Canadian lender tests machine learning models for near-instant home loan decisions
Toronto-Dominion Bank is turning to artificial intelligence to give customers faster service and decisions on mortgages, insurance and other banking products through a series of pilot projects.
The projects are spearheaded by Layer 6, TD's in-house AI research lab. The bank is utilizing AI to speed up mortgage and insurance approvals, enhance customer service responses, and boost the efficiency of its engineers.
"We're looking at all the ways we can really help our colleagues," Luke Gee, head of AI and analytics at TD, told the Financial Post. "We're excited about these human augmentation opportunities where we can make the roles more interesting and quicken up our response, which gives us better customer outcomes as well."
One key area where TD is deploying AI is mortgage and home equity line of credit (HELOC) applications. A new AI model introduced in August 2023 has been trained on previous application data and the bank's underwriting practices. For straightforward applications like those from first-time homebuyers, the AI system can now issue pre-approvals within seconds.
“Buying a home is really stressful and so we want to make sure we can give people answers as quickly as possible. It happened that AI is a good solution to that scenario,” Gee said. “A lot of customers have very simple requirements … and so we wanted to make sure we can give them an answer and they can move on to the next part of the process.”
The mortgage AI has already granted thousands of pre-approvals in its first months, though TD declined to provide an exact number. The rapid pre-approvals have also freed up human underwriters to handle more complex files requiring additional skill and analysis.
Gee said Layer 6 is examining more potential applications across retail banking, investing and insurance.
TD is also using machine learning models to instantly approve straightforward term life insurance applications, addressing a "significant" increase in demand that had slowed response times.
Beyond mortgages and insurance, the bank's engineers are leveraging Microsoft's GitHub co-pilot AI programming assistant to increase their coding speed and catch bugs faster.
"If our engineers are able to build more quickly, then the feature is able to be released more quickly to the customer," said Gee. In a pilot from September to December 2023, half of surveyed engineers said the AI tool saved them up to 20 hours over a two-week span.
TD has also developed a generative AI virtual assistant to help customer service agents answer inquiries more quickly and accurately by surfacing relevant policies and procedures. The AI responds conversationally and directly quotes sources so agents can verify the information.
"Our agents want to be able to give a good answer to our customers at a quick rate, and they feel confident about the answer they've given," Gee said. "To be able to provide information that is easily absorbable and easily relatable is where generative AI has helped us."
Read next: OSFI eyeing stricter rules concerning the use of AI by banks
While introducing AI carries risks around accuracy and compliance, TD said its virtual agents are instructed to say "I don't know" if required information can't be found. Agents can also rate the AI's responses to further train the models.
The push into AI comes as banks globally ramp up related research and patent filings. TD said its AI patent portfolio has grown 110% since 2020 to 2,500 filings, one-fifth tied to artificial intelligence.
Make sure to get all the latest news to your inbox on Canada’s mortgage and housing markets by signing up for our free daily newsletter here.
