ASIC Had No Concerns With the Recorded Reasons on 74.1% of Files Where the Extra Cost Was Under $1,000. Above $20,000 It Was 34.1%

The number most of the coverage missed

ASIC Report 840 landed on 13 October, and the figure that travelled fastest was this one: seven of the eight aggregator groups reviewed reported that their brokers did not recommend the lowest-cost loan in more than 20% of cases over the two-year period.

It is a striking number, and it is also close to meaningless on its own. ASIC says so itself. The report notes that licensees were calculating loan costs in different ways — some over the full 25 to 30-year term, some over five years to reflect how long borrowers actually stay — and that in some cases the loan with the higher total cost was probably the cheaper loan over the short term, because of introductory rates that later reverted. Some aggregators reported on their credit representatives only; others included brokers operating under their own licence. It is not a like-for-like figure, and recommending something other than the cheapest loan is not a breach of anything.

The number that should change your Monday sits in Appendix 2.

ASIC took the files where the broker recommended a loan that was not the cheapest on their own shortlist, calculated how much extra that recommendation was projected to cost the client over five years, and then graded the rationale the broker had recorded. Sorted by the size of the cost gap, the results form a clean gradient in the wrong direction.

Put plainly: the files that needed the most justification carried the least.

What ASIC actually looked at

Scope matters here, both for how much weight to give the finding and for how not to overstate it.

ASIC reviewed 13 credit licensees making up eight large aggregator groups — together responsible for roughly two-thirds of all broker representatives in Australia. The groups named in the report are AFG, Connective, Finsure and Beagle, Lendi Group (Aussie), Loan Market Group, Mortgage Choice and Smartline, Outsource Financial, and Vow Financial and Yellow Brick Road. The data covered 1 April 2023 to 31 March 2025. ASIC also reviewed more than 7,000 documents, including policies, training material, file review methodologies and assurance reports.

The file-level work covered 1,600 loan recommendations, split into two deliberately targeted samples of 800: files where the recommended loan was not the cheapest of the shortlisted options, and files where the broker presented only one option at the end of the shortlisting process. ASIC is explicit that these were recommendations which, under RG 273 and the licensees’ own policies, brokers should be supporting with evidence.

What “concerns” meant in ASIC’s own words

The grades are not a judgement about whether the loan was wrong. They are a judgement about the record. ASIC set out its criteria in Appendix 1, and the distinction is worth knowing before anyone quotes the 65.9% figure at you:

• No concerns — the documented rationale supporting the recommended loan was clear and reasonable.
• Some concerns — the rationale was specific to the consumer’s circumstances, but could have better demonstrated how or why the recommended loan was in their best interests, even though it was not the cheapest.
• Most concerns — there was no documented rationale at all, or the rationale was generic, inconsistent or poor quality.

So the middle band is a quality gap, not a failure: the broker wrote something real about that client and stopped a sentence or two short. The top band is the one with teeth. On 43.9% of the files where the recommendation was projected to cost the client more than $20,000, ASIC found no recorded rationale, or one that was generic, inconsistent or poor quality.

Elsewhere in the report, ASIC lists the specific patterns that put brokers at risk of being unable to demonstrate compliance: failing to record reasons at all; generic or non-specific reasons; a lack of detailed cost comparison notes; documented reasons limited to the features of each product; a claim that the recommended loan was the lowest cost option when the compared data suggested otherwise; and reasoning that was inconsistent with the client’s stated preferences.

That fourth one catches a lot of otherwise careful files. “Client required an offset account” is a product feature. It is not a reason why this client, with this balance and this behaviour, is better off paying more for it.

Why the gradient probably exists

ASIC does not offer an explanation for the pattern, and neither should anyone else with certainty. But two mechanics in the report point at it.

The first is that nothing in the workflow was escalating by stakes. ASIC found licensees used exception reports that flagged when the lowest-cost option was not selected — a binary trigger. What was largely missing was the size of the gap. In the report’s words, there was “a general absence of metrics that estimated the difference in cost of the compared loans and the recommended loan.” Two of the eight groups could not readily extract total-cost data from their comparison software at all, leaving them no portfolio view of cost differences and no way to do risk-based monitoring on loan costs except by manual review.

If the system treats a $400 gap and a $40,000 gap identically, the broker gets no signal that this file is the one to slow down on. The note that satisfies a minimum character count passes either way.

The second is that large cost gaps tend to arise in the hardest files. A client with non-standard income, a tight servicing position, or a settlement deadline gets placed with the lender that can actually do the deal, and the cheapest lender on the panel often cannot. The reasoning in the broker’s head is usually sound. It is also the most complex to write down, and it gets written down at the end of the longest, messiest files.

The comparison-tool blind spot

One finding in the report deserves singling out, because it can make your file say something untrue without anyone intending it.

ASIC found that some licensees could provide interest rate data for the products compared, but were not always sure whether that rate was the one the client was actually expected to receive. The reason is mechanical: a broker has to manually enter any discount agreed with a lender for it to show up in the comparison tool — but the tool’s data does not need to be updated for the client to receive the discount.

So the client gets the sharper rate, and the tool keeps ranking on the undiscounted one. Your file then records a cost comparison that no longer matches the deal. ASIC’s action item for licensees is to ensure their systems capture accurate information relevant to loan recommendations, “including lender discounts, promotional offers, cashbacks and lender timeframes for processing loan applications.”

Until the systems do that, the manual entry is the control. If you negotiate a discount and do not feed it back into the tool, the record of your own reasoning is wrong in a way that is very hard to explain two years later.

