Mortgage Broking and AI: Best Interests Duty File Notes Explained

An Australian mortgage broker can use AI to turn call notes, the fact find and a lender comparison into a draft best interests duty file note, then check and sign it. The draft must give specific reasons tied to the client's objectives, show the cost comparison and say why a dearer loan was recommended, if it was.

The reason to get this right is recent. In October 2026 ASIC published Report 840, a review of 13 credit licensees in eight large aggregator groups, and found that licensees generally had processes designed to support compliance, but that some brokers recommended significantly more costly loans without clearly documenting why they were in the consumer's best interests. File notes are where that gap shows, and AI makes either the good version or the bad version faster.

Follow me on Instagram@sagnikteaches

What the duty asks and what a file note has to prove

The best interests duty sits in the National Consumer Credit Protection Act. Report 840 summarises it this way: licensees must act in the best interests of consumers (section 158LA) and take reasonable steps to ensure their representatives do (section 158LE(2)), and representatives have the obligation themselves (section 158LE(1)). ASIC's Regulatory Guide 273 explains what the regulator looks for and what steps reduce the risk of non-compliance. Read the guide itself before you change your notes, because a summary is no substitute for the original.

Connect on LinkedInSagnik Bhattacharya

For a file note, the practical test is simple: could someone who was not on the call understand, from the file alone, why this loan, from this lender, with these features, suited this client better than the alternatives? Report 840 says licensees generally required brokers to document why the recommended loan was not the lowest-cost option. Where it fell down was in the detail. The review listed weaknesses such as failure to record reasons for the recommendation, generic or non-specific reasons, a lack of detailed cost comparison notes, and documented reasons limited to the features of each product. It also noted a recommendation rationale that was generic, inconsistent or poor quality.

Subscribe on YouTube@sagnikteaches

That list tells you what a good note contains:

• The client's stated objectives and circumstances, in their words where possible.
• The options you considered and the lowest-cost one available to them.
• The expected costs compared, with figures.
• The reason for any more costly recommendation, linked to a quantifiable benefit where there is one.
• What the client was told and what they decided.

Where AI helps and where it harms

AI is good at turning a messy record into a structured draft: a call transcript, your bullet points, the fact-find fields and the comparison output. It is poor at supplying the reason. Ask a model to "write a best interests file note" with nothing to go on and it will produce fluent boilerplate, exactly the "generic or non-specific reasons" the report criticises, plus the occasional invented detail. The rule is that the AI formats the facts you give it and the broker supplies the judgement.

ASIC's report also comments on AI directly. It notes that large licensees are increasingly adopting AI-assisted tools, that outputs from decision tools or AI agents should be explainable and documented, including the data used to create them, and that brokers and licensees are still responsible for the quality of the credit assistance they provide regardless of the technology they use. Those are statements for you to build on: a human decides, and the file shows how.

The inputs the AI needs, and what to leave out

Gather before drafting:

1. Client objectives in two or three sentences, such as "wants repayments under a set figure while a partner is on parental leave, and wants an offset account".
2. Circumstances that matter to the choice: income changes, other debts, plans for the next five years.
3. The options from your comparison tool: lender, product, rate, fees, features, repayment figure.
4. Your reasoning in rough bullets. This is the part you cannot skip.
5. What you told the client and what they decided.

Leave out identity document numbers, tax file numbers, full account numbers and the clients' full names where a first name or file ID will do. Most of the note does not need them. Use only an AI tool that your licensee or aggregator has approved, on terms that keep your content out of model training, and follow their rules on client data. Privacy law applies to the personal information in a loan file, and your aggregator will have a policy, so check it before you paste anything.

A prompt that produces a note you can defend

You are helping a mortgage broker draft a best interests duty file note.
Use ONLY the facts below. Do not add reasons, figures or client statements that are not given.
If a fact the note needs is missing, list it under MISSING instead of guessing.

Structure the note under these headings:
1. Client objectives and circumstances
2. Options considered (table: lender, product, rate, fees, key features, monthly repayment)
3. Lowest-cost option available to this client
4. Recommended loan and why it suits the client's objectives
5. Cost difference compared with the lowest-cost option, with figures
6. Why any additional cost is in the client's interests (quantify the benefit if possible)
7. What the client was told and decided

Facts:
[objectives] [circumstances] [options table] [broker's reasoning bullets] [client decision]

The line "list under MISSING" is the safeguard. It turns the model's tendency to plug gaps into a visible list for you to fill. The headings force the two items that Report 840 found weakest: the lowest-cost comparison and the reason for any dearer choice.

A worked example with numbers

An illustration, not a real case. A couple want to refinance a $600,000 balance over 30 years. Both work, one is changing to part-time in eight months, and they want an offset account and the option to make extra repayments. Your comparison shows three loans. The lowest-cost one is a basic variable loan at 6.19% with no offset. The one you prefer is at 6.29% with a 100 per cent offset account and no fee for extra repayments.

The arithmetic: at 6.19% the repayment is about $3,671 a month, and at 6.29% about $3,710, a difference of roughly $39 a month, $468 a year, or around $14,000 over the full 30 years if nothing else changed. Those are illustrative figures from a standard repayment formula, and the real difference will depend on fees, rate changes and how the offset is used.

