AI Agents for Mortgage Brokerage

How the money actually works here

Time to close is the number that decides everything in brokerage — it determines how many files a processor can carry, whether the rate lock holds, and whether the borrower is still with you at the end. And it is decided overwhelmingly by document chasing rather than by underwriting. Underwriting takes as long as it takes; the waiting is the variable. A file that closes in three weeks instead of five is not a faster underwrite, it is a chase that happened continuously instead of on the days someone got to it.

The rhythm this has to fit

Volume in this business is set by rates, which means the workload arrives in waves nobody can staff for. A refinancing wave triples file count within a month, and the processors you would hire take longer to become useful than the wave lasts. That shape rewards anything that scales without headcount, and it punishes fixed capacity in both directions — you are either overwhelmed or carrying staff through a quiet period. The chase is the part of the work that scales, and it is the part that does not need judgement.

Where the repetitive load sits

Three roles absorb most of it. Median wages are from the US Bureau of Labor Statistics, grossed up by the 1.43× private-industry loaded-cost ratio:

That is roughly $203,532 of annual compensation covering work that is, in large part, the same sequence repeated. 41% of that — the share Slack’s Workforce Lab found desk workers spend on work that is “low value, repetitive or lacks meaningful contribution to their core job functions” (n=10,281, six countries) — is about $83,448 a year.

Read across those three roles, the work itself is customer onboarding, expense management, contract management, order management, customer support and meeting follow-up. Each of those is a loop with its own failure points, and the page for it says where the loop breaks rather than where the job title sits.

Against BreathingRoom at $30,000 a year, recovering just 36% of that covers the engagement.

What we would build first

• Borrower document collection and chasing
• File completeness checking
• Status updates at every stage

Which goes first is decided in the audit rather than in advance. Anyone who tells you the order before looking at your volumes is guessing.

How this goes wrong in mortgage brokerage

Automated status updates create a specific trap here. The borrower is told the file is in underwriting, which is true and has been true for eleven days, so they stop asking — and a file nobody is asking about ages quietly. The reassurance has replaced the pressure that used to move it. The guard is that status messages are tied to real state changes rather than sent on a cadence, plus an ageing alert on any file that has not moved within a defined window. If the only thing that changed this week is that the borrower feels better, something is wrong.

When you should not do this yet

A broker doing a handful of files a month will not recover the build cost, and a broker whose lenders all require document upload through their own portals gets much less from this than the volume suggests — the chase can be automated but the submission may not be. Check where the manual work actually sits before assuming. If it is mostly inside a lender's portal, the saving is the chase alone, which may still be worth it but is a smaller number than it first appears.