What is an AI underwriter?

"AI underwriter" is a name that promises more than most tools deliver, which is part of why brokers are sceptical of it. The useful version is narrower and far more practical than the phrase suggests: software that reads the documents in an application package the way an experienced assistant would, checks them against each other, does the arithmetic, and tells you what looks off before a lender does.

What does an AI underwriter actually do?

It starts with reading — not scanning for keywords, but actually parsing the documents a borrower sends you: paystubs, notices of assessment, T4s, bank statements, credit reports, MLS listings. On a normal file that is dozens of pages arriving in three different formats over four days, and the traditional process is a broker or an assistant opening each one and typing numbers into a worksheet.

The second step is where the value is: cross-referencing. Does the income on the paystub reconcile with the T4 and the NOA? Does the deposit pattern in the bank statements match the employer named on the application? Does the credit report show an obligation the client never mentioned? A document read in isolation looks fine; the problems live in the gaps between documents, which is exactly where a tired human reader stops looking.

Third, it does the numbers. GDS and TDS are calculated with the OSFI B-20 stress test applied, so the ratio you are looking at is the one the lender's system will produce rather than the optimistic one from the contract rate.

Finally, it flags and summarises. Income mismatches between sources, NSF activity in the statements, undisclosed liabilities, gaps in employment, and deposits that need a source-of-funds explanation all get surfaced as specific items to verify. The output is a broker-ready summary you can actually read — the numbers, the flags, and the reasoning behind them, in a form you can use to have a straight conversation with the client before the file goes anywhere.

What it does NOT do

It does not approve or decline deals. This is the line that matters, and any tool blurring it should worry you. An AI underwriter has no lending authority, no lender risk appetite, and no accountability for the outcome. It performs document analysis; the broker decides what the file means, which lender it suits, and how to present it.

It also does not replace your judgment about the client. A flag is an item to check, not a verdict. Plenty of perfectly good files contain an income structure that looks inconsistent on paper until someone asks one question. The tool's job is to make sure that question gets asked before an underwriter asks it for you.

Why it matters for Canadian files

Canadian mortgage files are unusually document-heavy and unusually rule-bound. B-20 means qualifying is done at a stressed rate, so the affordability conversation you had at the contract rate is not the one the lender will have. GDS and TDS caps turn small income and liability discrepancies into approval-or-not differences — a car loan the client forgot about can move TDS past the line on its own.

At the same time the volume of supporting paper keeps rising: self-employed income across multiple years, rental offsets, gifted down payments needing source-of-funds documentation. The failure mode is not that brokers cannot do this analysis; it is that doing it thoroughly on every file, at eleven at night, in a market where speed wins deals, is not realistic. Automating the reading is how the analysis actually gets done consistently.

What to look for in one

It reads all document types together. A tool that only handles paystubs is an OCR feature. The value comes from cross-referencing across the whole package, so it needs to ingest statements, credit reports and tax documents in the same pass.

It shows its reasoning. A number without an explanation is unusable — you cannot defend it to a lender or explain it to a client. Look for output that says where a figure came from and why an item was flagged, so you can check it in seconds rather than redoing the work.

It keeps data in Canada. You are feeding it a borrower's most sensitive financial records. Where those documents are stored is a question your brokerage's compliance process — and eventually a client — will ask, and "hosted in Canada" is a much simpler answer than the alternative.