Your bullets say: partner moving to part-time in eight months, expects to hold about $40,000 in savings in the account for most of the year, plans to use the offset rather than a separate savings account. The AI draft should then show the benefit as a sum: $40,000 held in an offset at 6.29% saves interest of about $2,516 a year in the loan (40,000 x 6.29%), against an extra cost of about $468 a year. That is a quantified reason, not "client wants flexibility".

A weak draft would read: "The recommended loan offers better flexibility and features suited to the client's needs." A strong draft reads: "The lowest-cost option is a basic variable loan at 6.19% (about $3,671 a month). Recommended loan: 6.29% with 100% offset (about $3,710 a month), $39 a month more. The clients expect to keep about $40,000 in savings in the offset, which would reduce interest by roughly $2,500 a year, more than the extra $468 a year. They also want fee-free extra repayments for the period of reduced income. The clients were shown both options and chose the recommended loan." You then check that the $40,000 came from the client, not from the model, because a wrong assumption here would undermine the whole note.

The review step: two minutes that matter

Treat the AI draft like a junior's first attempt. Before you sign it off, check these seven things, and tick them in the file:

1. Every figure matches the comparison output and the client's declared amounts.
2. Every client statement in quotation marks was actually said, per your notes or transcript.
3. The lowest-cost option shown really is the lowest-cost one available to this client, not just on your panel.
4. The reason for any more costly loan is specific to the client and quantified where possible.
5. Nothing in the note was added by the AI that you cannot support.
6. The note says what the client decided, and when.
7. The note will make sense to someone reading it in three years.

Edit freely. If the draft's style makes every note sound alike, that is a risk in itself: Report 840 criticised generic reasons, and a licensee reviewing 30 of your notes will notice identical phrasing. Change the opening of each, and delete filler such as "as discussed" when it hides what was discussed.

Where the credit guide and the file note meet

Your credit guide, the disclosure a client gets from you early on, typically sets out your licence details, how you are paid, the lenders you use most and how to complain. The file note should be consistent with it. If you recommend a lender where you earn a different commission to another suitable lender, the note should show that you considered the client's interests first. The review found that brokers generally received similar commissions on standard home loans and did not appear to favour lenders that paid more, but that data on commissions and clawbacks was not always easy to extract. An AI step can help here: ask it to compare the commission details you paste in against the chosen loan and flag any case where the recommended loan pays more than the cheapest alternative, so that you write a stronger note for those files. It cannot decide the answer. It can only make the file harder to overlook.

Using a call transcript as source material

Many brokers now record or transcribe client calls through a meeting tool. The transcript is raw material, not the file note. A one-hour call produces 8,000 words, most of it chat, and a note that pastes the lot into a field proves nothing. A better route is two passes. First, ask the assistant to extract only the statements that bear on the client's objectives, circumstances and preferences, each with a timestamp. Second, you add your reasoning and generate the note from the extract.

From the transcript below, list every statement by the client about: (a) what they want from the loan,
(b) income, expenses or changes expected, (c) features they asked about or rejected, (d) any concern.
Give the speaker, the approximate time stamp and the exact words. Do not summarise or interpret.
If nothing is said about a topic, write NOT DISCUSSED.

A realistic extract line: "00:14, client: 'We don't want to be locked in if rates fall, we'd rather pay a bit more to be able to move.'" That sentence is gold in a note, because it is the client's own reason for choosing flexibility over the cheapest rate. Because the extract is quoted, you can check it against the recording in seconds. Before recording any call, make sure the client knows and agrees, and that your recording and storage arrangements meet your licensee's policy. The risks of listening tools are covered in whether it is safe to let AI listen to mortgage advice calls, and the choices for the transcript tools themselves in AI note takers for financial advisers.

A second example: a first home buyer with a gap in income history

A younger client is buying a first home and has been self-employed for fourteen months. The broker's comparison shows that two lenders will consider the application on the current income evidence, and one major lender would want two years. The lowest-cost option on paper is the major lender's loan, but the client would not qualify for it for another ten months. A weak note would say "recommended Lender B as the client's circumstances suit its policy." A strong one says: "Lender A has the lowest rate (illustrative 6.14%) but requires two years of trading history, which the client does not have until [month]. Waiting would risk losing the property under contract. Lender B accepts 12 months of trading history with accountant's letter at 6.34%. Cost difference about $55 a month on the $425,000 loan over 30 years. The client was told that a refinance to a lower-cost lender may be possible once they have two years' history, and agreed to a review at that point." Every sentence is checkable. The AI can draft the skeleton in a minute, but only you know the contract timeline and the refinance discussion.

The advice in that note, to review in ten months, is also an action. Ask the assistant to add it to a follow-up list with a date, so the promise in the note becomes a task in your CRM. That connects the file record to the client's later experience, which is where complaints often start.

A quick checklist for the licensee conversation

If you work under an aggregator or a licensee, send them these five questions before you roll out AI drafting. Their answers decide what you can do.

