AI automation for NZ accounting firms: what it costs

A small NZ accounting firm can automate client onboarding, reconciliation exceptions, and the monthly reporting pack for $5,000 to $15,000 NZD as a first scoped build, after a fixed-fee discovery at $3,000. Here's how the numbers break down, and what to automate first.

The month-end wait

I'll start with my own accountant. Good firm, sharp people. And every month-end I waited days for numbers that were already sitting in Xero, because somewhere in their office a person was assembling my reports into a PDF by hand.

That wait isn't a people problem. It's a workflow problem. And if your clients are waiting on you the way I was waiting on them, it's costing more than hours. The wait is what a client remembers when the fee conversation comes around.

Why this is suddenly a live question

Three things converged this year. Xero's AI assistant, JAX, now reconciles high-confidence bank matches on its own, and Xero reported more than 100 million transactions auto-reconciled by August. CA ANZ's own survey found 92% of chartered accountants want AI training and only 30% have had any. And the government's AI Advisory Pilot has been extended again, now running to 30 June 2027.

One more number for context. New Zealand has roughly 5,600 accounting-services businesses, and industry commentary suggests no more than about 80 of them clear $5 million in revenue. Nearly every practice in the country is small. Which means nearly all the AI advice written for "firms", the enterprise pilots and big-firm innovation programmes, is written for somebody else.

What to automate first

Don't try to fix the whole practice at once. Three workflows have the cleanest payback:

1. Client onboarding. Engagement letters, ID checks, IRD authority, and the document chase. It's the same sequence every time, which is exactly what makes it automatable, and it's the first impression every new client gets of how your practice runs.
2. Reconciliation exceptions. Xero already matches most transactions. It's the exceptions that eat your Fridays: the odd payment references, the split invoices, the client who pays three bills in one transfer. An exception-handling workflow routes them, resolves the known patterns, and learns the new ones.
3. The monthly reporting pack. Assembling, checking, and sending. This is my month-end wait from the other side of the desk, and it's usually the fastest win of the three.

Intuit's research puts the saving from AI-powered accounting tools at around 12 hours a month for nearly half of users. That's a vendor number, not mine, but the direction matches what I see in practice: the hours are in the repetitive assembly work, not the advice.

"Why pay anyone when Xero's AI is free?"

Fair question, and it deserves a straight answer.

Platform AI automates what every firm shares. When JAX reconciles automatically and predicts cashflow, it does it for every Xero subscriber in the country on the same day. That raises the floor. It doesn't move your practice relative to the one down the road, and it can't, because it doesn't know anything specific about how your practice runs.

The standard practice stack works the same way. Xero Practice Manager (now folding into Xero Partner Hub), FYI, Ignition, often Karbon at $59 to $99 USD per user a month and Dext, which prices its practice plans per client rather than per seat. These are connectors: pre-built links that handle the workflows everyone has. Useful, and worth having, though which of them is still worth backing in two years is a separate question. A custom build starts where your workflow stops matching the template. The onboarding sequence that's specific to your practice. The exception rules only your senior knows. The reporting pack your best clients actually read.

Platform AI raises the floor. Custom automation raises your margin.

And here's the part the vendors don't say: even firms that understand all this stall. In CA ANZ's survey, 45% of senior decision-makers said they don't feel confident using AI. That's not a capability gap, it's a confidence gap, and honestly it's a rational one. You carry the liability if a tool does something silly with a client's GST.

The fix for a confidence problem isn't another subscription. It's a scoped build with a definition of done, one metric that matters, and a documented handover your team actually understands.

How it fits the stack you already run

You don't rip anything out. Xero stays the ledger, XPM or Karbon keeps running jobs and time, FYI keeps the documents, Ignition keeps the proposals and billing. A custom build sits between them and does the practice-specific steps none of them own: moving a client from signed proposal to set-up job, sorting the exceptions Xero couldn't match, pulling the reporting pack together from three places.

One rule makes it safe, and it's the rule I build to. The automation drafts, a person approves. Nothing posts to a client's ledger and nothing goes out in a client's inbox until someone in the practice has looked at it. The system does the assembly; your people keep the sign-off, and with it the liability they already carry.

The practical upshot is that the first build is smaller than people expect. It's usually one workflow, wired into the tools you already pay for, with a person at the approval step.

What it actually costs

Discovery first, always. A fixed fee of $3,000 NZD that maps how work actually moves through your practice and finds the leverage. If a consultant wants to start building before they've seen your workflows, that tells you something.

From the map, most small practices land between $5,000 and $15,000 for a first build covering one or two of the workflows above. Touch more systems and it climbs; the honest ceiling for this kind of work is around $25,000. Practices with a genuine backlog sometimes move to a retainer from $4,000 a month, but that's a decision the map makes, not the sales call. I've written up how AI consulting pricing works in NZ if you want the longer version, and how we work with accounting firms specifically.

Software subscriptions sit on top, but they're small: $30 to $250 a month for most stacks. The spend that matters is the scoping and integration, not the subscriptions.

I'm building this kind of exception-first workflow for a Queenstown chartered accountancy right now: the system proposes how each account should be mapped, a person confirms only the ones it isn't sure about, anything it accepted on its own stays listed so they can scan it, and the confirmed map is kept for that client so next year's review starts from it. It's not finished, so I won't dress it up as a result yet.

The closest finished build I can show you isn't an accounting firm, and I'd rather say that plainly than imply otherwise. My Wellness is a Queenstown wellness practice that ran on the same admin shape: bookings by phone, invoicing by hand, records scattered across tools. We wired Cal.com, Stripe, and Xero together and built the practice platform around them. Ten-plus admin hours saved per practitioner, every week. Same integration depth, different kind of practice, and the accounting version of that build is the onboarding, reconciliation, and reporting stack above.

/ From the workshop · Health platform · Queenstown

My Wellness: 10+ admin hours saved per practitioner every week.

Read the case study →

Client data and the Privacy Act

The Office of the Privacy Commissioner's AI guidance recommends a privacy impact assessment before starting an AI project, and for a practice holding client financials that's not bureaucracy, it's hygiene.

In practice, for a small firm, it comes down to three things. Use business-tier AI accounts with model training switched off. Know exactly which client data enters which tool. And get your vendors' data-processing terms in writing before anything sensitive flows through them. A decent advisory firm builds this into the scoping rather than treating it as an afterthought.

What stays human

Judgment. Tax advice. The conversation where a client asks whether they can afford to hire, and you know their numbers and their nerve well enough to answer.

Automation buys those hours back. The advice is why clients stay, and no reconciliation bot is coming for it.

The takeaway

Don't start by buying AI tools. Start by mapping how work actually moves through your practice, then automate the three admin sinks: onboarding, exceptions, reporting. That's the step most firms skip, and it's why so many practice-management subscriptions end up as expensive shelf-ware.

If your month-end still ends with someone assembling a PDF by hand, a short call is the fastest way to find out what a first build would cost your practice.