Australian mortgage brokers settled 92.06% of the home loan applications they lodged in 2025, down from 93.06% in 2024, according to the MFAA State of Mortgage & Finance Broking Report 2026. Roughly one application in twelve is fully prepared and never paid, and AI reduces that cost by fixing files before submission rather than after.
The idea
The commercial problem in Australian mortgage broking is not lead volume, it is the cost of work that never converts. The MFAA State of Mortgage & Finance Broking Report 2026 recorded 911,150 home loan applications lodged across an eight-aggregator sample in 2025 against 838,815 loans settled, putting the settlement-to-application ratio at 92.06%. An application that does not settle has already absorbed document collection, serviceability work, file notes and a submission, and none of it earns a cent.
Volume growth has been masking the slip. Residential settlements through Australian brokers reached $495.55 billion in 2025, up 23.5% on 2024, and broker numbers grew 9.1% to 24,116 (MFAA, 2026), so most brokerages have seen revenue rise whilst their conversion quietly went backwards. Brokers also held a record 81.0% of Australian residential home lending in the March 2026 quarter on Cotality data reported by the MFAA (2026), so the channel now carries both the volume and the unpaid processing attached to it. It is the same pattern that sits behind why Australian brokers are losing settlements they have already won, measured this time at industry level.
Why it matters
A one percentage point move in the settlement-to-application ratio is worth real money to an owner, and it is measurable inside this quarter. The arithmetic below uses the MFAA's 2025 industry figures plus three Alvo assumptions, stated so an owner can substitute their own.
| Step, a brokerage lodging 600 applications a year | Calculation | Result |
|---|---|---|
| Settled at the 2025 industry ratio (MFAA, 2026) | 600 × 92.06% | 552 |
| Settled at the 2024 industry ratio (MFAA, 2026) | 600 × 93.06% | 558 |
| Settlements lost to the one-point slip | 558 − 552 | 6 |
| Average settled loan, 2025 (MFAA, 2026) | $495.55 billion ÷ 838,815 loans | ~$591,000 |
| Upfront commission per settlement | ~$591,000 × 0.65% (Alvo assumption) | ~$3,840 |
| Upfront revenue lost | 6 × $3,840 | ~$23,000 |
| Applications prepared and not settled | 600 − 552 | 48 |
| Unpaid preparation cost | 48 × 6 hours (Alvo assumption) × $45 an hour loaded (Alvo assumption) | ~$12,960 |
| Combined annual cost | ~$36,000 |
Clawbacks sit on top of that. Median gross clawback per broker rose to $11,442 in 2025 from $9,820 in 2024 on a matched five-aggregator sample (MFAA, 2026), so revenue already recognised is also moving backwards.
Where AI changes the number is pre-submission, not post-approval. The table below separates the two.
| File stage | Common process today | With AI applied pre-submission |
|---|---|---|
| Document collection | Support staff chase payslips, statements and tax returns by email, then eyeball them | Documents are read, classified and checked for completeness on upload, with gaps flagged the same day |
| Serviceability | Broker runs calculators lender by lender, often after the client meeting | Serviceability is pre-screened across lender policy before a lender is nominated |
| File notes and credit memo | Written from memory or from notes, hours after the appointment | Drafted from the source documents and the recorded conversation, then edited by the broker |
| Policy check | Discovered at assessment, as a recondition or decline | Checked against lender policy before lodgement |
| Compliance record | Assembled at audit time | Assembled as the file is built |
The Alvo take
Alvo's position is that an Australian brokerage should measure its own settlement-to-application ratio before it buys anything, because AI applied to an undefined process simply produces faster mess. The pathway that works is narrow: define what a complete file looks like, instrument the ratio and the hours per lodged file, then apply AI to pre-submission checking, which is the one place unpaid rework actually sits. That is where Alvo's work with mortgage brokers starts, and an Assess engagement exists to build that baseline before any tool goes in.