1. Which AI tools are approved for client data, and on which plan?
2. What client data may be pasted into them, and what may not?
3. Are AI-drafted notes acceptable, and must the file show that AI was used?
4. How long must the transcript, the draft and the final note be kept?
5. Who reviews a sample of notes, and what do they look for?

Write the answers into your own procedures, date them and update them when the licensee's policy changes. Then your notes, your process and your records say the same thing.

Realistic mistakes and how they show up

The confident invented reason. The draft says "the client prioritised speed of settlement", which nobody said. It shows up when you search your notes for the phrase and find nothing. Fix: the "use only these facts" line, and the review step.

The copied template. Twenty notes in a row with the same sentence about "flexibility". It shows up in a licensee's file review. Fix: require the quantified benefit line in the prompt and vary the opening.

The missing alternative. The note compares your recommended loan with only the loan the client first mentioned. It shows up when a reviewer asks about the lowest-cost option. Fix: heading 3, filled from the comparison tool and not from memory.

The client data leak. A full bank statement is pasted into a consumer chatbot. It shows up in an audit, or never, which is worse. Fix: an approved tool and an extraction step that leaves out the data you do not need.

The sign-off that is not one. Notes are saved unread because "the AI wrote it". It shows up when a client complaint prompts a file review and the note contradicts the facts. Fix: the seven-point check, recorded in the file.

Fitting this into the rest of your process

File notes are the end of a longer chain: enquiry, fact find, comparison, recommendation, application. If the earlier steps are tidy, the note almost writes itself. An AI workflow from enquiry to fact find for mortgage brokers covers the first half, the best AI tools for mortgage brokers at each stage of a case shows where tools fit, and what brokers can automate with AI and what stays advice draws the line this tutorial relies on. If your CRM can hold the draft and the approved note together with an audit trail, so much the better; choosing a mortgage CRM with AI built in has the questions to ask.

Before you change your process, speak to your aggregator or licensee about their expectations for AI-assisted notes, because they are the ones the regulator will examine. The review noted that licensees are investigating AI-assisted tools for compliance, including checks on file quality, and expects governance to keep pace. That means your approach may soon be reviewed by someone using similar tools, which is a good reason to make the notes specific, honest and checked.

A short routine you can start this week

1. Pick your next three completed files and rewrite the recommendation note with the prompt above, then compare with what you wrote originally.
2. Count how many of the original notes quantified the cost difference and the benefit. If it is under all three, you have found the improvement.
3. Agree the approved tool and the data rules with your licensee in writing.
4. Save the prompt where the whole office can use it, and review it when RG 273 or your licensee's policy changes.

None of this makes the duty lighter. It makes the record better and faster to produce, provided the thinking in it is yours. This is general information and not legal or compliance advice, so confirm the specifics with your licensee or a qualified adviser.

Questions brokers ask about AI and file notes

Can AI write the whole file note for me?

It can draft one from the facts you give it, but the reasoning must be yours. A note that gives generic reasons is the weakness regulators have pointed to. Treat the AI draft as a typing aid, then check every figure and reason against the case before you save it.

Is a transcript enough as a file note?

No. A transcript records what was said, not why the recommendation was in the client's interests. The note must connect the client's objectives and circumstances to the loan chosen and, where it costs more than another option, explain why. Use the transcript as source material for the note.

Can I paste client documents into a chatbot?

Only into a tool your licensee or aggregator has approved, on terms that keep data out of model training. Do not paste identity documents, tax file numbers or full bank statements into a consumer chatbot. Where unsure, give the AI the extracted facts instead of the documents.

Does using AI change my best interests duty?

No. ASIC's October 2026 review said brokers and licensees remain responsible for the quality of credit assistance regardless of the technology used. If AI drafts something wrong and you sign off, the responsibility is yours and your licensee's.

Further reads

• Can Financial Advisers Use ChatGPT? What Compliance Allows — What compliance allows when advisers use a chatbot.
• AI CRM for Financial Advisers: What Is Built In and What Is Bolted On — What a CRM does natively and what needs add-ons.
• Can AI Keep Mortgage Clients Updated During an Application? — Keep clients updated mid-application with AI help.
• Using AI to Spot Remortgage Opportunities in Your Client Bank — Find review opportunities in your existing client base.
• AI Answering Service for Mortgage Brokers: Cost and What It May Say — What an AI answering service may and may not say.
• AI Tool Approval Process: How Staff Request a New AI Tool — A model for agreeing an approved AI tool with your licensee.
• Can AI Write File Notes From Adviser Meetings? Draft, Check, Sign Off — Draft a structured file note from a meeting transcript, then make adviser sign-off a named step with consent wording, a checklist and a retention rule.
• Can Claude Read Payslips and Bank Statements for a Mortgage Case? — A broker's document-extraction workflow with source pages, salary-credit tie-outs and exception review.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ASIC Report 840, Mortgage brokers: Review of best interests duty and complaints handling practices (published 13 October 2026, read 4 November 2026); ASIC Regulatory Guide 273 as cited in that report; National Consumer Credit Protection Act 2009 sections as cited in the report; a published broker credit guide for typical contents. This is general information, not legal advice; confirm requirements with your licensee.