Where AI now sits in the duty

The second thread in REP 840 is the one with the longer tail. ASIC used this report to state, for the first time in a broking-specific document, how it expects AI-assisted tools to sit inside the best interests duty.

The tone is not hostile. ASIC says it is “encouraged to see the credit industry and service providers exploring the development and rollout of tools that aim to identify better products and savings opportunities for consumers,” and that if adopted more widely such tools “may help licensees demonstrate how consumers are being provided with suitable products, including lower-cost options where appropriate, consistent with RG 273.55.” On policy-matching specifically, the report notes the potential for AI to compare lender policies to determine eligibility, observing that understanding and interpreting lender policy is “a core part of the service that mortgage brokers offer,” and that ASIC is interested in how such tools might help the industry “not only work smarter and faster, but also to more confidently and accurately identify loans that meet consumers’ needs.”

Then it draws the line. Pointing back to its earlier Report 798 on AI governance, ASIC writes:

Outputs from any decision tools or AI agents should be explainable and documented (including the relevant data used to create them).
ASIC, Report 840, ‘Governance and accountability’

And immediately after:

Brokers and licensees are still responsible for the quality of the credit assistance they provide, regardless of the technology they use.
ASIC, Report 840, ‘Governance and accountability’

Read those two sentences against the table at the top of this article and the practical consequence is clear enough. If a tool shortlists, ranks or filters the products you then recommend from, the tool’s output has become part of your reasoning — and ASIC expects that output, and the data behind it, to be explainable and on the file. A rationale that amounts to “the system ranked it first” is a generic reason wearing better clothes, and generic reasons are what put 43.9% of the big-gap files in the bottom grade.

The flip side is the opportunity. The same documentation burden that is being handled badly on complex files is exactly the burden a well-governed tool can carry — provided it records the inputs, the point-in-time data and the logic, rather than producing a tidy paragraph with nothing underneath it.

What RG 273 actually asks you to keep

None of the obligations changed on 13 October. The best interests duty sits in the National Credit Act — section 158LA for licensees, with section 158LE(1) placing the duty on representatives directly and section 158LE(2) requiring licensees to take reasonable steps to ensure their representatives comply. RG 273, published in June 2020, is still the guidance ASIC assesses you against, and its record-keeping section is more specific than most brokers remember.

RG 273.165 says ASIC generally expects records to include “the options and ultimate recommendation you gave and the reasons why (including a detailed description of your decision-making process).” RG 273.167 lists the forms those records can take, and explicitly includes “outputs of product and feature comparison tools, including graphical presentations and calculators” — which is where your software’s shortlist lives.

Two further paragraphs are worth re-reading in light of this report. RG 273.169 says taking notes contemporaneously “may help you to accurately capture your thinking and reasoning at each relevant point in time,” and that “just drafting notes and creating records at the end of the process may not be as effective.” RG 273.170 suggests a concise narrative summary connecting the various records — one that is “evidence based and supported by contemporaneous records.”

And on cost, RG 273.54 is the sentence the whole table turns on: “Any situation where a higher cost loan is recommended will need to be supported by evidence demonstrating why that recommendation is in the consumer’s best interests.” RG 273.56 is the balance — the product with the lowest interest rate is not necessarily the lowest cost option, and features such as an offset or redraw may save certain clients a considerable amount. The guidance has always allowed you to recommend the dearer loan. It has always asked you to show the working.

The rationale test: what to review this week

This is a report about records, so the useful response is a records exercise. It does not need a project.

1. Sort your own files by cost gap, not by exception flag

Pull the files from the last six months where you did not recommend the cheapest option on your shortlist, and rank them by the projected cost difference rather than treating them as one pile. ASIC’s own action item for licensees is to be able to “readily identify loan recommendations with limited or no recorded rationale and that are not for the cheapest loan compared.” Doing it at your own desk first is cheaper than being asked.

2. Apply ASIC’s three grades to your top five

Take the five largest gaps and grade them honestly against the criteria above. Is the rationale clear and reasonable? Is it specific to that client, or specific to the product? Would a reader who had never met the client understand why the extra cost bought them something worth having?

3. Convert every feature statement into a client statement

“Offset account required” becomes “client holds approximately $X in savings and is paid monthly, so the offset is expected to reduce interest by roughly $Y a year, which exceeds the $Z annual package fee.” Same decision, documented. That is the difference between the middle grade and the top one.

4. Check your discount hygiene

Where you negotiated pricing, confirm the discounted rate was entered back into the comparison tool. If it was not, the cost comparison sitting on your file does not describe the loan your client took.

5. Write down what your tools did

If an AI assistant, policy-matching engine or servicing calculator shaped the shortlist, record which tool, what you asked it, the date, and the client-specific data you gave it. Keep the output. ASIC’s expectation is that the output and the data behind it are explainable and documented — which is a filing habit, not a technology project.

6. Ask your aggregator one question

Can their system show you the projected total cost difference between the loan you recommended and the cheapest loan you compared? Two of the eight groups ASIC reviewed could not extract that data readily. If yours cannot, your own notes are carrying the entire evidentiary load, and you should know that before someone tells you.

What to watch next

ASIC says it expects mortgage broker licensees to consider the findings and close the gaps in their own practices, and that all licensees — not just the thirteen reviewed — should use the report to assess whether they need to change. In practice that means the controls will arrive through your aggregator: more mandatory fields, cost-gap thresholds that trigger a second look, and quality-assurance sampling weighted toward the expensive files rather than spread evenly.

Brokers who tighten the rationale on their own big-gap files now will find those controls a formality. Brokers who do not will meet them as friction at exactly the moment a complex deal is trying to settle.