One year before TD acquired Layer 6 in 2018—then an artificial intelligence (AI) startup and now a leading AI research and development centre—the Bank was already taking steps to build an AI ecosystem.
AI tools had yet to become ubiquitous, but the Bank was looking ahead to innovate and anticipate future client and colleague needs.
TD became a founding sponsor of the Vector Institute, an independent non-profit AI research facility, in 2017. Since then, the Bank has strategically developed a network made up of AI start-ups, research institutions, universities and more that are giving it insight into the next generation of talent, research and new technologies.
"We can't do it all alone," said Maksims Volkovs, chief AI scientist at TD. "We've developed a broad network of relationships so that we can keep our fingers on the pulse of innovation, identify promising, young talent and develop new solutions to power our AI strategy going forward."
Here are five institutions that TD is working with to advance AI.
Vector Institute
Vector was founded by Toronto-based artificial intelligence (AI) entrepreneurs Jordan Jacobs and Tomi Poutanen (who co-founded Layer 6, alongside Volkovs) and Geoffrey Hinton, a University of Toronto professor sometimes referred to as the "godfather of AI."
The organization supports the adoption of AI across Canada, helps develop new Canadian talent, and conducts new research into the field. Since its launch, Vector and TD have worked on developing and sustaining responsible AI innovation to support the growing AI industry in Canada
Together, they have run industry-focused AI training programs on topics such as "explainability in larger language models (LLMs)" and "practical AI governance," with more than 500 TD colleagues attending.
They have also worked on applied AI research and projects for the Bank. Volkovs said working with Vector on AI research and solutions has allowed Layer 6 to accelerate the development of AI use cases at TD.
Their joint research led to the development of the Bank’s first causal foundational AI model, a large-scale model that's trained on hypothetical what-if scenarios. Using this data, the model can help TD predict the outcomes of making certain business decisions such as those that deal with pricing.
Before developing the causal foundation model, Layer 6 scientists would have to build a separate model for every product, a process that took multiple months.
"Each use case was a resource-intensive and time-consuming engagement," Volkovs said. Now, the causal foundation model can simultaneously predict variables in the Bank's products, significantly accelerating time to market.
Building relationships with research institutes like Vector are important to the Bank's AI strategy, Volkovs said. TD also recently began working with Mila, an AI research institute in Quebec. TD will work with Mila professors on AI research projects, gain access to its student network for internship and co-op opportunities and participate in training programs.
Radical Ventures
TD continued to build out its AI ecosystem in 2019 when it invested in Radical Ventures, a venture capital firm founded by Jordan Jacobs.
Radical invests in early-stage companies and entrepreneurs that are applying deep tech, such as AI, to transform their industries. Their goal is to identify these companies early enough that they can help support their development and help build — and maintain — a homegrown AI sector in Canada.
Investing in Radical Ventures helps TD play a role in supporting the next generation of AI talent in Canada.
Signal 1
In 2021, TD invested in Signal 1, an AI health start-up company that's researching and developing solutions to integrate AI into hospital workflows.
Founded by Tomi Poutanen, Signal 1 is introducing AI into hospital workflows to improve patient outcomes, enhance hospital flow and reduce stress on staff. The company's first solution, CHARTWatch, uses machine learning to predict whether hospital patients will recover, remain stable or deteriorate. Signal 1 uses Layer 6 technology to enhance and maintain CHARTWatch.
"We essentially found that the same techniques that work for client and bank-facing data translate to healthcare because they have similar patterns," Volkovs said. "At a bank, it's transactions and account changes. In healthcare, it's hospital visits and lab results. Even though the data sounds different, it's structured very similarly."
As part of the agreement, TD made Layer 6's leading AI solutions available to Signal 1 to enhance and maintain CHARTWatch.
The general internal medicine unit at St. Michael's Hospital in Toronto, an early adopter of the technology, has seen unexpected deaths decline by 26% during a 19-month period between 2020 and 2022, according to the Canadian Medical Association Journal.
The collaboration with Signal 1 allows TD to use its cutting-edge technology to make a difference in its communities, Volkovs said. But it also gives TD the opportunity to learn from Signal 1's results and vice-versa, which can help Layer 6 further develop its AI solutions at the Bank.
Cohere
Through the Radical Ventures Fund, TD invested in Toronto-based AI startup Cohere in 2024 to test the company's LLMs and determine whether they could help the Bank refine its generative AI use cases.
Volkovs first met Cohere CEO Aidan Gomez in 2019. He remembers the promise he saw in Cohere's ideas, even at an early stage in the company's formation. TD and Cohere would cross paths again when they collaborated at the 2020 ACM RecSys Challenge, a global AI competition where leading researchers and technology companies have the opportunity to benchmark their AI capabilities against the world’s best. Volkovs was impressed by the company's advancements in such a short time.
When TD began working with Cohere in 2024, it did so because the AI startup had become a leader in language models and in training them. The agreement with Cohere also gave TD access to medium-sized language models, which are more cost-effective and quicker than LLMs. These models, unlike LLMs, can also be efficiently finetuned, Volkovs said. Over the past year, TD has been studying whether finetuning these models can make them as accurate as LLMs to see if there might be an application for them at the Bank.
TD has also been testing Cohere's retrieval augmentation generation capabilities as part of their collaboration. These capabilities ground a language model, like a chatbot, to a particular dataset, such as a series of policies and procedures. This means that a chatbot can't source its answers from anywhere outside a particular dataset, which enables a model to be more accurate and can help to limit its potential to hallucinate.
Columbia Business School
TD has relationships with multiple Canadian academic institutions that help it identify and assess the next generation of AI talent. This year, the Bank also began to seek out these collaborations in the U.S. and started working with the Columbia Business School in New York.
The relationship with Columbia, Volkovs said, allows TD to connect to promising AI talent who are completing their studies at the university. TD brings in students for co-op programs and internships, where they have the opportunity to participate in months-long research projects. Students are given experience on the front lines of an AI research lab and TD is given the opportunity to monitor their work and see if some might consider a career at the Bank.
Layer 6 hires about 30% of its talent straight out of university, Volkovs said. Talent comes from leading academic institutions in Canada, the U.S. and around the world.
Together, TD and Columbia also run a generative AI executive education program to help some of the Bank's leaders develop foundational AI knowledge to further AI innovation across many areas at TD.
Want to learn more about Innovation?