Can AI Improve Financial Advice?
Australia has a wide advice gap and a shrinking pool of advisers, and people are already asking chatbots about money. Rules-based robo-advice has some peer-reviewed evidence of better portfolios. General chatbots give generic advice in published case studies, and the claim that grounding fixes this is largely untested with real clients. The duty to act in the client's best interests stays with the adviser and the licensee.
On this page
Better: Note and confidence
Low confidenceRobo-advice shows gains; chatbot advice still generic in placesOne peer-reviewed study found robo-advice adopters diversified more and showed fewer behavioural biases. Case studies of ChatGPT found generic recommendations that often overlooked priorities; a newer model was more thorough but still too broad on some topics. We found no peer-reviewed trial of grounded AI advice with real clients.Faster: Note and confidence
Low confidenceAnswers are instant; we found no measured adviser time savingsA model answers in seconds, but we found no independent study measuring how much time AI tools save Australian advisers in preparing advice.Cheaper: Note and confidence
Low confidenceDigital advice can cost less; uptake has been lowThe Quality of Advice Review was told that digital plans are available for a modest fee, but it found adoption slow, and only 1% of participants in a 2019 ASIC survey had used digital advice. We found no independent cost study.Safer: Note and confidence
Governance-gatedSafe only inside a licensee's tested, supervised systemASIC applies the same obligations to digital and human advice and expects algorithms to be tested, documented and reviewed. Moneysmart says AI platforms are generally not licensed to give personal advice.The problem, in numbers
Australia has a wide advice gap and a small advice workforce. The Quality of Advice Review, delivered to government in December 2022, put the number of financial advisers at around 16,000 and said their numbers were declining.Source 1 It cited a 2019 ASIC survey in which 27% of participants had ever received financial advice, and an Investment Trends estimate that 61% of consumers had unmet advice needs in 2021.Source 1 Both are second-hand figures, reported by the Review.
People are already filling the gap with general-purpose AI. Research commissioned by ASIC's Moneysmart found that 18% of Gen Z Australians use AI platforms for financial information and guidance, and 63% are confident in its accuracy; both are self-reported, and ASIC did not publish the sample.Source 4 Moneysmart's own guidance says AI platforms are generally not licensed to give personal financial advice.Source 4
This review asks whether AI makes advice better: more accurate, better suited to the person and more widely available. It looks at the advice industry, and it is not advice to readers. Turnaround in credit is covered in our review of loan decisions.
What improves, and why
Automated advice is not new. ASIC defines digital advice, also called robo-advice, as automated financial product advice given without the direct involvement of a human adviser, and its guidance dates from 2016.Source 3 What has changed is the interface. A large language model can take a question in plain words and answer in plain words, which a fixed questionnaire cannot.
In our reading, the gain depends on what the model is grounded in. A rules-based engine draws its recommendations from a documented algorithm that the licensee tests and can reconstruct; ASIC expects licensees to keep records of changes to their algorithms over the past seven years.Source 3 A general chatbot draws on whatever it absorbed in training. Grounding a model in the client's actual data, current product terms and the licensee's approved advice rules should narrow that gap, and an adviser who uses the tool to prepare rather than to decide keeps judgement where the law puts it.
The Review saw the potential. Its author wrote that she was convinced digital advice tools can make good quality financial advice widely available, and that a full financial plan and ongoing advice can already be had digitally, some providers charging a modest fee.Source 1 She also found adoption had been slow, for two reasons: the pace of regulatory reform, and the complexity and uncertainty of personal advice obligations.Source 1
A model can answer a money question in seconds. Whether the answer suits the person asking is a separate question, and the law asks it of the licensee, not the model.
What the evidence actually says
The strongest peer-reviewed evidence concerns rules-based robo-advice, not generative AI. A study in the Review of Financial Studies, of a robo-adviser that built portfolios around each investor's holdings and preferences, found that previously undiversified adopters ended up with less volatile, higher-return portfolios, and that adopters showed fewer behavioural biases such as trend chasing.Source 5 Investors who were already diversified traded more after adopting.Source 5 We read the abstract only.
For general chatbots the evidence is case-based and mixed. Finance researchers gave ChatGPT personal finance cases they had written themselves. In the published study, its recommendations were a suitable starting point but tended to be generic, often overlooking alternative solutions and the priority of recommendations.Source 6 In a peer-reviewed follow-up on the same 21 scenarios, a newer model gave more thorough suggestions and paid closer attention to tax, but still overlooked important details, stayed too broad on some topics and occasionally cited the law misleadingly.Source 6 The authors call it a supporting tool, not a replacement for professionals. We read both abstracts only, and these are small, expert-assessed case sets, not trials with clients.
