The AI question is coming. It won't have ASIC's letterhead on it.

Six years of compliance work fixed the floor. Nothing has touched the ceiling.

Assured Support has published its Quality of Advice Index, built on 9,315 advice files reviewed between 1 July 2020 and 30 June 2026. Poor and very poor advice, taken together, fell from 20.9 per cent to 10 per cent. Very poor on its own dropped from 2.4 per cent to 0.3 per cent. Sound advice climbed 11.2 points, from 77 per cent to 88.2 per cent. That's a real achievement and it happened through a royal commission, an education standard and a decade of legislative churn. Worth saying out loud, because the profession rarely gets told it did something well.

Then there's the number nobody quotes. Advice rated ‘good’ sat at 1.8 per cent of files in both 2024-25 and 2025-26. ‘Exceptional’ never got above 0.2 per cent in the whole six years. Six years of investment bought a much higher floor and did nothing at all to the ceiling, and Assured Support says so themselves: raising the floor and raising the ceiling are two different stages, and only one of them has happened. Now hold that next to what's being sold into your practice at the moment. Fewer errors, better file notes, cleaner records, nothing missed. Every bit of that is floor work, and the floor is largely done. The other use for this index is as a baseline: next time somebody tells you AI-drafted advice is risky, the fair question is risky compared to what, and this is the honest answer to that question rather than the one most people assume.

One caveat, and Assured Support raise it themselves rather than burying it. These are files from licensees and practices that chose to commission an independent review, which is not a random sample of the profession. Read it as their index of a self-selecting group, not as a measure of Australian advice.

Vanguard paid around US$4 billion, and almost none of it was for the AI

Vanguard has agreed to buy Altruist, a US adviser technology and custody platform it describes as AI-forward, founded in 2018 by Jason Wenk and one Vanguard had already invested in since 2020. Terms weren't disclosed and the reported figure is around US$4 billion. It's expected to close later this year subject to regulatory approvals. This is a US deal with no Australian advice offering attached and no local timetable, so it changes nothing you have to do.

It tells you something useful anyway, and it's about price. Language models are close to free and getting cheaper every quarter. So four billion dollars didn't buy intelligence. It bought custody, account opening, trading, rebalancing, billing, reporting and the adviser relationships that sit on top of them, which is to say the pipes and the data running through them. The AI is only worth anything because it can see all of that. That rule scales down to a three-person practice without changing shape: every AI tool that has genuinely saved a practice hours had reach into real data, and every one that disappointed was a clever model looking at nothing. So the question to put to a vendor isn't how good is your AI. It's what can it see, and who has to do the work to let it see that. The second answer is usually the one that decides whether the thing ever gets used. There's a bonus in there too. Asked why it didn't just build its own, Vanguard said replicating Altruist internally would take “significant time and investment”. Worth reading twice if you've been meaning to have a crack at your own tooling over a long weekend.

There's a new job at the front of every meeting, and nobody has priced it

Geoff Brailey, a research director at McCrindle, described the adviser's job now as partly to “almost tease out and disentangle what's unhelpful financial information that they might have received from an AI platform that may not know the latest regulations and policies”. In the same piece, ASIC's own Corporate Plan records Moneysmart traffic for super, insurance and retirement planning falling to 9.1 million visits in 2025-26, down from more than 11 million. People haven't stopped looking things up. They've stopped looking them up there.

The part worth sitting with is the asymmetry underneath it. McCrindle also found 55 per cent of Australians are wary of putting too much personal information into an AI platform, which is a sensible instinct and also the whole problem. The answer your client walks in holding was built without the facts that would have changed it. Wrong would be easier. What they've got is confidently generic, and unwinding a confident wrong answer takes longer than answering a blank question, so the work went up rather than down, and it landed at the front of the meeting where nothing in your fee model, your meeting template or your capacity numbers accounts for it. No tool fixes that. What does is a standing opening question about what they've already read or been told, asked before you start, so you find out in minute two instead of minute forty. Issue 04 covered clients using AI as a first draft and then going somewhere else to verify it. This is what it costs you when ‘somewhere else’ turns out to be your meeting room.