AI mortgage underwriting tools in Canada: the 2026 landscape.

At a functional level, AI underwriting tools read and extract data from income documents, bank statements, and credit reports; recalculate GDS/TDS ratios at the qualifying rate rather than the contract rate; and flag inconsistencies or missing items against a checklist — the same broad tasks a skilled fulfillment specialist performs manually, run faster and more consistently.

Done well, the output looks like the pre-underwriting review covered in how mortgage underwriting works in Canada: a written calculation and a flagged list of issues, produced in a fraction of the time a manual first pass takes.

It can never make the credit decision. In Canada, only a licensed underwriter at the lending institution — or, for insured deals, the underwriter alongside the mortgage insurer — approves a mortgage, under OSFI Guideline B-20. Any tool, description, or vendor claiming otherwise is describing something other than what it actually does.

That distinction matters for how a brokerage frames the tool internally, too: it's a pre-underwriting aid, not a replacement for the lender's review, and the brokerage's own regulatory accountability for the file doesn't move because software touched it first.

Without naming specific products, the tools showing up on the Canadian broker side generally fall into a few categories: document-extraction assistants that pull and organize data from a file without doing much calculation; ratio-and-checklist tools that recalculate GDS/TDS and flag gaps against a defined rule set; and fuller pre-underwriting review platforms that combine both, producing something closer to a complete pre-submission package.

Which category fits a given brokerage depends less on the label and more on how much of the manual review it's meant to replace, and how much a human is still expected to check afterward.

1. 01Does it show its work? A usable tool produces a visible calculation and a specific list of flags — not just a pass/fail score with no explanation.
2. 02Is it built against Canadian rules? GDS/TDS at 39%/44%, the MQR formula, and Canadian document conventions are not the same as US underwriting standards — a tool trained on the wrong rule set will confidently produce the wrong answer.
3. 03What does it escalate to a human, rather than guess at? Non-standard income, gifted deposits, and bureau anomalies need a flagged handoff, every time — a tool that silently resolves ambiguous cases is a liability, not a feature.
4. 04Who's accountable if it's wrong? The brokerage retains regulatory responsibility regardless of the tool's output — confirm that's understood, in writing, before relying on it for volume.

OSFI's Guideline E-23 on model risk management, finalized and set to take effect in May 2027, sets expectations for how federally regulated financial institutions — banks, insurers, and similar — govern the models they use, explicitly including AI and machine learning. It doesn't regulate mortgage brokerages directly; brokers aren't federally regulated financial institutions in the sense E-23 applies to.

What it does signal is direction: Canada's prudential regulator is treating AI model governance as a real supervisory priority for the institutions that ultimately make the credit decision on every file a broker submits. A brokerage adopting an AI underwriting tool today is reasonable to anticipate that lenders will expect more transparency about how AI-assisted files were prepared, not less, as this framework comes into force.

Engage's AI mortgage underwriting, built specifically for Canadian brokers and agents, is currently in early access with an email waitlist rather than general availability — a tool a broker or agent uses themselves to pre-underwrite a file faster and more cheaply, not a done-for-you service where someone else underwrites on the brokerage's behalf.

For the fuller comparison against a human contract underwriter or a fulfillment partner, see Outsourced Mortgage Underwriting in Canada, and for how AI and human review specifically compare task by task, continue to our companion piece on AI vs. human underwriting.