The case for grounding is largely untested in advice itself. We found no peer-reviewed study measuring whether grounded AI tools improve the advice real clients receive, or their outcomes. Nor did we find independent measurement of AI tools that Australian advisers use to prepare advice. Confidence is low.
Where it falls short
Generic by default. The published cases found advice that overlooked alternatives and priorities; a newer model was more thorough but still too broad on some topics.Source 6
A thin fact-find. SEC staff observed robo-advisers relying primarily, if not solely, on online questionnaires, some with no way for clients to add context or for staff to ask follow-up questions.Source 7
Questions outside scope. ASIC expects digital advice to filter out clients whose needs fall outside what the tool considers, such as a mortgage.Source 3 In our reading, a conversational model invites exactly those questions.
Unlicensed by default. General AI platforms are generally not licensed to give personal advice.Source 4
Where governance decides the outcome
Australian law does not carve out a separate regime for machines. ASIC's guidance says the Corporations Act is technology neutral and the obligations for traditional and digital advice are the same.Source 3 It expects at least one person who understands the rationale, risks and rules behind the algorithms, testing before advice is first given and regularly after, a way to suspend advice when a detected error is likely to cause client loss, and a sample of advice reviewed by a suitably qualified person.Source 3 The Review recommended a "good advice" duty for all personal advice and, for financial advisers, a new fiduciary best interests duty without a safe harbour.Source 1 In December 2023 the government said it would introduce a modernised best interests duty for all providers of advice and remove the "safe harbour" steps.Source 2 In August 2026 it narrowed that plan: it now proposes to keep the existing duty and safe harbour steps and remove only the broadest step, which it sees as a barrier to scaled advice.Source 2 We found no bill introduced by the end of October 2026, so the current duty and safe harbour still apply. In the United States, SEC staff say robo-advisers carry the same fiduciary obligations as other advisers, including a duty to give only suitable advice.Source 7 In our reading, suitability and fiduciary duty sit with a licensee or adviser who can be held to them, and a model cannot hold either.
How to think about adoption
For licensees, the evidence supports narrow, grounded uses over open chat. Define the advice a tool may give, connect it to the client's verified data and to approved advice rules, and filter out anything beyond that scope to a human adviser. Test it before release and after every change, keep versions for seven years, and review a sample of real outputs. Measure advice quality and client outcomes before and after, not only time saved. Treat vendor accuracy claims as hypotheses to test, and remember that the duty to the client does not pass to the software.
BFCS.ai does not fabricate figures. This is an evidence review of AI in the advice industry, not financial advice. The peer-reviewed outcome evidence concerns rules-based robo-advice; the generative AI evidence is two small, expert-assessed case studies by the same authors, which we read as abstracts only. Survey figures from 2019 and 2021 are cited second-hand from the Quality of Advice Review. We found no peer-reviewed trial of grounded AI advice tools with real clients and no independent measure of adviser time saved.
Sources
- 1
Treasury, Michelle Levy, "Quality of Advice Review: Final Report," December 2022 (provided to Government 16 December 2022, released 8 February 2023). treasury.gov.au
(PDF: treasury.gov.au
Back to textWhat we took from it
"There are only around 16,000 financial advisers in Australia and their numbers are declining" (p. 70). Second-hand figures cited by the Review (p. 47): in 2019, ASIC found that 27 per cent of survey participants had previously received advice (ASIC Report 627); in 2021, Investment Trends estimated that 61 per cent of consumers had unmet advice needs; a 2019 ASIC report found that 1 per cent of participants had used digital advice services and 19 per cent said they were open to doing so. Chapter 10 (pp. 177 to 179): the author is "convinced that digital advice tools can make good quality financial advice widely available"; it is possible to get a full financial plan and ongoing advice digitally, some providers charging a modest advice fee; adoption in Australia has been slow for two reasons, the rate of regulatory reform and regulatory complexity; there is no need for regulation specific to digital advice. Recommendation 4 proposed a "good advice duty" for anyone giving personal advice; Recommendation 5 proposed replacing the existing best interests duty and related obligations with a new statutory best interests duty for financial advisers, "a true fiduciary duty" that "will not include a safe harbour" (p. 6).
- 2
Treasury, "Delivering Better Financial Outcomes: Detailed Overview," December 2023 (page dated 7 December 2023). treasury.gov.au
(PDF: treasury.gov.au
Later position: treasury.gov.au
(PDF: treasury.gov.au
Back to textWhat we took from it
The Government will introduce a modernised and flexible best interests duty which will apply to all providers of advice; the existing primary obligation to act in the best interests of the client and to prioritise the client's interests in the event of a conflict will remain at the core; the existing "safe harbour" steps will be removed; the requirement to provide appropriate advice will be retained (responds to Review recommendations 4 and 5).
