How AI is Permeating Work Throughout TD Bank, and Lessons Learned

By Steve Cocheo, Senior Executive Editor at The Financial Brand

Published on February 20th, 2026 in Banking Technology

Simple Subscribe

Subscribe Now!

TD Bank Group has been investing heavily in multiple forms of artificial intelligence for years. Milestones include its purchase of a leading AI developer, Layer 6, in 2018 and the mid-2025 launch of TD AI Prism, an AI foundation model designed to ramp up the bank’s ability to predict customer needs and personalize their experiences.

TD has roughly 2,500 people working on AI development, including engineers, scientists, data analysts and other experts. This includes a New York City center for the Canada-based Layer 6 operation. But that’s only part of the picture.

Earlier forms of AI have been used for some time at TD for functions like credit analysis and siting branches (“stores” in TD’s parlance), but applications for some of the newest forms of AI, GenAI and agentic AI, are still emerging.

Need to Know:

During the company’s 2025 fall investor day, Raymond Chun, group president and CEO, said TD was targeting $1 billion in annual value from ongoing, increasing application of AI technology.
Chun said that half of that value would come from cost savings and half would come from improving revenue.
In an investor presentation in January, Chun spoke of 2026 as the year of agentic AI, which would be spread through many parts of the bank’s operations.

Lessons from TD’s AI Strategy

In the company’s operation in this country, TD Bank U.S., Ted Paris, SVP and head of analytics, intelligence and AI, heads the group that connects users in U.S. consumer banking, wealth management, and commercial banking with the tools the bank is supplying and developing.

Key insight: AI demands an ecosystem. Paris says that success with AI requires a continuum of effort, drawing not only on developmental talent.

Management of the business side, to guide adoption and application where tech meets task, is one essential element, he says. Another is the multiple types of protection that must precede adoption, including evaluation of model risk, assurance of compliance and fair banking, maintenance of customer data privacy, and keeping bank data insulated from external exposure.

Bankers are concerned about all of this, but the regulators are also watching, because it’s all new to them as well.

Another key insight: “The most important thing that we’re all collectively solving for, that we all have a mutual interest in, is the trust factor,” says Paris.

-- Article continued below --

Maintaining Guardrails to Maintain Trust

The recent rapid expansion of AI from its historical, arcane roots to something anyone with an internet connection can use raises the stakes for trust and control. “Our people can’t just download any model they want” onto company machines and systems, says Paris. The bank maintains guardrails to be sure that unauthorized AI can’t be turned loose on its systems and records.

“Our people are not having conversations with clients based on what ChatGPT says,” says Paris. “We might leverage such foundation models, but we build them for what are solutions are designed to do.”

Paris adds that models are built such that queries out of character with the purpose of the model can’t be made. So, for example, the bank’s model to assist contact staff to come up with instantaneous answers to customer questions — including citations to sources in bank policies and procedures — can’t be bent to some other purpose.

Staying on the same page. The bank disseminates a shared view of what is appropriate for AI applications and how it can be applied, Paris says. That drives principles that address AI in its various forms, including GenAI and agentic AI.

A critical element for AI applications seen through much of the bank’s usage is the “human in the loop” concept, according to Paris. The idea is that in anything that touches the customer, there is a human staffer between them and the AI.

TD’s Blueprint for Building Trust in AI

Trust is a consideration in the AI building process, whether the bank is adapting existing models or crafting unique solutions for TD Bank’s needs.

Key point on AI trust: A Layer 6 blog entry of late last year addressed this issue at length: “It is incredibly important that these tools are built to be aligned with human values, which is the overall goal of the field of research known as Trustworthy AI (TAI).”

Key tenets of trustworthy AI design, according to the blog, include: fairness, privacy, robustness — the ability of a model to maintain performance and not degrade over time; explainability — understanding and validating decisions made by models, rather than just trusting the “black box,” and uncertainty quantification — determining the range of confidence that a model justifies, and deciding when the model should be ignored.

Key insight: The art in the science of AI design, the blog goes on to say, is that measures aiming to achieve those five tenets have to be “harmonized,” to achieve a balance.

Why? Because sometimes they can actually fight each other. An example from the blog: “Applying training methods to improve privacy can amplify biases in the data as a side effect, undermining fairness. While not always acknowledged, these unintended interactions are very common.”

The blog explores solutions for building trust during design in some depth, which can be read here.

Read more:

Tapping AI’s Transformative Properties from the Top Down

“Generative AI and agentic AI are about transforming our operations, making them simpler and faster,” says Paris. “Hopefully, we can evolve the way we support our colleagues, so TD employees can gain greater capacity, evolve their roles and the way they deliver services, and improve the experiences of our customers — and ultimately deepen the relationships with them.”

As TD Bank U.S. has pushed forward into this, Paris says key steps have been taken to bring employees along at multiple levels.

For example, in cooperation with Columbia University, the bank has provided a series of executive training programs for close to 150 senior executives, as well as some executives at the next level, according to Paris.

AI in a nutshell. “It’s to give them a sense of what the technology is, what it’s about, what its potential is, how it’s evolving, and what it means for them,” says Paris. Time is also devoted to explaining the AI ecosystem the bank has already built out and what is coming and where in the bank it will be coming first.

Then the program turns things around. It challenges attendees to suggest what opportunities they see for AI implementation for their own areas.

-- Article continued below --

Spreading AI Application Around TD

Further down the hierarchy, the bank has been providing co-pilot and personal productivity AI tools in many areas, as well as training in how to frame prompts for them.

Two additional areas the bank has been pursuing for employees:

Content generation tools, for producing reports and analyses from huge bodies of bank data, as well as marketing content.
Knowledge management tools, for functions such as branches and contact centers.

“This greatly accelerates the learning curve and reduces the amount of times that someone has to escalate a query,” says Paris. “And it creates greater consistency across thousands of contact people.”

What’s coming: Such tools are “democratizing data” at TD Bank, says Paris, and spreading the ability to analyze and draw on qualitative data — concepts — instead of only quantitative data — words and numbers.

The bottom line. “We’re changing the way work is getting done,” says Paris. “And we’re just at the nascent stages of what GenAI and agentic AI are going to allow us to do.”