AI for Mortgage Brokers Wins Deals You Were Going to Win Anyway

The discharge authority lands on a Tuesday morning.

A client you settled three years ago. Clean file, a Christmas card most years. She has refinanced through another broker, into a rate you could have arranged yourself if she had rung. She did not ring. Next month the trail stops, and it does not come back.

That same week, the AI you pay for answered a lender policy question in seconds rather than across five lender portals. A genuine saving, on a deal you were always going to write.

That is the shape of AI for mortgage brokers as it is sold today. Nearly all of it points at the live deal: servicing across dozens of lenders on one screen, a policy assistant that reads the lender guide for you. The tools are real, and most are cheap. They are also aimed at the half of your business that already pays.

I build these systems for Australian businesses, so let me be blunt about the half nobody demos. A broking business does not bleed on the deal in front of it. It bleeds behind it.

What AI for mortgage brokers actually does today

Split it in two, because treating it as one thing gets brokerages to buy the wrong thing.

The first half is deal work: serviceability comparison, lender policy search, payslip handling, a first pass at the file notes. Useful, priced per seat, and where every vendor starts. It is also the work you already get paid for. You still read every answer, because the duty sits with you rather than the software.

The second half is the book and the chase: watching settled loans for the moment a rate drifts out of the market, collecting the documents that stall a live file for a fortnight, following up the pre-approval that went quiet. That work was always mechanical. It never had anyone spare, so it slipped, and it kept slipping, because none of it feels urgent until the discharge lands.

The trail book is the asset nobody watches

You are paid a trail every month for as long as that loan stays on the lender's book. Write it once, get paid for years. It is the closest thing a broking business has to an annuity, and most brokers could not name the ten clients most likely to leave this quarter.

When a client refinances away, the trail simply stops. Nobody sends a warning. The discharge authority is the notice. If the loan is young enough to sit inside the lender's clawback window, you hand back part of the upfront as well.

The industry has a tidy word for it: run-off. A polite way of saying the asset is leaking.

You get paid every month for a loan you wrote once. Most brokers ring that client once, if they remember.

Start with the book you already wrote

Pick one workflow, not the whole business. For most brokerages the best place to start is the one that gets dropped the moment the pipeline gets busy: watching the settled book.

A client is eighteen months into a loan, on a rate the same lender no longer offers new borrowers. Instead of nobody noticing until she does, the drift gets flagged, a repricing request goes to the lender, and she hears from you first. If the lender will not move, that becomes a refinance conversation with you, not an ad she clicked at eleven at night. What the system cannot settle goes to a person.

A repricing email you sent beats a discharge authority you did not see coming.

That gives you a number to defend: the trail you kept, against the upfront you did not repay. Arithmetic, not a vibe, and I set out how to do it honestly in the AI ROI framework.

The second workflow, once the first is proven, is the document chase: nudging the clients who go quiet, catching the bank statement that arrived as a photograph of a screen. It is what stalls a good file, and it is almost pure mechanics.

Best interests duty makes this easier, not harder

Here is where brokers talk themselves out of it. Best interests duty has applied since the start of 2021, and the guidance is plain: where your interests conflict with the client's, theirs come first. Read quickly, that sounds like a reason to leave your own book alone.

It is the opposite. A client who is better off elsewhere is leaving whether or not you were watching. The only question is whether they leave through you or around you. The broker who spots the drift first has the conversation, does the right thing, and keeps the relationship. The broker who finds out at discharge loses the trail and the client in the same envelope.

A client who leaves through you is still a client. A client who leaves without you is gone twice.

The duty sets the guardrail too. Every inquiry has to leave a record, because ASIC expects to see how you acted in the client's interest. A system that drafts the file note and files it is doing compliance work. A system that nudges clients toward whatever pays you best is a licence problem wearing a chatbot.

What AI for mortgage brokers costs to build and run

The bands are real. A per-seat policy or serviceability subscription is cheap enough to trial on one broker for a month. A retention-and-document workflow built to fit your business and talk to your CRM usually lands between a few thousand and the mid teens of thousands of dollars to build and go live. A connected setup across CRM, lodgement and marketing starts higher again.

What catches owners out is the running cost, not the build: the subscriptions underneath, the model usage, and the staff time still spent reviewing output and handling what the system escalates. I broke those layers down in what AI actually costs a small business. Never pay for a custom build where a setting in the CRM your aggregator already gives you would do the job. A vendor whose every answer is their priciest option is solving for their invoice.

The Australian layer: your aggregator's CRM and client data

Two things separate a system that works in an Australian brokerage from a generic overseas template.

The first is integration. You run on Mercury Nexus, AFG Flex, MyCRM, Infynity or Salestrekker, and you lodge through ApplyOnline or Loanapp. A tool that cannot read and write your real client list has not removed the double handling, it has moved it to your loan processor. Ask what it writes back, and where, before you fall for the demo.

The second is duty of another kind. The moment a system touches payslips, bank statements and credit files, it handles personal and credit information under the Privacy Act. Send that offshore and the cross-border rules keep you accountable for what the overseas recipient does with it. Ask whether it is stored in Australia, whether it trains someone else's model, and whether you can delete it on request. A vendor who cannot answer has told you how carefully they build.

Common questions about AI for mortgage brokers

What should a broking business automate first?

Retention on the settled book, then document collection on live files. Not lender selection. That is where the trail quietly leaves and where good deals stall. Prove the clients kept on one workflow before you widen, because automating five things at once means you cannot tell which one paid.

Will AI replace mortgage brokers?

No, and brokerages that aim it that way point it at the wrong target. Best interests duty is carried by a person. ASIC and AFCA hold your licence, not your vendor. The software clears the watching, the chasing and the note taking, so your people can do the advising the client pays for.

Is a custom build worth it over what my aggregator gives me?

Only once you have turned their tools on and hit the limit. The custom value sits in the retention-and-document workflow that fits how your business runs, and only if you own the system and the client data at the end.

If you want a straight read on where your book is leaking, and whether it needs a custom build at all, that is what a first call is for. Book a strategy call and bring last quarter's trail statement and the discharges you did not see coming. We will find the clients you are about to lose before we talk about building anything.