Treasury, "Protecting Consumers and the Promise of Superannuation in an Evolving Financial Ecosystem," fact sheet, 19 August 2026, item 5.3: "Simplifying the Best Interests Duty reform, by maintaining the existing obligation and safe harbour steps, and removing only the broadest safe-harbour step that is a barrier to scaled advice." The accompanying ministerial release says the Government will continue consulting to progress legislation. As at 31 October 2026 we found no bill introduced for these changes; the existing duty and safe harbour remain law.
- 3
Australian Securities and Investments Commission, Regulatory Guide 255, "Providing digital financial product advice to retail clients," issued 30 August 2016 (PDF version dated 28 March 2022). asic.gov.au
(PDF: download.asic.gov.au
Back to textWhat we took from it
Digital advice (robo-advice) is automated financial product advice without the direct involvement of a human adviser; "The Corporations Act is technology neutral. The obligations that apply to the provision of traditional financial product advice and digital advice are the same" (RG 255.86); at least one person who understands the rationale, risks and rules behind the algorithms (RG 255.61); robust testing before advice is first provided and regularly after, records of changes to algorithms over the past seven years, and controls to suspend advice if an error is likely to cause client loss or a breach (RG 255.73); a sample of advice reviewed by a suitably qualified individual (RG 255.74); a robust triage or filtering process to filter out clients for whom the advice is not appropriate, for example a client wanting advice about repaying a mortgage when the tool does not consider mortgages (RG 255.100 to RG 255.102).
- 4
Australian Securities and Investments Commission, "Moneysmart publishes tips on using AI for financial issues," news item (no date shown on the page): asic.gov.au
Moneysmart (ASIC), "AI and money decisions" (last updated 17 August 2026): moneysmart.gov.au
Back to textWhat we took from it
research commissioned by Moneysmart found "nearly one in five Gen Z Australians (18%) are using AI platforms for financial information and guidance" and "nearly two thirds (63%) of GenZ are confident in the accuracy of financial guidance from AI platforms"; the page does not give the sample size, method or age range. Self-reported survey.
"AI platforms are generally not licensed to give personal financial advice"; AI tools can produce inaccurate or biased information; check information against trusted, independent sources before acting.
- 5
D'Acunto, F., Prabhala, N. & Rossi, A. G., "The Promises and Pitfalls of Robo-Advising," Review of Financial Studies 32(5), 2019, pp. 1983-2020, doi:10.1093/rfs/hhz014 (peer-reviewed; abstract only read, via RePEc IDEAS). ideas.repec.org
Back to textWhat we took from it
A robo-adviser building portfolios tailored to investors' holdings and preferences: adopters were similar to non-adopters in demographics but more active and with more assets under management; previously undiversified investors increased their stock holdings and ended up with lower volatility and higher returns; already diversified investors held fewer stocks, saw a small fall in volatility and traded more; all adopters showed less of the disposition effect, trend chasing and the rank effect.
- 6
Schlosky, M. T. T., Karadas, S. & Raskie, S., "ChatGPT, Help! I Am in Financial Trouble," Journal of Risk and Financial Management 17(6), 241, June 2024 (peer-reviewed journal; abstract only read, via RePEc IDEAS and Crossref; the publisher's site refused automated access). ideas.repec.org
Follow-up: doi.org
Back to textWhat we took from it
The authors wrote their own personal finance cases; ChatGPT "serves as a suitable starting point" but its recommendations tend to be generic and often overlook alternative solutions, viewpoints and the priority of recommendations.
Schlosky, M. T. T. & Raskie, S., "ChatGPT as a Financial Advisor: A Re-Examination," Journal of Risk and Financial Management 18(12), 664, published 23 November 2025, doi:10.3390/jrfm18120664 (abstract only, from the publisher's Crossref record; the full text was not opened). The authors revisited "the 21 personal finance scenarios" with ChatGPT-4o, using expert qualitative assessment: it often gave more thorough suggestions and paid closer attention to tax, though it still overlooked some important details; generalisations remained too broad on certain topics and legal references were occasionally misleading; ChatGPT-5 gave similar recommendations with better accuracy on the numerical problems; the authors call ChatGPT "a more useful supporting tool" but "not a replacement for financial professionals".
- 7
US Securities and Exchange Commission, Division of Investment Management, "IM Guidance Update No. 2017-02: Robo-Advisers," February 2017 (staff guidance). sec.gov
Back to textWhat we took from it
Robo-advisers, like all registered investment advisers, are subject to the substantive and fiduciary obligations of the Advisers Act; an adviser's fiduciary duty includes an obligation to act in the best interests of its clients and to provide only suitable investment advice; staff observed robo-advisers providing advice based primarily, if not solely, on online questionnaires, some not designed to let clients add context or to allow follow-up or clarifying questions.