TD launches AI model to speed mortgage applications

TD has launched its first agentic AI model for mortgage and home equity line of credit applications in its real estate secured lending business.

The model automates the pre-adjudication stage by producing application summary memos for underwriters. Work that previously took an average of 15 hours can now be completed in less than three minutes.

Developed by Layer 6, the bank's AI research and development centre, the model was built in collaboration with teams across technology, data, lending, and risk management. It classifies client documents, extracts information, calculates income, checks figures against selected policy requirements, performs consent checks, searches for discrepancies and then creates a summary for underwriters.

TD is applying the technology first to mortgages and home equity lines of credit, or HELOCs, as part of a broader overhaul of its real estate secured lending operations. It has mapped the full lending process, from document submission to the release of funding, and plans to introduce agentic AI at each stage.

Speed gains

Mohit Veoli, Senior Vice President, Real Estate Secured Lending at TD, said the bank is using the model to address long-standing customer concerns about wait times in the home lending process.

"Agentic AI is enabling us to deliver what clients tell us matters most - speed and simplicity," Veoli said. "By providing confident decisions earlier in the homebuying process, one of life's milestone moments, we're meeting our clients where they need us most and making the experience simpler and faster. This is just the beginning of how we're using AI to evolve the mortgage experience and bring our promise of being more human to life."

TD described the launch as an early step in a broader AI strategy across the bank. It is working to generate USD $1 billion in annual value from AI in the coming years while extending the use of agentic AI across other business lines.

Agentic AI refers to autonomous software agents that use generative AI models to carry out tasks with limited human intervention. In TD's lending process, the tools are intended to support underwriters rather than replace them, handling document-intensive preparation work before a person reviews the file.

Human oversight

Luke Gee, Chief Analytics and AI Officer, TD, said the model is part of a hybrid approach that combines staff and AI systems.

"Agentic AI will supercharge our work and put us in a position to deliver more accurate and faster results for our clients," Gee said. "We're building a hybrid future where our colleagues and AI work together to help our clients get to a 'yes' faster on some of their most important decisions. We're going to approach our vision responsibly so that we maintain the trust of our clients and colleagues."

Oversight of the model sits with TD's Trustworthy AI team, which reviews systems for privacy, security, fairness, accountability and explainability before deployment. The team continues to monitor models after they are introduced into live operations.

The move highlights how large lenders are starting to use newer forms of AI in core banking processes rather than limiting them to customer service tools or internal pilots. Mortgage underwriting has been a natural target because of the volume of documents involved, the number of policy checks required and the pressure on banks to shorten approval times without increasing risk.

TD is one of North America's largest banks by assets and serves 28.1 million clients across Canadian personal and commercial banking, US banking, wealth management and insurance, and wholesale banking. It had assets of $2.1 trillion at the end of January.

For TD, the real test will be whether the system can maintain speed and accuracy as it expands beyond pre-adjudication into the rest of the lending journey while remaining under the scrutiny of its internal Trustworthy AI controls.