TD Bank's AI Cuts Mortgage Review From 15 Hours to 3 Minutes

TD Bank Group just gave the rest of North American banking a uncomfortable benchmark. On May 21, TD launched its first agentic AI system for mortgage and HELOC (Home Equity Line of Credit) applications — and early data shows it's compressing a 15-hour underwriter task into less than 3 minutes.

The system was built by Layer 6, TD's in-house AI research lab, and is now live across the bank's Real Estate Secured Lending (RESL) operations. It's the most consequential "AI replaces actual work" deployment any major North American bank has shipped this year.

The Context: Why Mortgages Are the Killer Use Case

Mortgage processing is the kind of work agentic AI was designed to eat. It's high-stakes, document-heavy, and full of cross-referencing tasks that follow predictable rules — but variable inputs. A single application can include 40+ documents: pay stubs, tax returns, bank statements, identity verification, property appraisals, and a dozen more.

For decades, banks responded with armies of mortgage processors and underwriters. The work is slow, error-prone, and structurally expensive. Industry estimates put the all-in cost of processing a North American mortgage at $8,000–$11,000 per loan — most of it human labor.

Layer 6 has been operating inside TD since the bank acquired the Toronto-based startup in 2018. Until now its work surfaced as quiet ML improvements: better fraud scoring, smarter offers, faster anti-money-laundering flags. The mortgage agent is the first time Layer 6 has shipped a customer-facing deployment that visibly changes how a core banking workflow runs.

What the Agent Actually Does

Layer 6's agentic AI handles what underwriters call the "summary memo" — the document that gets prepared before a human makes the lending decision. It runs four jobs autonomously.

First, it classifies every document an applicant submits. A pay stub gets tagged as income evidence, a property tax bill as collateral context, a notice of assessment as a CRA confirmation.

Second, it extracts the key data: employer name, gross income, debt obligations, property valuation, identity fields. Third, it validates that data against the bank's policy rules — debt-service ratios, income stability minimums, identity verification requirements — and flags anything that fails or sits in a gray zone.

Fourth, it runs consent and discrepancy checks: did the applicant authorize the right data sharing? Do the bank statements match the income claim? Does the property appraisal align with the loan amount? Then it generates a concise memo for a human underwriter to review and sign off.

The result is the same artifact underwriters used to produce by hand — just made in minutes instead of business days.

Industry Impact: The Productivity Math Is Brutal

The implications for the rest of banking are stark. If TD's results hold at scale, a single underwriter who used to clear 2–3 applications per day could clear 50+ — and spend their time on the judgment-heavy edge cases instead of paperwork. Banks that don't follow risk being underpriced and out-paced on speed.

The competitive pressure is already visible. RBC, BMO, and Scotiabank all have AI labs of comparable maturity but haven't publicized agentic deployments at this scale. South of the border, JPMorgan, Bank of America, and Wells Fargo have all signaled mortgage AI roadmaps but have moved more cautiously, partly because U.S. fair-lending regulation makes any automated underwriting input legally fraught.

For consumers, the obvious benefit is speed. A pre-approval that used to take a week could be back in hours. For mortgage brokers, who already compete on response time, it raises the floor of what borrowers consider acceptable.

Expert Perspectives

Industry analysts framed the launch as a meaningful proof point rather than a one-off. PYMNTS called it "the AI deployment that finally made mortgage waiting obsolete." Canadian Mortgage Trends noted that TD's framing was deliberately conservative — the agent prepares the memo; humans still make the decision — which is the model regulators across G7 jurisdictions have signaled they'll accept.

TD itself was careful to position this as a productivity tool rather than a replacement. "TD has mapped every step of the RESL journey, from when a client submits their documents to when funding is released, and will be introducing agentic AI in each one," the bank said in its release.

What's Next: Beyond Mortgages

TD has telegraphed that mortgages and HELOCs are the starting point, not the destination. The bank says it's "exploring opportunities to introduce agentic AI in other businesses across the Bank" — code for commercial lending, wealth onboarding, and small-business credit, all of which involve similar document-heavy approval flows.

For Layer 6, the launch is a coming-of-age moment. After eight years inside TD as a quiet research lab, its work is now sitting at the front of a $400 billion-a-year Canadian mortgage market. Expect the playbook — agentic AI on the underwriter side, humans on the decision side — to show up at every major North American bank by year-end.

Bottom Line

The mortgage industry just got a public benchmark for what agentic AI does to its core workflow: a 99.7% time reduction on the most labor-intensive step. The technology is now boring enough to ship at a Big Five bank, conservative enough to satisfy regulators, and dramatic enough to redefine what consumers expect. If you're a banker betting AI is overhyped, TD just made the bet harder to win.