AFG’s AI Will Pre-Fill the Fact-Find From Payslips Next Year. RG 273.167(f) Makes That Fact-Find a Record of Your Inquiries

What AFG actually announced

The tool is called AssistAI Document Intelligence. AFG announced it at AFG Next, its national conference, in Melbourne on 10 November 2026, and published its own release the following day. It sits inside BrokerEngine Plus, AFG’s CRM, and AFG says brokers will be able to access it through that platform in the new year.

On AFG’s description, the mechanics are straightforward. A client uploads supporting documents — AFG names payslips, residential history and financial information. The tool reads those documents, extracts the relevant details and writes them into fields in the client’s mortgage fact-find. It then runs automated verification across the uploaded paperwork to flag incomplete information and possible discrepancies between documents. The client can check and amend what has been pre-filled before the application is submitted. AFG says the broker reviews the information and progresses the application.

AFG’s chief digital officer, Sam McCready, framed it in terms most brokers will recognise. “AI isn’t going to replace mortgage brokers, but it will change how they work,” he said. Broker Daily, reporting the announcement on 11 November, quoted him saying the pilot had proven the tool saves significant time for both the broker and their clients. AFG has not published a figure for that time saving, and neither has the trade coverage, so the scale of the gain is the company’s characterisation rather than a measured number.

It is worth being clear about what is not in dispute here. Document reading and field population is a well-understood use of the technology, the gain is real, and the admin it removes is the part of the job nobody values. AFG also puts the human in the right place in its own description: the broker reviews, the broker progresses. The question is what that review has to produce.

The fact-find is already a compliance record

Most brokers think of the fact-find as the thing you fill in before the work starts. ASIC’s record-keeping guidance treats it as part of the work.

RG 273.167 lists the forms records may take, and paragraph (f) names “fact-finding documents used when making inquiries into the consumer’s circumstances.” That sits inside a broader expectation set at RG 273.162: ASIC expects brokers to keep records of how they have acted when providing credit assistance, including records of the inquiries they make into the consumer’s circumstances and the consideration, investigation and assessment of the products they recommend. RG 273.163 adds that ASIC expects those records to clearly show how the broker has complied.

The list at RG 273.165 is more specific again. Among the records ASIC generally expects are relevant conversations with the consumer, information showing that you acted in the consumer’s best interests, and — at paragraph (g) — the options and ultimate recommendation you gave and the reasons why, including a detailed description of your decision-making process.

None of this is new. RG 273 was published in June 2020 and has not been rewritten for tools that populate the form themselves. That is precisely why the paragraphs are worth reading now rather than after the rollout: they were drafted on an assumption about how a fact-find comes to be filled in, and that assumption is about to stop holding for a large part of the channel.

RG 273 does not tell you how to gather information

The first thing to say is that there is nothing in the guidance that makes automated pre-fill a problem. RG 273.33 is explicit: the best interests duty does not prescribe how you gather information about the consumer or their circumstances. RG 273.38 says there is no prescribed manner for making inquiries and that the process may be iterative. A broker who lets software read a payslip is not thereby in breach of anything.

What the guidance does say is that the information you start with is usually not enough. RG 273.35 states that the information initially disclosed by consumers will often be insufficient to identify their individual circumstances and do the analysis necessary to recommend products. RG 273.37 says that where instructions are unclear, or requests seem inconsistent with the client’s circumstances, you should make further inquiries. RG 273.39 adds that further inquiries are particularly important, and may need to be more detailed, where the client has a low level of financial capability or the structure sought is complex — ASIC’s own example contrasts a first home buyer with an experienced borrower refinancing an owner-occupied property.

Those paragraphs are about the gap between what a client hands over and what you need to know. A tool that closes the data-entry gap does not close that one. Separately, under the responsible lending obligations, RG 273.31 notes that brokers or their licensees are required to make reasonable inquiries about the consumer’s requirements and objectives and about their financial situation, pointing to sections 117, 130, 140 and 153 of the National Credit Act. Whether and how a particular workflow satisfies those sections is a question for your licensee or compliance adviser against the current text of the Act — this article is general information, not advice on your obligations.

What REP 840 found when it read the recorded reasons

On 13 October 2026, ASIC published Report 840, Mortgage brokers: Review of best interests duty and complaints handling practices. It is the closest thing the channel has to a map of where files are thin, and it landed four weeks before AFG’s announcement.

ASIC reviewed 13 credit licensees making up eight large aggregator groups, which it says are responsible for approximately two-thirds of all broker representatives. It looked at 1,600 loan recommendations: 800 where the recommended loan was not the cheapest of the shortlisted loans presented, and 800 where only one option was presented at the end of the shortlisting process. Across the aggregator groups, brokers presented only one option in an average of 6.8% of cases, though that ranged from just under 1% to just over 15%. Seven of the eight groups reported that, between 1 April 2023 and 31 March 2025, brokers did not recommend the lowest-cost loan in more than 20% of cases.