Two regulatory points are worth flagging, because both land inside the next twelve months. From 10 December 2026, Australian Privacy Principle 1.7 requires an entity to disclose in its privacy policy where a computer program makes, or substantially informs, decisions that significantly affect a person's rights or interests, according to the OAIC (2026), and a brokerage using AI to score or pre-assess applicants is inside that scope. ASIC also wrote to Australian banks on 29 September 2026 opening a review of new and proposed AI use cases in lending and customer decisions (2026), so lender-side policy keeps moving whilst brokerages build.
The adoption gap is the opportunity, and it is wider than most owners assume. The ABS Business Characteristics Survey for 2024-25, released in 2026, found around 12% of Australian businesses using AI, rising to 22% of medium businesses and 35% of large ones, so a founder/owner-led brokerage gets ahead of its peer group by doing one thing properly. CSIRO (2026) found that Australian firms adopting AI posted 36% more non-AI job advertisements over time than non-adopters across more than 4,000 firms, so the Australian pattern is capacity growth rather than headcount reduction. Globally, BCG (2026) projected AI could lift bank profitability by 30% and cut costs by 30% to 40% by 2030, which is a lender-side number, and the Australian read-across for brokers is that lender assessment keeps getting faster whilst an incomplete file carries its friction on the brokerage's own margin.
Australian brokers settled 92.06 per cent of the home loan applications they lodged in 2025, down from 93.06 per cent, whilst volumes grew 18.4 per cent and hid the slip. On a brokerage lodging 600 applications a year, one point of conversion is six lost settlements worth ~$23,000 in upfront commission, plus ~$12,960 of preparation on 48 files that were never paid, ~$36,000 a year in total. The unpaid work sits before submission, which is where AI document checking and pre-lodgement policy checks actually move the number.
Sources: MFAA, State of Mortgage & Finance Broking Report 2026, published September 2026, covering calendar 2025, reported by The Adviser, Mortgage Professional Australia and Australian Broker. MFAA and Cotality, March 2026 quarter market share, June 2026. OAIC, transparency for automated decision-making under APP 1.7, September 2026. ASIC review of AI use in bank lending, September 2026. ABS, Business Characteristics Survey 2024-25, released June 2026. CSIRO, research into firms adopting AI, April 2026. BCG, How Retail Banks Can Put Agentic AI to Work, March 2026 (global). The worked-example inputs (600 applications, 0.65 per cent upfront commission, 6 hours per file, $45 an hour loaded) are Alvo assumptions, labelled as such, not published figures.
Common questions
Will AI replace mortgage brokers in Australia?
AI is not replacing Australian mortgage brokers, because the broker holds the credit licence, the best interests duty and the client relationship. Mortgage brokers held a record 81.0% of Australian residential home lending in the March 2026 quarter, according to the MFAA. AI changes the cost of preparing and checking a loan file, not who owns the advice.
What can AI actually do in a mortgage broking business?
In an Australian mortgage broking business, AI reliably reads and classifies payslips, bank statements and tax returns, drafts file notes and credit memos from the source documents, checks a submission against a lender's policy before lodgement, and flags missing items. AI does not choose the loan, sign the credit advice or carry the best interests duty, which stay with the broker.
Do I have to tell clients my brokerage uses AI?
From 10 December 2026, Australian Privacy Principle 1.7 requires an entity to disclose in its privacy policy where a computer program makes, or substantially informs, decisions that significantly affect a person's rights or interests, according to the OAIC in 2026. An Australian brokerage using AI to score, rank or pre-assess applicants should update its privacy policy before that date.
How much does it cost to get AI working in a mortgage brokerage?
Most Australian brokerages start inside tools they already pay for, because AI document reading and file-note drafting now ship inside mainstream CRM and lodgement platforms rather than as a separate purchase. The larger cost is process work: agreeing what a complete file looks like, and training support staff to work to it. Budget time before budgeting software.
Where should a mortgage brokerage start with AI?
An Australian mortgage brokerage should start by measuring its own settlement-to-application ratio and the hours spent per lodged file, then apply AI to pre-submission checking, which is where unpaid rework sits. Measuring first matters, because the MFAA reported the industry ratio slipping from 93.06% in 2024 to 92.06% in 2025 whilst volumes grew.