LMG Brokers Are Saving 50 Minutes a File With AI. The Risk Is What the AI Writes in Your BID Notes

1. What LMG actually reported

Loan Market Group has published the first substantial efficiency data from an Australian aggregator’s own AI tooling, and the numbers are better than the sceptical case would have predicted.

Measured in July 2026, approval times improved 18 per cent year-on-year. Separately, the median time from lodgement to approval fell 10 per cent over the preceding 12 months. More than 3,800 users accessed the group’s AI tools. Brokers reported saving an average of 50 minutes per application, with some reporting close to an hour.

The tools sit inside a platform called MyCRM Intelligence, launched in December 2025. Two components do most of the work. MyNoteWriter generates lender rationales, product selection notes and exit strategies from application data. MyQualityAssurance reviews a deal across more than 500 data points before lodgement, flagging missing information, funding gaps and product mismatches.

LMG chief product and technology officer Whitney Cali linked the tooling to submission quality rather than speed alone: “More brokers are using MyCRM Intelligence and reporting faster time-to-lodgment, greater compliance confidence, and cleaner submissions that are driving higher first-time approval rates.”

Two brokers were quoted with specifics. Tania Richardson, operations manager at Loan Market Ellenbrook in Western Australia: “We’re saving 50 minutes, on average, on every application with MyCRM Intelligence.” Sid Malhotra, director of Loan Choice Australia in Queensland, whose settled deals were up 49 per cent year-on-year: “We still do our own checks, but the MyQualityAssurance tool takes it to another level.”

One caution on the arithmetic. LMG did not publish a total broker count alongside the 3,800 figure, so it cannot be converted into an adoption rate. And the 18 per cent and 10 per cent figures cover different periods and different measures — a July-on-July comparison and a rolling 12-month median respectively. They should not be added together or used interchangeably.

2. This is not an LMG-only story

Two other aggregators have moved in the same direction, and one has published comparable data.

Connective reported in April 2026 that its Client Portal had been adopted by more than 1,400 brokerages and 3,500 brokers, handling over 27,000 client requests with a 59.8 per cent completion rate in March. It cited up to one hour saved per deal, and income and expense verification falling from around two hours to 10 to 15 minutes. Chief technology officer Chin Hui Yeo framed the design principle sharply: “Brokers do not need more disconnected tools or more noise layered over their business.”

Finsure launched Metanoia, described as an AI-native CRM replacing Infynity, in March 2026, with a staged rollout beginning in New Zealand and no forced migration. Chief executive Simon Bednar described it as “effectively renovating the house room” by room. No adoption or time-saving statistics have been disclosed.

For AFG, no 2026 AI adoption or efficiency data could be verified.

The pattern is that platform capability is becoming an aggregator differentiator with published numbers attached, rather than a roadmap slide. If you are choosing or reviewing an aggregator this year, this is now a comparable metric rather than a vibe.

3. The bottleneck is moving, but not necessarily to you

Here is the part of the picture the aggregator announcements do not cover.

Experian research reported in August 2026 found that 72 per cent of lenders are using agentic AI for underwriter decision support — but only 3 per cent have fully AI-ready data systems, 67 per cent say their data is not ready or only partially ready, and just 11 per cent have AI widely implemented in underwriting. Experian’s managing director of software solutions, Mathew Demetriou, put it plainly: “The foundations underneath AI still need to catch up.”

Read that against LMG’s numbers and a specific conclusion follows. The broker-side gains are real and measurable, because the work being automated — note writing, document checking, data validation — is well-defined and sits entirely within the broker’s own system. The lender-side gains are far less certain, because underwriting AI depends on data infrastructure that most lenders admit is not ready.

The practical implication is that the win to chase is not turnaround. It is first-time approval rate. A submission that does not come back for more information avoids the delay entirely, regardless of how AI-ready the lender’s credit engine is. That is exactly what Cali pointed to, and it is a more durable benefit than raw speed.

