Draft the boring bits, keep the judgement

A 6-minute read · for accountants, bookkeepers and advisors

AI won't do your professional judgement, and you wouldn't want it to. What it does well is the scaffolding around the judgement: the first draft, the summary, the follow-up. You review and sign; it just gets you to the review faster.

Most NZ practices already run Xero, and many bolt on Dext or Hubdoc for source documents and Karbon or FYI for workflow. The good news is that a lot of the AI value below plugs into that stack rather than replacing it - and a fair chunk is about pointing tools you already own at the jobs that quietly eat your team's week. Take them in order; the first two pay for themselves fastest.

1. First drafts of letters and emails

Advisory letters, engagement notes, "here's what your numbers mean" client emails - the blank page is the slow part. Give AI the figures and a few bullet points and it drafts in your firm's tone, ready for you to correct and add the judgement. You're editing, not composing from scratch.

The trick that makes this work in a practice: build a small library of your own approved templates - the year-end summary, the "your provisional tax is due" note, the fee-increase letter - and have AI fill each one from the specifics of the client in front of you. Consistent tone, no blank page, and a junior can produce a first draft that reads like the firm. You still read every word before it goes out.

Roughly: 2-3 hours a week across a busy practice.

2. Meeting notes and action lists

The client meeting is the easy part; writing it up afterwards is where the evening goes. An AI notetaker turns a recorded call into tidy file notes and a clear action list, so the write-up is a check-and-save rather than a start-from-memory. Save the notes straight to the client's file in Karbon or FYI and the follow-up email half-writes itself. See the worked example →

One caution worth naming up front: tell clients you're recording, and keep the recordings where client data belongs, not in a random consumer app. More on that below.

Roughly: most of the post-meeting admin, gone.

3. Chasing records at year-end

The annual scramble for client records is a job nobody enjoys and everybody repeats. Automated, polite reminder sequences - "we still need X to finish your return" - keep the chasing consistent without a team member owning the nag. It's the same idea as invoice reminders, pointed at missing paperwork. See how automatic reminders work →

Practice tools like Karbon and FYI can run these as part of a job checklist, so the chase escalates on a schedule and stops the moment the records land. The win isn't just time - it's the clients who file earlier because the reminders were steady rather than dependent on someone remembering to send them.

4. Coding and bank rec, faster

For bookkeepers, the daily grind is coding transactions and reconciling. Xero's bank rules and the machine-learning suggestions in Dext and Hubdoc already handle a lot of the repetitive coding - the value is in setting the rules up properly and reviewing the exceptions, not touching every line. Get the rules right on your highest-volume clients first; that's where an hour of setup saves an hour a week, every week.

Roughly: 1-2 hours a week per high-volume client, once the rules are tuned.

5. A knowledge assistant for the FAQs

Clients ask the same things: provisional tax dates, what's deductible, how to send records, when GST is due. An assistant trained on your own guidance answers those instantly and consistently, freeing your team for the questions that actually need a person. Built on your documents, so the answers are yours, not generic. See the example →

What it costs

Less than most people expect. A business-tier AI subscription (with training switched off) runs roughly $20-$30 per user a month, and the notetaker and reminder features are often already inside the tools you pay for - Xero, Dext, Karbon, FYI. The real cost is the afternoon spent setting up templates and rules properly, plus a clear internal policy on what may and may not be pasted into an AI tool. Spend on the setup and the policy; be sceptical of anyone selling a bespoke "AI for accountants" platform before you've exhausted what your existing stack already does.

What to be careful with - this matters more for you

• Client data is the whole ballgame. Never paste client financials or personal details into a free public chatbot. Use business-tier tools with training turned off, and know where the data sits. This isn't optional in a practice with Privacy Act obligations.
• Judgement stays human, always. AI drafts; a qualified person reviews and signs. Nothing goes to a client or IRD unreviewed.
• Watch for confident nonsense. AI can state a wrong figure or tax rule with total confidence. Treat every output as a draft to verify, never as an answer - especially on anything IRD will see.

The honest version

Used well, AI gives you back the hours you lose to drafting and chasing, so more of your week is the advisory work clients actually value. Used carelessly with client data, it's a real risk. The whole job is knowing which is which - and that's exactly the line the check helps you draw.

Written by the Even Odds team · Practical AI for Kiwi business · Last updated 18 July 2026