M&T Bank is moving artificial intelligence out of the pilot phase and into everyday use across the organization, marking a major step in a technology strategy that has been reshaped for years. The bank said its expanded enterprise AI push is designed to make operations more efficient, give employees faster access to data, and help customers manage their finances in a period of rapid digital change. Rather than treating AI as a set of standalone experiments, M&T is integrating models into the same systems and workflows that support deposits, lending, payments, and the bank’s internal control infrastructure. Executives described the work as one of the largest technology initiatives in the bank’s recent history, with AI now embedded in decisions ranging from customer service to risk monitoring.

The expansion follows a period of concentrated technology investment at M&T. For years, the bank worked to replace older, fragmented systems and create a digital backbone that can support advanced capabilities. That multiyear overhaul involved consolidating data platforms, moving more workloads into the cloud, and designing common application services that can be reused across business lines. The underlying goal was to reduce complexity, quicken development cycles, and put data in a condition that would allow the bank to use newer forms of automation safely.

A technology foundation built for AI

M&T’s enterprise AI efforts did not begin with the much-publicized rise of generative AI. The bank has used machine learning in areas such as fraud prevention and credit modeling for years. What changed recently is the scale and scope of those tools. Across the industry, banks have discovered that pilot projects can be difficult to move into production without standard data controls, common platforms, and the ability to measure outcomes. M&T’s technology overhaul was intended to solve those problems before making AI expansion a strategic objective.

One of the central undertakings was the creation of a more flexible data environment. In older banking systems, data often sits in product-specific silos, making it difficult for a customer relationship manager to gain a full picture of a household’s relationship with the bank. M&T aimed to tear down many of those silos by updating its data governance and adopting cloud-native tools. Better data organization allows AI models to be trained and used in ways that reflect customer behavior and risk across different products, not just in isolated channels.

The cloud has also changed what is technically possible for the bank. Instead of waiting for large on-premises systems to send batch files, cloud-based applications can provide faster access to data and computing power. This is particularly important for AI workloads that need to process large amounts of information quickly. M&T has been selective about moving sensitive banking data, but it has steadily increased its use of cloud services in areas where controls can meet regulatory expectations. The result, according to technology executives, is an environment where AI can be developed with the same security and compliance requirements applied to traditional banking systems.

Beyond pilots: use cases across the bank

M&T’s expanded enterprise AI strategy is not centered on a single product. Instead, the bank is deploying a portfolio of capabilities that can be applied across operating divisions. Bank leaders say this approach avoids the risk of creating one high-profile AI tool that cannot connect with the rest of the organization. The use cases are selected based on whether they have a clear business outcome, sufficient data, and controls that allow humans to review the output.

• Fraud detection. M&T has long used machine learning to flag suspicious transactions. The next phase of enterprise AI builds on that expertise by allowing models to identify patterns in real time and adapt to changes in criminal behavior. The goal is to reduce false positives, lower the time spent reviewing alerts, and protect customers without slowing legitimate transactions.
• Customer service. Conversational assistants and intelligent routing tools are being used to handle routine requests, answer questions about accounts, and direct customers to the right team. Employees in contact centers see AI-generated summaries of prior conversations, enabling them to resolve issues faster without asking customers to repeat information.
• Document processing. Banking requires enormous volumes of document review, especially in commercial lending and account opening. M&T is using AI to read structured and unstructured data from loan applications, tax returns, financial statements, and legal contracts. The bank then routes that information into approval workflows with human oversight.
• Risk and compliance. AI tools scan transactions and communications for possible regulatory issues, monitor changes in the portfolio, and help model stress scenarios. The goal is not to automate risk decisions entirely but to give risk managers better signals and more time to investigate sensitive areas.
• Software engineering. The bank’s internal technology teams are using generative AI assistants to write, review, and document code. This is intended to reduce repetitive work, let engineers focus on complex problems, and accelerate the pace of feature releases.
• Operations and finance. Internal finance teams are using machine learning to improve forecasting, manage liquidity, and detect anomalies in expense and vendor data.

