M&T Bank didn’t rush into generative AI. The Buffalo-based regional bank initially blocked employee access to public large language models, worried that someone would paste customer data into ChatGPT. Then Chief Data Officer Andrew Foster evaluated enterprise providers, picked Microsoft Copilot, and ran a pilot with about 800 employees before flipping the switch for roughly 16,000 of the bank’s 22,000 workers.
The scale is notable for a regional player. Employees now use Copilot to draft emails and reports, while the bank leans on AI to summarize call-center conversations — a task Foster says saves about six minutes per call. Developers use GitLab tools to generate code, and the bank is already probing agentic AI for cybersecurity and fraud detection. But M&T’s approach isn’t hands-off. Its 2026 Code of Business Conduct and Ethics requires employees to review AI-generated work, and prohibits confidential information from entering unapproved systems.
None of this would work without plumbing. M&T’s AI push sits on top of a technology overhaul that started in 2018, when more than half its tech specialists were outside contractors. Today, 80% of the workforce is in-house, spanning roughly 2,000 technologists across 300-plus agile teams. Technology spending hit $1.2 billion in 2025 — nearly triple 2017 levels — and outages have dropped more than 80% since 2018.
Foster also built a data-lineage program using Solidatus and Monte Carlo to trace how information moves through the bank’s systems. He told American Banker that work wasn’t driven by generative AI, but it’s become the foundation for it. M&T now uses retrieval-augmented generation with internal, governed data, and has an internal repository called Edison for authoritative policy documents. A Data Academy has drawn about 2,000 employees into data governance training. The bet is that AI built on governed data beats AI bolted onto chaos.