ASIC’s overall verdict was not an indictment. It found licensees generally had processes designed to support compliance, that their documented processes largely aligned with RG 273, and that they generally required brokers to document why the recommended loan was not the lowest-cost option. The problem was in the execution on individual files. ASIC lists what it saw where brokers were at greater risk of being unable to demonstrate compliance: failure to record reasons for the recommendation; generic or non-specific reasons; a lack of detailed cost comparison notes; documented reasons limited to the features of each product; and reasons apparently inconsistent with the client’s stated preference or requirements.

The sharpest number is what happened as the money got bigger. ASIC graded the adequacy of the recorded reasons against the projected extra cost of the recommended loan over five years. Where that gap was less than $1,000, it had no concerns with 74.1% of files. Where the gap was more than $20,000, it had no concerns with 34.1% — 22.0% drew some concerns and 43.9% drew most concerns. That top band is a small sample, 41 loans in all, and ASIC is careful to say a larger cost difference does not mean a loan was unsuitable. But the direction is unambiguous: the recorded reasoning got weaker exactly where it needed to be strongest.

One methodological detail matters for how brokers should read this. ASIC says it sought the data for the loans compared and the reasons recorded in the aggregators’ systems, and that it did not consider other information or documents that may have been held on file. Some licensees said as much in response — that there could be other information elsewhere in their records indicating the actual reasons. ASIC’s answer was that relying on periodic file reviews to catch thin reasons, rather than using data to monitor for them, was itself the weakness. The practical translation: what is in the structured fields is what gets looked at first.

The specific risk: completeness that reads as verification

Put the two documents side by side and the risk is narrow and specific. It is not that AI will fill in a payslip wrongly — discrepancy flagging is one of the things AFG says the tool does. It is that a fact-find which arrives complete, internally consistent and confirmed by the client looks, on a file, exactly like a fact-find produced by a broker who asked good questions. The artefact is the same. The inquiry behind it is not.

There is a related point in ASIC’s broader work. In Report 798, Beware the gap: Governance arrangements in the face of AI innovation, published on 29 October 2024, ASIC reviewed 624 AI use cases across 23 AFS and credit licensees and found that governance and risk management often lagged AI adoption. That report is addressed to licensees, not to individual brokers, and it predates the current generation of tools. But the shape of its finding is the shape of the thing worth avoiding here: the capability arrives, the process around it catches up later.

The other asymmetry worth noticing is who the pre-filled data passes through. In AFG’s described flow, the client can check and amend the pre-filled information before submitting. That is sensible product design. It also means a value in a field may have been produced by a model, reviewed by a borrower, and never discussed with you — and a borrower confirming a number is not the same event as a broker making a reasonable inquiry about it. Where the fact-find is the document carrying that inquiry, this is the join that needs a process behind it.

Seven things to settle before the rollout

The wider build, and what to watch

AFG’s announcement was not the only one of its kind in the same week. On 10 November, Bluestone Home Loans launched a proprietary loan origination platform built by its in-house technology team, with Australian Broker reporting that it cuts turnaround from roughly five days to two. Bluestone’s chief commercial officer, Tony MacRae, said the goal was to make it easier for brokers to do business with the lender, and that brokers can keep applying through their existing CRM platforms. Earlier in the month, AFG also put utility and telco referrals inside BrokerEngine Plus through its Compare & Connect integration.

The pattern across those three is the same: time is being taken out of the file. That is good for brokers, and in a market where mortgage demand is falling it is close to necessary. But every one of those minutes comes out of a step where a broker used to be doing something, and some of those steps were where the evidence got created.

Two things worth watching from here. The first is whether aggregators publish guidance on documenting AI-assisted fact-finds before the tools ship, or after. The second is whether ASIC says anything about it at all; REP 840 gave licensees actions on monitoring the quality of recorded reasons, and the obvious next question is what those reasons look like when the inputs are machine-generated.

Sources: AFG, “AFG launches AI document tool to cut broker admin and save time,” 11 November 2026; Broker Daily, “AFG announces new AI tool to slash broker admin,” 11 November 2026; ASIC Report 840, Mortgage brokers: Review of best interests duty and complaints handling practices, October 2026; ASIC Regulatory Guide 273, Mortgage brokers: Best interests duty, June 2020; ASIC Report 798, Beware the gap: Governance arrangements in the face of AI innovation, 29 October 2024; Australian Broker, “Bluestone launches proprietary loan origination platform,” 10 November 2026; Australian Broker, “Australian Finance Group automates utility referrals for brokers,” 5 November 2026.