Each use case is evaluated in stages. A model may begin with a small group of users, then expand only if the outcomes meet expectations and the controls remain strong. M&T says the most successful applications are those where employees can see tangible value, such as shaving hours off a manual process or identifying a suspicious transaction more quickly.

Risk, governance, and responsible AI

As banks scale AI, risk management becomes more important than model development. M&T has worked to align its approach with evolving guidance from bank regulators. There are several layers of governance: model risk management, data governance, software development standards, and business-level control functions. The bank says no AI system is allowed to make a final decision about a customer without human review in areas that require judgment, such as credit denials or suspicious activity reports.

The governance structure begins long before a model is put into production. If business units propose an AI use case, they must describe the data being used, the potential for bias, what could go wrong, and the controls that will be in place. A centralized team reviews those proposals and ranks them based on risk. Low-risk automation can move more quickly, while higher-risk applications require more rigorous testing and sign-off from senior risk committees.

One challenge is managing third-party models embedded in vendors’ products. M&T increasingly relies on AI capabilities offered by software providers, but lenders remain responsible for how those tools affect customers. The bank has added due-diligence questions to its vendor management process, including how a model was trained, how frequently it is updated, and whether a vendor will permit independent testing. This transparency is especially important as banks deploy AI in areas where consumer protections apply.

Responsible use of data is also central to M&T’s expansion. Many AI techniques depend on large data sets, but banking data is subject to privacy laws and customer expectations. The bank says it continues to separate beneficial uses of customer data from uses that could harm trust. For example, AI can improve fraud detection by analyzing transaction history, but it must not be used in ways that would discriminate against protected groups. The bank has created model documentation standards and fairness reviews to test for unintended outcomes.

People and operating model changes

Scaling AI is not just an engineering challenge; it requires new skills across the bank. Employees are being trained to use AI tools, understand their limitations, and question model outputs. The bank’s technology teams have hired data scientists, machine learning engineers, and AI product managers, but it is also retraining existing employees who know banking processes. This combination of technical and banking expertise is critical for selecting the right problems to solve.

M&T’s operating model has also changed to speed delivery. In the past, a technology project might require a long requirements phase, followed by a development phase and a separate launch. Enterprise AI asks teams to work in smaller increments, test quickly, and involve business users throughout. That shift has led M&T to organize around product groups rather than technology silos. Each product group includes business leaders, designers, data scientists, and engineers accountable for a specific outcome.

Change management is particularly important because employees may be skeptical of AI. Bank executives say the strongest adoption has occurred where the bank focuses on reducing friction for workers. A banker faced with a cumbersome document review process is more likely to welcome an AI assistant that drafts a summary and lists the files needed for approval. When employees see that automation is making their jobs less repetitive rather than replacing them, they are more willing to help improve the data and controls underlying the system.

The road ahead for enterprise AI

M&T’s enterprise AI expansion is open-ended. The bank expects the capabilities to evolve as models improve, customer expectations change, and regulators provide clearer guidance. Technology leaders at the bank are not treating AI as a project with a final completion date, but as a core part of how the institution will operate from this point forward.

In the near term, M&T plans to increase the number of production-ready applications and connect more systems to the enterprise AI platform. That will require more data quality work, ongoing risk validation, and training for employees who are not yet comfortable with AI. The bank is likely to pay close attention to costs as well, because running large language models can be expensive, especially in an industry with thin margins and heavy regulatory oversight.

The expansion comes at a moment when banks are being forced to choose whether to lead with their own innovations or rely primarily on vendors. M&T has taken a hybrid approach, building some models internally while working with commercial AI providers for tools that require scale and deep technology investment. The bank says the deciding factor is whether the AI application involves a competitive advantage. Areas that shape customer relationships, such as financial wellness, tend to warrant internal work, while commodity functions may be better served by a vendor.

For customers, the benefits of enterprise AI should be visible in faster digital experiences, fewer alerts that require verification, and more accurate responses when they contact the bank. For the bank, the payoff is expected to be lower operational costs, better mitigation of financial crime, and a stronger ability to spot problems in the loan portfolio before they become large losses. Those benefits may not appear in a single quarter, but they underpin a long-term strategy that has been under construction since the earliest days of the bank’s technology overhaul.

Source:AI News News