in 2018 to 65,000 in 2025. His current role spans both technology and operational functions across the bank.

Data Governance Came First

Foster, who arrived at M&T in 2023, started building a data-lineage program to track where information originates, how it is used, and how it moves between systems. He emphasized that this work was not created in response to generative AI. Rather, he described it as a core capability for understanding the bank's data estate.

The bank established a Data Academy focused on data governance and skills. Around 2,000 employees took part. But M&T didn't stop there; they also built an internal repository called Edison, which now holds the bank's authoritative documents and official policies in one searchable place. Solidatus and Monte Carlo provide data-lineage software, and that technology traces information as it flows through databases, applications, and business-intelligence systems, so nothing gets lost in the shuffle. It's a lot of moving parts.

That infrastructure hands M&T a clear view of each data element's source, meaning, quality, and governance. But it's the practical payoff that matters. Foster cited a real-world example: the bank's Copilot deployment uses retrieval-augmented generation that pulls from internal, vetted information, so employees don't get answers from unvetted sources. Short and sharp.

Three Pathways to Enterprise AI

Wisler outlined a three-pronged strategy for generative AI adoption. The first route covers general employee use through tools like Copilot, while the second identifies AI capabilities already embedded within the bank's more than 1,800 third-party applications. And the third involves proprietary AI systems built around M&T's own data and processes. It's a layered approach. But each prong targets a distinct layer of the organization's tech stack, from everyday workers to the deepest, most custom-built infrastructure.

Early proprietary applications target repetitive operational work, software development, fraud prevention, and cyber defense. But that's not the whole story. The bank continues to assess both internally developed systems and external tools, including general enterprise software and technology designed specifically for financial institutions, so they can weigh what works best against what's already in place. It's a careful process. They don't rush.

Important. The earliest use cases focused on drafting, summarization, call-center work, and coding. The newer applications are more sophisticated, identifying customer needs and flagging portfolio risks before they become problems.

Agentic AI on the Horizon

M&T is also examining agentic AI applications in cybersecurity and fraud detection.

M&T is not alone in this push.

JPMorganChase launched its internal LLM Suite platform to more than 200,000 employees in 2024. By 2025, over 65,000 employees in its Corporate and Investment Bank were actively using the platform, while more than 90 percent of its engineers used AI coding assistants. The bank's AI-based transaction screening reviewed more than twice the previous transaction volume while cutting manual operator checks in half.

Bank of America has deployed a generative AI system called EricaAssist with more than 18,000 customer service employees, and the tool summarizes why a customer is calling, retrieves relevant information, and recommends next steps while keeping the employee responsible for the interaction. It's fast. In July 2026, Bank of America said EricaAssist delivers contextual guidance in under three seconds, and it has cut average call times by nearly one minute, a clear operational win for the bank's call centers. So the bank plans to extend the system to additional servicing scenarios later this year. But they can't stop there.

M&T's approach reflects a broader industry recognition that enterprise AI success depends less on the model itself and more on the data infrastructure, governance policies, and human workflows surrounding it. They spent years building that foundation. But the payoff didn't come overnight. Only then did they scale AI across their workforce, and the result is a deployment that touches thousands of employees daily while maintaining strict oversight over what AI can and can't do with sensitive financial information. That's the real lesson.

Frequently Asked Questions

What is the current scale of M&T Bank's enterprise AI deployment?

M&T Bank's enterprise AI deployment has crossed a major milestone with AI copilots now active for more than 15,000 employees. This rollout spans internal workflows, customer service channels, software engineering, and risk management, touching nearly every corner of the institution.

Why did M&T Bank initially restrict employee access to public large language models?

M&T Bank deliberately restricted access to public large language models before committing to enterprise tools because employees might accidentally type sensitive company data into public-facing tools that lacked governance safeguards. This hesitation gave way to a structured evaluation of enterprise providers, leading them to choose Microsoft Copilot.

How does M&T Bank ensure human oversight in AI-assisted work?

Employees at M&T Bank remain responsible for reviewing all AI-generated output, and this human-review requirement is codified in the bank's 2026 Code of Business Conduct and Ethics. The policy mandates using only approved AI tools and prohibits entering confidential or regulated information into unapproved systems, with workers bearing final responsibility for accuracy and appropriateness.

What was a key outcome of M&T Bank's data governance initiatives?

M&T Bank built a data-lineage program to track data origins and movement, and established a Data Academy that about 2,000 employees participated in. They also created an internal repository called Edison for authoritative documents, and use software like Solidatus and Monte Carlo for data-lineage tracing, providing a clear view of each data element's source and governance.

Who are some other banks mentioned in the article as also deploying enterprise AI, and what are their achievements?

JPMorganChase launched its internal LLM Suite platform to over 200,000 employees in 2024, with more than 65,000 in its Corporate and Investment Bank actively using it by 2025. Bank of America deployed EricaAssist to over 18,000 customer service employees, delivering contextual guidance in under three seconds and cutting average call times by nearly one minute.