TD has launched a predictive artificial intelligence (AI) foundation model, called TD AI Prism.
When it comes to AI technology, the creation of TD AI Prism represents the next major step forward for TD as it looks to use AI to help the Bank better anticipate customer needs.
Traditionally, AI technologies are designed to perform a single task, such as identifying which customers are most likely to need to open a savings account next.
Predictive foundation models, such as TD AI Prism, go to the next level and can predict a multitude of events; for example, which customers may need a range of savings or investment products to help suit their needs. With versatility, efficiency, and the ability to analyze massive amounts of data and identify nuanced patterns and signals, foundation models represent the next evolution of predictive AI.
The Bank's AI team – Layer 6 – recently announced the breakthrough, which represents a giant leap forward when it comes to providing AI supports and solutions for banking clients. Because this new model is cutting-edge technology, there's a lot to understand about how TD AI Prism works and what it means for TD clients and colleagues.
Here are three things to know about TD AI Prism.
The foundation model operates differently
Traditional AI has already started to reshape the way the Bank delivers services to clients. For example, the Bank uses AI to speed up processing for some mortgage pre-approval applications.
But traditional AI has limits. The technology performs well on the specific task it was trained to do. But it typically needs to be rebuilt each time it is deployed to take on a different task, or when it is used to work on a new product or solution for clients, otherwise known as a use-case.
A predictive foundation model is different. What sets it apart is how adaptable it is. The same model can solve new challenges, without having to be rebuilt each time.
One reason for this progress is that predictive foundation models are able to process significantly more data inputs than traditional AI models, allowing them to understand customer needs more holistically. Traditional AI requires a lot of prep work – called feature engineering – before a curated set of data inputs is used to train a model for a specific task.
Predictive foundation models like TD AI Prism don't have this need, and can thus adapt to solve new tasks with no additional engineering or input curation. Essentially this means predictive foundation models don't need to be trained from scratch to perform new tasks each time they are presented with a new form of data.
The development of TD AI Prism is also enabled by the Bank's secure cloud-based data platform, which is making it easier to access data at scale (which means it's able to handle and analyze large volumes of data quickly and efficiently). This cloud-based system supports a more streamlined process which unlocks TD AI Prism insights, helping connect TD customers with the products and services that can help suit their needs.
TD AI Prism is built with the Trustworthy AI approach
Bank researchers know AI needs to be built on a foundation of trust. That means the foundation model was built in alignment with the Bank’s governance best practices, which aim to address various aspects of trust, including security, privacy, and fairness.
TD has a specialist team that focuses on Trustworthy AI, and works within the Bank’s AI center, Layer 6.
Trustworthy AI is an approach to developing AI that includes checks and balances every step of the way. Because AI technology is designed to spot patterns and make recommendations, as the technology advances, making sure that its recommendations enable TD to stay true to its values and legal requirements is critical. That is why TD AI Prism was built based on Trustworthy AI considerations from the start.
In May 2025, TD's Trustworthy AI approach was named Best Responsible AI Program in the Americas at the DataIQ awards, across all industries.
TD AI Prism can provide a more personalized banking experience
One of the advantages of TD AI Prism is that it is customer-centred.
With traditional AI, each model is limited to one use-case, and it takes lots of time and effort to build it just to address one facet of a customer's banking profile, for example their mortgage.
TD AI Prism, on the other hand, is designed to break down those barriers and analyze needs from every angle simultaneously, so one model can see how the Bank can support customers in dozens of different areas, from debit products to lines of credit, and more.
That means a more personalized experience for the customer across a wide range of products and actions. For instance, the predictive foundation model can better evaluate what type of lending product could be right for a particular customer. Based on testing thus far, TD AI Prism is 20-30% better at identifying customer needs compared to existing models currently running in production.
With a combination of comprehensive coverage and predictive performance, TD AI Prism can help to better anticipate customer needs and deliver more personalized solutions. The Bank will roll out TD AI Prism in a series of use cases this year, scaling it for wider use, all while upholding the Bank's commitment to Trustworthy AI.