AI for Financial Advisers in Australia: Keep the Advice Yours

Advisers adopted AI faster than anyone. The duty didn't move.

74% of Australian financial advice practices now use AI or plan to within the year — above the global average, and up from 45% a year earlier. The admin win is real, and most of it is going exactly where it should: file notes and meeting documentation, the paperwork that keeps advisers at the desk long after the last appointment. But ASIC's position hasn't shifted an inch. The tool can draft the file. It can't hold the duty. When the advice is wrong, the name on it is still yours.

Advisers adopted first and asked permission later

The adoption figures come from Adviser Ratings' 2025 Australian Financial Advice Landscape: 74% of practices are using AI or planning to — 59% already have it in the workflow, 15% are planning to add it — up from 45% in 2024. The Financial Planning Standards Board's survey of more than 6,200 planners across 24 territories puts the global figure at 64%. Australian advisers are out in front. And the top use, by a wide margin, is file notes and meeting documentation: 86% of the practices using AI point it there first.

Regulators saw the gap coming. In October 2024, ASIC published REP 798 — "Beware the gap" — after reviewing 23 licensees and 624 AI use cases. The finding, in plain terms: adoption is running ahead of the risk and compliance frameworks meant to govern it. ASIC didn't tell advisers to stop. It said the existing rules still apply — including the obligation to provide financial services "efficiently, honestly and fairly" — and that a model which can't explain how it reached a decision is a problem the licensee owns, not the vendor.

Where the hours actually are

For a typical advice practice, the cost isn't the advice — it's everything wrapped around it. The meeting that has to be written up. The file note accurate enough to defend three years later. The fact-find, the research, the first draft of a statement of advice. This is Admin Leverage territory: work that must happen, that a client never pays for directly, and that quietly stretches every adviser's day.

That's precisely where the 86% are pointing AI, and it's the right instinct. A model that drafts the file note from a meeting recording, or assembles the research and a first cut of an SOA, hands back hours an adviser can spend in front of clients — the only part of the day that grows revenue. The win is capacity, not headcount: the same adviser, seeing more clients, without the paperwork debt. If you're not sure how much of your week is going to work a tool could carry, map where AI already fits your practice.

The line ASIC won't let you cross

Here's where advisers get into trouble. The admin is fair game. The advice is not — or rather, the advice can be AI-assisted, but responsibility for it can't be handed to a model. The duty to act in the client's best interests, to give advice that's appropriate, to provide services "efficiently, honestly and fairly" — all of it attaches to the adviser and the licensee, not the software. Like mortgage brokers before them, advisers are learning that AI can draft the file but can't hold the duty. If a tool pulls the wrong figure into an SOA, favours a product because its training data did, or writes a file note that doesn't match what was said, the breach is yours.

Two things sharpen this. First, explainability: ASIC has flagged that AI can produce outcomes that are hard to explain, and "the computer said so" is not a defence an advice file can lean on. Second, disclosure: from 10 December 2026, Privacy Act changes require covered organisations to tell people, in their privacy policy, where personal information feeds an automated decision that significantly affects them. A practice running client data through AI tools needs that line in its policy before the date, not after.

What to do about it

Start where the 86% started, but deliberately. Put AI on the admin that never touches the recommendation first — file notes, meeting summaries, research collation, document assembly — and keep a human sign-off on anything that shapes the advice itself. Make the review real: the adviser reads the AI-drafted SOA and file note against the source, and the file records that they did. Check what the tool does with client data — where it's stored, whether it trains the vendor's model, whether that's disclosed — and update your privacy policy before 10 December. ASIC's REP 798 is free and worth reading, but it tells you the duties exist; it won't design the workflow that keeps your practice on the right side of them. That last stretch — from knowing the rule to running a process that proves you met it — is the work.

The advisers pulling ahead aren't the ones who adopted AI fastest. They're the ones who put it on the admin, kept the judgement human, and can show their working if ASIC ever asks. If you'd rather build that discipline in from the start than retrofit it after a complaint, a short conversation will map where AI fits your practice and where it can't.