TD Bank just put its first agentic AI model into production — and picked an unglamorous, high-friction target: mortgage and HELOC pre-adjudication. The before-and-after number is the pitch. Pre-adjudication used to average roughly 15 hours before a human underwriter could even open a file. That window is now under three minutes.
## What the agent actually does
Built by Layer 6, TD’s internal AI research arm, the system runs the document-handling grind end to end before any human looks at the file: it classifies borrower documents, extracts and validates financial data, calculates income, runs policy and consent checks, flags discrepancies, and writes a summary memo for the underwriter. The human still adjudicates — the agent clears the 15 hours of prep.
## Why it matters
Most “agentic AI” announcements are demos. This is a regulated bank shipping one into a core lending workflow, with a measurable cycle-time cut on a process customers feel directly. TD has mapped the full mortgage journey from document submission to funding and plans to add agents at each stage. The first live application is the document bottleneck — the part that was pure latency, not judgment, and the safest place to let an agent go first.

Leave a comment