AI can safely help Australian financial advice practices under the Delivering Better Financial Outcomes (DBFO) reforms by drafting advice documents, structuring fact finds, summarising research and preparing reviews. It must not form recommendations. The best interests duty in section 961B of the Corporations Act belongs to the relevant provider, so the adviser forms the advice and a person reviews every figure.
Advice practices face two pressures at once. The cost of compliant advice keeps rising, driven by the time it takes to document personal advice. Meanwhile DBFO is rewriting the rules for that documentation, under the eye of the Australian Securities and Investments Commission.
This guide is for principals, responsible managers and compliance leads at licensees and multi-adviser advice groups who want to use AI seriously without loosening the obligations under every client file.
Key takeaways
- Use AI for drafting, never deciding. The recommendation must originate with the relevant provider.
- DBFO tranche two (replacing the SOA with a client advice record) is not settled, so build around adviser reasoning and data, not a fixed template.
- Every figure in an AI draft must trace to verified client data before it goes out.
- Advice files are highly sensitive: keep client data out of consumer AI tools and on infrastructure you control.
- Put a licensee-wide governance layer in place before AI moves from experiment to standard practice.
What problem are the DBFO reforms trying to solve?
Producing personal advice is slow, and the Statement of Advice (SOA), required by section 946A of the Corporations Act 2001, has become the symbol of that. Years of defensive drafting turned a communication tool into a long compliance document few clients read in full.
The Quality of Advice Review, led by Michelle Levy and delivered to Government in late 2022, found the system had drifted toward documents that protect licensees more than they inform clients. The Government's response became DBFO, aiming to make quality advice more accessible and affordable while keeping consumer protections.
DBFO has been delivered in stages:
- Tranche one, through the Treasury Laws Amendment (Delivering Better Financial Outcomes and Other Measures) Act 2024, streamlined ongoing fee arrangements and consent, gave more flexibility on the Financial Services Guide, and tidied several conflicted remuneration provisions.
- Tranche two, released in draft during 2025, contains the change that matters most for documentation: replacing the SOA with a more flexible client advice record, plus clearer rules on how super funds deliver and charge for limited advice.
Tranche two is not settled. Public reporting through 2025 and 2026 indicates the timetable has slipped, and ASIC has said it will keep issuing guidance as reforms land. Build AI capability that improves advice production today and can flex to a lighter standard later.
Which part of the advice lifecycle consumes the most adviser and paraplanner time?
- Drafting the advice document itselfAI drafts from file notes and data, adviser owns the reasoning
- Fact find and data collection from clientsAI structures intake, humans verify accuracy
- Research and product comparisonAI summarises, licensed adviser makes the recommendation
- Ongoing service and review schedulingAI tracks obligations and drafts review agendas
- File review and compliance monitoringAI flags gaps for a human compliance decision
What can AI safely do inside an advice workflow?
The most valuable role is drafting. Personal advice carries the best interests duty (section 961B) and the obligation to give appropriate advice. Those duties belong to a person and cannot be delegated to a model. What can be delegated is turning a completed fact find, file notes and the adviser's reasoning into a clear document.
The adviser records their reasoning during or straight after the meeting. A model assembles a first draft reflecting that reasoning, summarises the client's circumstances and writes plain-English explanations. The paraplanner or adviser reviews, corrects and owns the final content.
- CaptureAdviser records reasoning and client goals in the meeting
- DraftAI assembles a first-pass advice document from notes and data
- ReviewAdviser verifies facts, strategy and appropriateness
- ApproveRelevant provider signs off and owns the advice
Four repeatable jobs suit this pattern:
- Advice document drafting. The model works from your inputs, not the open internet, so the recommendation originates with the adviser.
- Fact find structuring. Converting scanned statements and email threads into a clean summary for the adviser to confirm. It should flag what is missing, never invent it.
- Research synthesis. Summarising product disclosure statements and drafting the comparison narrative against stated needs. The adviser still selects.
- Review preparation. Drafting review agendas, flagging clients due for annual review and pre-filling updated circumstances, which keeps ongoing fee arrangements defensible.
