Services Automations Our Work Bax OS About AI-Readiness Check Common Questions The Advisers AI Brief Book a Call

Issue 02 · 21 July 2026

Most compliance findings aren't misconduct. They're mess.

Why the fortnight's most useful sentence about AI in advice is really an argument for fixing your process before you buy anything.

Julia Vojkovic from 3Lines wrote the most useful sentence published about AI in advice this fortnight, and I don't think it'll be read the way she meant it.

After years of reviewing advice files, she says most of what turns up isn't a major breach or deliberate misconduct. It's process gaps. An overlooked disclosure, a template nobody updated, a strategy that isn't quite clear on the page. Not bad intent, not bad knowledge. Time pressure.

Fair enough, honestly. I've been on that side of the desk.

Here's where I'd part ways with how this gets read. The obvious move is to go shopping for an AI compliance tool. I'd do the opposite first. If the process itself changes depending on who's running it that week, then AI checking your files just finds the same gaps faster, and now you're paying a subscription to be told what you already suspected.

The order that works is boring. Write down how the job actually gets done today, warts and all. Fix the two or three steps that vary depending on who's on them. Then automate what's left.

That's why the first deliverable in my own assessment isn't a tech recommendation. It's the practice's own process, written up plainly, every inconsistent step named and given a fix they can do themselves. No tool, no licence. The automation conversation gets shorter after that, and cheaper. Odd thing for me to be arguing, and I'd still rather argue it than sell someone a tool that automates their mess.


If a client asks an AI who to talk to, are you in the answer?

Sean Graham, Assured Support, in Professional Planner, 17 July 2026

Graham's argument is that AI assistants have quietly turned into a referral channel, so being absent from those answers is now a commercial choice rather than a neutral one. He points at Similarweb's 2026 finding that users who received a brand recommendation from ChatGPT were 2.5 times more likely to visit that brand's website within seven days, and once there viewed nearly twice as many pages and stayed about twice as long. His answer isn't publish everything or publish nothing. It's what he calls governed authority: separating public expertise from regulated advice, so a practice can be visible without the licensee wincing.

This is the most under-rated item of the fortnight for small practices, mostly because the fix is so unglamorous. An AI can only put you in an answer if something of yours exists, publicly, that clearly isn't personal advice. Plenty of practices have neither the material nor that separation, and some of the ones that do have it locked inside a client portal where nothing can see it. If you do one thing with this, take the three questions you answer in every single first meeting and write your answers up publicly, in your own words, with your name on them. That's the whole play, and it's the cheapest marketing you'll do this year.

17% of Australians use AI for money decisions. The number that matters is 59%.

AustralianSuper, research by Quantum Market Research, published 20 July 2026 (n=1,009, nationally weighted, ±3.1% at 95% confidence, fielded April and May 2026)

Seventeen per cent of Australians now use AI to inform financial decisions, rising to 28 per cent of 18 to 29 year olds and falling to 6 per cent of the over-60s. Thirty-seven per cent say they don't trust AI for financial information, and of that group, 59 per cent put it down to not knowing where the information came from. A quarter of AI users have acted on what it told them without checking anything else, though most do check: 54 per cent cross-reference other sources and 32 per cent take it to a financial professional.

The headline number is the 17 per cent and it's the least interesting one here. Sit with the 59 instead. The objection isn't "AI gets it wrong", it's "I can't see where this came from". That's a provenance problem, and it's the same one waiting inside your own practice: whatever you end up running, if you can't show where an output came from, you'll lose that argument twice, once with a client and once with your licensee. The 32 per cent is worth a second look too. A third of these people are carrying AI's answer to a professional, which makes this top-of-funnel behaviour rather than a substitute for you. They're turning up better briefed and more sceptical. That's a different meeting, not a lost one.

Your advised clients use AI more than your unadvised prospects do

CoreData and Conexus Financial, Best Possible Retirement 2026, reported by Simon Hoyle in Professional Planner, 16 July 2026 (n=3,481 Australians aged 45 and over, fielded 19 March to 1 May 2026)

Thirty-four per cent of pre-retirees and retirees have used a generative AI tool, though that average hides an enormous spread: 49.6 per cent of pre-retirees against 20.7 per cent of retirees, and 55.9 per cent of 45 to 49 year olds against 13.5 per cent of the over-70s. The number that stopped me was the advised comparison. People who already have an adviser use AI more than people who don't, 38.1 per cent against 32.9 per cent. Same pattern on fund tools, 47.7 against 33.1. And nearly a quarter of AI users are using it to prepare questions for their adviser.

The fear here is obvious and, on this data, backwards. If AI were eating advice you'd expect the unadvised to be the heavy users, quietly self-serving instead of booking a meeting. It's the opposite. The people paying for advice are the same people curious enough to go looking, which fits everything else we know about engaged clients. They're doing homework before they see you, not shopping for a replacement. Practically, that changes the meeting rather than threatening it: expect better questions, expect the occasional confidently wrong thing picked up somewhere, and be ready to say where your own answer comes from. A client who has already asked a machine will notice whether you can show your working.

That's the fortnight. If you reckon I've read any of it wrong, hit reply and tell me. I'd rather be corrected than agreed with. Past issues live here.

Issue 02 covers the fortnight to 21 July 2026 and was published to this archive on 2 August 2026, when the Brief launched. It was not emailed at the time.

Get It In Your Inbox

Don't want to miss the next one?

One email a fortnight. Three or four things worth knowing, read in five minutes. Unsubscribe whenever you like.

We'll email you a confirmation link first, so nobody ends up on the list by accident. Your address is used for the Brief and nothing else.