4. The compliance problem hiding in the productivity gain

MyNoteWriter, on LMG’s own description at launch, generates lender rationales, product selection notes and exit strategies from application data. That is a genuinely useful capability. It is also, described precisely, a system for producing plausible reasoning from structured inputs.

Now place that beside what ASIC has said it will be looking for. Commissioner Alan Kirkland, at the MFAA conference in Melbourne on 22 July 2026: “If the reasons for a recommendation are boilerplate factors that could apply to anyone, then it will be hard to demonstrate that the recommendation was in that customer’s best interests.” He also warned that “acting in the customer’s best interests emphatically does not mean simply taking orders when you know a product isn’t right for them”, and confirmed ASIC expects to publish its best interests duty report “in the final quarter of this calendar year”.

The risk is not that AI-generated notes are wrong. It is that they are generically right. A rationale assembled from application data will accurately describe the loan, the product features and the client’s stated objectives — and will read almost identically to the rationale on the next file, and the one after that. That is the definition of boilerplate, produced faster and at greater volume than a human could manage.

There is a second-order version of the same problem. If an aggregator’s tool produces the notes for thousands of brokers, then a reviewer sampling files across that aggregator will see the same structure and often the same phrasing repeatedly. Consistency that looks like quality control from inside can look like an absence of individual reasoning from outside.

5. Human in the loop is a philosophy, not a control

LMG has been explicit about its framing. Cali, setting out the group’s AI roadmap in March 2026: “AI tools aren’t taking away your job — they’re just taking away two hours of busy work each day.” And: “The human in the loop is our philosophy in which the broker is the final decision maker as AI tools work behind the scenes.” At the December launch she added: “Our vision is to give brokers intelligent tools that save them time without ever compromising trust.”

That framing is the right one, and it also quietly transfers the responsibility. If the broker is the final decision maker, then the broker owns the output. A generated rationale on your file is your rationale. An exit strategy drafted by a tool is your exit strategy. There is no version of a review in which “the system wrote that” is a useful answer.

The roadmap makes the point sharper still. LMG has flagged automated document upload and categorisation, a natural-language query tool for performance reports, fraud risk detection and churn prediction. Each of those moves a judgement further from the broker’s direct attention. The efficiency compounds. So does the distance between you and the reasoning on your own file, unless you deliberately close it.

6. A working policy for AI in a broking business

Most brokerages using these tools have no written position on them. That is worth fixing before ASIC’s report lands, not after.

1. Write down which tools are approved for which tasks. Note generation, quality assurance, document handling and client communication are different risk categories and should be treated separately.
2. Require a human-specific addition to every generated rationale. Make it a step in your process that at least one client-specific sentence is added before lodgement, and say so in your policy.
3. Sample your own files monthly. Pull five recent files at random and read the rationales side by side. If you cannot tell them apart, a reviewer will not be able to either.
4. Keep the client’s own words somewhere. Objectives and requirements captured verbatim during the meeting are the strongest antidote to generated boilerplate, because they cannot be produced from structured data.
5. Record what the tool does with client data. Where it is processed, who can access it, and what your obligations are if that changes. This matters more, not less, as tooling gets embedded.
6. Track first-time approval rate, not just time saved. Time saved is a private benefit that never appears in a client outcome. First-time approval rate is the measure that connects the tooling to something a client would notice.

7. What to watch next

• ASIC’s best interests duty report, expected in the final quarter of this calendar year, and whether it addresses AI-assisted file notes directly.
• Whether other aggregators publish comparable efficiency data — Connective has, Finsure has not, and AFG’s position is unverified.
• Lender data readiness — with only 3 per cent of lenders reporting fully AI-ready data systems, any improvement there is where turnaround gains will come from.
• First-time approval rates as a published metric, which would be a far more meaningful channel benchmark than time saved.
• Aggregator AI roadmaps moving into credit judgement — fraud risk detection and churn prediction sit closer to decisions than note writing does.