For example, an advice group with 12 advisers each producing four to six pieces of personal advice a month could lift capacity meaningfully if drafting time falls, provided human review stays genuine. Measure it on your own files.
- Adviser and paraplanner hours per advice document (baseline)Measure it
- Hours per document with AI drafting plus full reviewMeasure it
- Material errors caught in reviewLog every one
- Capacity redirected to client-facing workThe real result
Where does AI add licensee risk, and how do you contain it?
Fabricated facts and figures. Models can produce confident text with numbers or projections that were never in the source. In an advice document, an invented fee or balance can breach best interests and appropriate advice obligations. Every figure must trace to a verified source.
Advice drifting from the adviser's intent. Ask a model to recommend and it will produce a plausible recommendation no licensed person formed. That is a document dressed as advice. The control: the recommendation and reasoning must come from the relevant provider and be recorded before drafting.
Privacy and data location. Advice files are dense with personal and financial information governed by the Privacy Act 1988 and the Australian Privacy Principles. Feeding them into consumer AI tools that may retain or train on inputs is a serious risk. See how the wrong AI tools leak business data.
Tax advice boundaries. Advisers providing tax (financial) advice services operate under the Tax Practitioners Board Code of Professional Conduct. AI content that strays into unauthorised tax positions or misstates an outcome creates a separate compliance problem.
| Risky | Defensible | Impact | |
|---|---|---|---|
| Who forms the recommendation | The model suggests it | The relevant provider | Duty stays human |
| Source of figures | Model-generated | Verified from client data | No fabrication |
| Client data handling | Consumer tool, unclear retention | Controlled, no training on inputs | Privacy protected |
| Final document | Sent largely as drafted | Reviewed and owned by adviser | Accountability clear |
AI handles drafting; a qualified person keeps the decision and the accountability. Our framework for AI agent governance, data access and human override translates directly to advice.
What governance does a licensee need before scaling AI?
One adviser experimenting is manageable. Twelve advisers using their own tools with no shared controls is how client data ends up scattered across consumer platforms.
- Weeks 1-3AssessMap where time is lost and where client data currently flows
- Weeks 4-8ControlChoose tools with proper data handling, write the usage policy
- Weeks 9-14PilotRun AI-assisted drafting with a small adviser group and tight review
- Weeks 15+ScaleRoll out with training, audit trails and ongoing monitoring
The governance layer has four parts:
- A written AI usage policy, including the hard rule that recommendations originate with the adviser.
- Controlled tools that do not retain or train on client data, running on infrastructure in Australia you control.
- An audit trail of what the AI drafted and what the human changed, so file reviews can show judgement was exercised.
- Training, because the biggest practical risk is an adviser trusting a fluent draft without checking the figures.
Accounting firms face close parallels under their own codes, covered in our guide to AI in accounting firms and tax compliance.
How do the DBFO reforms change the calculation?
The reforms do not change who owns the advice. They change the documentation standard. If tranche two lands as drafted, a flexible client advice record rewards practices that produce clear, tailored documents quickly, which an AI drafting layer supports.
Build around your reasoning and data, not a fixed template. A practice that separates captured adviser reasoning, a controlled drafting layer and human review is ready for whatever final shape the record takes. One that hard-wires AI to a specific SOA template will rebuild when the standard moves.
Every advice document is also a set of representations to a client, so Australian Consumer Law misleading conduct provisions apply alongside the Corporations Act. See how consumer law applies to AI implementations.
Where should an advice practice start?
Target the drafting bottleneck with tight guardrails and leave reasoning entirely with advisers. Start with a small group, insist on line-by-line review, log what the AI produced and what changed, and check whether time saved holds once genuine review is included. If review keeps catching material errors, you have learned that before it reached a client file.
Related obligations include Privacy Act compliance for AI and APRA and ASIC expectations on financial services technology.
How Solve8 can help
Our consultants and engineers build controlled advice drafting and fact-find tools that run on your infrastructure in Australia, keep an audit trail of every change, and leave the recommendation with your advisers. See our financial services work and custom AI development service, or contact us to scope a pilot.
