MYOB + AI — what's coming, what to test, what to ignore

If you run a small NZ business, there's a decent chance MYOB is somewhere in your stack. Maybe it's the books, maybe it's payroll, maybe it's the whole back office. And in the last few months MYOB has signed a five-year partnership with Microsoft to bring AI features into the platform. That's a real shift, and most owners I talk to either haven't heard about it or assume it means nothing changes for them.

Both reactions miss the practical point. The interesting part of "AI in MYOB" isn't the press release. It's that AI is showing up in software you already use, which means the adoption question stops being "should we pick a new AI tool?" and starts being "which of our current MYOB workflows should be ready for it first?"

This post is the honest version of what I'd tell a small NZ owner sitting across the table. What's actually useful right now. What to test on next month's run. What's still oversold. What I'd wait on.

What's actually changing

The headline is the partnership: MYOB and Microsoft signed a five-year deal that puts Microsoft's AI capabilities (the Copilot stack and underlying models) inside MYOB's products. In practice, that means the AI features won't sit in a separate app you have to log in to. They show up where invoicing, payroll, reporting, and account management already happen.

The reason that matters for a small team is simple: the biggest reason AI projects fail in small NZ businesses is that nobody actually adopts the new tool. Adding another login is a tax. Adding capability inside the tool people already open every morning is a much shorter path to value.

But — and this is the part that gets glossed over in the announcement coverage — "AI in MYOB" doesn't mean every feature is good yet. Some of it is genuinely useful. Some of it is generic AI assistance bolted onto a screen. Knowing which is which is the difference between saving time and adding noise.

What to test on next month's run

The features I'd actually try in the next 30 days, in priority order:

Bank feed categorisation suggestions. This is the one that's been improving steadily across all the cloud accounting tools, and it's where the time savings show up first. Set the confidence threshold conservatively, let it auto-categorise the obvious recurring transactions, and review the rest. Within two weeks you should see your weekly reconciliation time drop noticeably. If it doesn't, the rule set needs work — not the tool.

Invoice and quote draft assistance. AI is reasonably good at drafting the boring parts of an invoice or quote — line item descriptions, payment terms, follow-up wording. Test it on a handful of routine jobs first. Don't test it on a complex variation or a sensitive client. Get a feel for what kind of editing you have to do before trusting it on anything that goes out under your name.

Follow-up reminders for overdue invoices. Less about generating the message, more about catching the trigger reliably. If MYOB's automated reminder rules now do a better job of catching overdue invoices and prompting a follow-up, that's worth setting up properly. The bottleneck for most small teams isn't writing the reminder — it's noticing the invoice is overdue.

Report summarisation. Asking a natural-language question against your own data ("which clients haven't been invoiced this month", "what's my average days-to-pay this quarter") is genuinely useful when it works. Test the answers against what you can verify manually. If the numbers line up, this becomes a real time-saver. If they don't, stop using it for anything you'd act on.

What I'd be skeptical of

A few things that are getting marketed inside accounting AI right now that I'd treat with caution:

Anything labelled "automated insights" that isn't drilling into your specific data. Generic dashboards that say "your revenue grew 8% — that's great!" aren't insights. They're summaries you could read off a graph. The bar should be higher: did it tell me something I didn't know, or that would change a decision I'm about to make? If not, ignore it.

AI that drafts customer-facing emails in your voice without review. I'm not against drafted emails. I am against drafted emails being sent automatically. The cost of one weird-sounding message to a long-standing client outweighs the time saved on dozens of fine ones. Keep a person in the loop on anything customer-facing for at least the first 90 days.

Anything that promises to "do your books" end to end. It can't, yet. It can do meaningful chunks of your books — categorisation, matching, drafting — but the judgment calls (is this a capital purchase or an expense, what's the right account, is this revenue recognition correct) still need a human, and ideally an accountant. Tools that imply otherwise are over-promising.

Tax and compliance automation. The part of accounting AI most likely to cause real problems is anything that automatically files, calculates GST, or makes IR-facing decisions without clear review. Use AI to prepare and check these things, not to submit them. The downside of an automated mistake here is a long way bigger than the time you saved.

What this means for your accountant relationship

A thing I get asked about often: does AI in MYOB mean I need my accountant less? My honest answer is no, but it changes what you should be paying them for.

If your accountant is mostly doing data entry, categorisation, and basic reconciliation, AI is going to compress that work. That's fine — it was always low-leverage time. What you should be doing instead is getting more of their attention on the things AI is bad at: structural advice, tax planning, decisions about how to set the business up, helping you read the numbers in the context of where the business is heading.

The accountants I see doing well with this shift are the ones leaning into the advisory side. The ones that aren't are the ones still pricing time spent on data entry. If you're paying for the latter, this is a good moment to renegotiate the engagement.

AI in software you already use changes the question from "which tool?" to "which process first?"

How to actually get value, not just features

The trap I see operators fall into with this stuff is treating AI features like a buffet — turn them all on, see what sticks. That doesn't work, because you don't have time to evaluate ten things at once and the ones that don't fit your process actively make things worse.

The pattern that works:

1. Pick one workflow. Not three. One. Bank reconciliation, follow-up reminders, or quote drafting are all good first picks.
2. Define what good looks like before you turn anything on. "Reconciliation should be done by 10am Monday with under five exceptions per week." If you don't have a measurable target, you can't tell if it's working.
3. Run it for two cycles. A month, two months. Compare against your baseline.
4. Decide: keep, adjust, or drop. If it's a keeper, document the new process. If it needs adjustment, change one thing and run it again. If it's not working, drop it and try a different feature. Don't keep noisy automations on out of inertia.
5. Then pick the next workflow. One at a time.

This is unglamorous. It's also how the teams that actually get value from accounting AI operate. The teams that turn on every feature and never measure are the ones who decide six months later that "AI didn't really do much for us" — usually because they couldn't tell signal from noise.

What to wait on

A few things I'd hold off on until they've matured a bit:

• Cross-tool agentic workflows. "Get the data from Stripe, push it through MYOB, summarise it, draft a report" — agents that span multiple systems are improving fast but still fragile. If you're using one of these in production, you need real monitoring. Most small teams don't.
• Voice and chat interfaces over your accounts. Cool demo, not yet a workflow improvement. Wait until you've heard a real operator say "this is how I run the books now" and not "I tried it once."
• AI-driven cash flow forecasting. Same caveat I'd give for any small-team forecasting: a spreadsheet built once with your accountant beats a model that's confidently wrong. Revisit in a year.

Where to go from here

If you're on MYOB and you want to actually do something this quarter, the move is small and specific: turn on bank feed categorisation properly, set the threshold conservatively, and put 20 minutes a week against reviewing what it does. That's the lowest-risk highest-payback first step for almost every team I see.

After that, look at where your team is losing time to admin handoffs. That's where the first-automation framework gets useful — picking the right second thing to add matters more than picking the first.

If you want help with the wider rollout — not just MYOB features but how AI fits across the rest of your stack — that's exactly what my workflow automation work is built for. The honest version of "AI in your business" is a few small, well-chosen assists that quietly run every week. Not a transformation. Just better defaults.