Canada's TD Bank, one of the five largest banks in North America, has revealed details of its first agentic AI tool. It's not powered by a typical GPT-style language model — the bank bet on its own Causal Foundation Model, a technology that models actual cause-and-effect relationships rather than mere correlation. The Layer 6 research center behind the development has already deployed over 60 AI solutions across banking processes and plans an expansion into New York.
Listen to this article:
Agentic AI in banking: more than just a smart chatbot
When people hear "AI in banking," most imagine a chatbot that answers balance inquiries. But agentic artificial intelligence is several orders of magnitude beyond that. It's a system capable of independently analyzing situations, anticipating client needs, and actively assisting employees — for example, by suggesting relevant solutions in real time during a client call. TD Bank has turned this concept into reality. Its AI chatbots are integrated across seven business divisions and serve as assistants for customer support staff. It's not about replacing people, but about a tool that speeds up and improves the accuracy of their work. For instance, Cassandra Aeichele, who works in customer support at TD, uses an AI tool that helps her serve clients faster and more easily — the system suggests responses in real time, retrieves relevant information, and shortens the time needed to handle requests.Layer 6: an AI research center inside a bank
TD Bank didn't bet on outsourcing. Back in 2018, it established its own AI center of excellence, Layer 6, located in Toronto's MaRS Discovery District — a tech hub adjacent to the University of Toronto. The center now employs hundreds of data scientists and researchers, and the bank plans to open a second branch in New York. The chief scientist at Layer 6 is Maksims Volkovs, co-founder of the center, who oversees the development of AI products for the new era of banking. Alongside him is a team led by Jesse Cresswell, who is responsible for the Trustworthy AI framework — the safety and ethical guardrails for all deployed models. Academic partnerships also play a key role. TD is a founding sponsor of the Vector Institute, an independent AI research center in Toronto. Over 500 TD employees have undergone training organized by the Vector Institute — from webinars to intensive bootcamps. Additional collaboration runs with the Fields Institute, which trains PhD and master's students in AI and cybersecurity. Over 60 students have gone through the program in its two years.TD AI Prism: a Causal Foundation Model that understands cause and effect
The core of TD's technology strategy is a model called TD AI Prism — a so-called Causal Foundation Model. To understand what makes it exceptional, we first need to distinguish two concepts: correlation and causality. Conventional predictive models (including large language models like GPT) look for statistical correlations — if two things frequently appear together in the data, the model assumes a connection. But correlation does not imply causation. The classic example: ice cream sales correlate with drowning deaths, but ice cream doesn't drown anyone — both are related to hot weather. The Causal Foundation Model goes a step further. It is trained on hypothetical "what would happen if" scenarios, so it learns to recognize genuine causal relationships between variables. In banking practice, this means the model can not only tell you that a client with a certain profile is likely to close their account, but also why — and what would happen if the bank intervened with a specific measure. Previously, experts had to manually study data and build causal models by hand. With TD AI Prism, you simply feed new data into the model and it immediately predicts causal effects. The result is less manual work, fewer resources, and faster decision-making — from personalized offers to predicting client churn. Technically, the model was created in collaboration between researchers from the Vector Institute, who brought deep theoretical knowledge of causal modeling, and scientists from Layer 6, who leveraged their experience in training large-scale foundation models.Trustworthy AI as a competitive advantage
When a bank deploys AI that influences decisions about clients, trustworthiness isn't optional — it's a requirement. TD Bank therefore built an internal Trustworthy AI framework under the leadership of Jesse Cresswell and his eight-person team. The framework ensures checks and balances at every step of AI development — from data collection through training to production deployment. Every model must pass an audit that verifies not only accuracy but also fairness and the absence of bias. In the context of the European AI Act, which from February 2025 introduces mandatory risk assessments for high-risk AI systems (which includes banking), this is a textbook example of how responsible regulation and corporate governance can go hand in hand.What Europe and the Czech Republic can take away from this
Czech banks are not falling behind in AI, but they are taking a different path. The Czech National Bank is building its own AI competence center — according to CzechCrunch, it has purchased powerful Nvidia chips and runs models from OpenAI, Mistral, and Alibaba on them for banking supervision purposes. Commercial banks like Česká spořitelna or Komerční banka are experimenting with AI chatbots and process automation, but developing their own foundation model at the level of TD AI Prism is not yet on the table. TD Bank's approach is inspiring in two ways. First, it shows that investment in proprietary AI research — not just purchasing off-the-shelf solutions — can give a bank a unique competitive advantage. Second, it demonstrates that responsible AI is not a brake on innovation, but its prerequisite. At a time when the EU AI Act sets clear rules of the game for banking AI, the emphasis on trustworthiness may well determine who succeeds in deploying AI and who gets left behind. For smaller Czech banks and fintechs, the TD model can be an interesting source of inspiration — even if they cannot afford hundreds of researchers, the principle of causal modeling and the emphasis on trustworthiness can be applied on a smaller scale. And that may be the key lesson to take away from this Canadian story.How does the Causal Foundation Model differ from a typical language model like ChatGPT?
ChatGPT and similar models look for statistical patterns in data — when they see words or phenomena that frequently co-occur, they assume a connection. The Causal Foundation Model goes further: it is trained on hypothetical "what would happen if" scenarios and models genuine causal relationships. In banking practice, this means not only predicting that a client will leave, but also saying why — and what would prevent it.
Is agentic AI at TD Bank available to ordinary customers, or only to employees?
TD Bank's agentic AI tools are primarily intended for employees — they assist customer support staff in serving clients. Ordinary customers don't interact with them directly, but they feel the effect: faster request handling, more accurate answers, and more personalized services. It's AI that works in the background, not a chatbot you'd chat with.
Does TD Bank have branches in the Czech Republic or Europe?
TD Bank operates primarily in Canada and the United States; it has no branches in Europe or the Czech Republic. Its approach to agentic AI is, however, relevant to the European banking sector — it shows how proprietary AI research can be combined with strict regulation, which is exactly the challenge facing European banks in the context of the EU AI Act.
