Where AI actually earns its keep in a small practice — the messy file, the working paper, the report pack — and the short list of work that has your name on it and must stay that way.
There is a contradiction sitting in the middle of the profession’s own research, and it is the most useful thing to start from.
CPA Australia’s Business Technology Report found AI use in business rising from 69 per cent in 2024 to 89 per cent in 2025, with the proportion using it “all the time” doubling over that period. Meanwhile the same body’s earlier report — a survey of 1,060 accounting and finance professionals across Australia, Mainland China, Hong Kong, Vietnam, Singapore and Malaysia — found just 22 per cent of Australian businesses had adopted AI to a moderate to significant extent, against 41 per cent across the other markets surveyed, and one in three Australian businesses saying they had never used it at all.
The two figures are from different reporting years and measure different things, and both are true. Together they describe the real state of play: almost everyone has tried it; almost nobody has built it into how the work gets done. In an audit-minded profession that is not surprising. A tool that produces an answer you cannot show your working for is worse than no tool.
So this is not a list of things AI can write. It is the three places in a month-end where it genuinely helps, and the checkpoint at the end of each.
Every practice knows the file. A client sends a spreadsheet with merged cells, a date column stored as text in four formats, three tabs with the same headings, a total row in the middle of the data, and a fourth tab that is somebody’s notes.
Cleaning that up is skilled work that is nonetheless not worth your charge-out rate, and it is the single best use of AI in a practice. Not “analyse this and tell me what it means” — that comes later — but the structural pass: normalise the dates, split the merged headers, flag the rows that don’t reconcile to the total, and tell you what it changed.
Two things make this work in a real spreadsheet rather than a chat window. First, the output has to be an actual sheet with real formulas, not a table pasted into a message that you then re-key. Second, you need the functions the work depends on — XLOOKUP, FILTER, LET and LAMBDA among roughly 397 of them, plus pivot tables, conditional formatting and real .xlsx round-trip, so the file opens where the client expects it to.
The checkpoint: ask for the change log, not just the clean file. What was reclassified, what was assumed, what was dropped. If it cannot tell you what it changed, do not use the file.
This is the part that separates a practice tool from a productivity toy.
A number in a report is only as good as the trail behind it. When you build a reconciliation, a depreciation schedule, a variance analysis or a cash-flow projection, the artefact you actually need is not the answer — it is the sheet that shows how the answer was reached, with the inputs visible and the assumptions written down where a reviewer can argue with them.
Use AI to build the structure of that working paper: the layout, the formula scaffolding, the assumption block at the top, the sensitivity rows, the check totals that must equal zero. Then put your own numbers through it. Tools like Goal Seek and Solver are useful here for exactly the reason they always were — they make the question explicit (“what utilisation rate makes this break even?”) rather than burying it in prose.
The checkpoint: every material figure traces to an input you can point at. If a number appears with no visible parent, it is not finished.
The third place is the one clients see. A trial balance is not a conversation. The pack that turns it into one — a short commentary document, three charts that show the trend rather than the total, a one-page summary, and a PDF that opens on a phone — is where a lot of practice hours go, and it is genuinely repetitive across clients.
This is one prompt’s worth of work when the numbers are already right: the commentary as a document with track changes so a partner can mark it up, the charts from the same data rather than pasted screenshots, the summary as slides if it is going to a board, and the whole thing as a PDF you can annotate and sign. Doing it once well and reusing the shape is the actual saving — create your first skill covers turning a good result into a repeatable one, and one prompt, multiple files covers producing the document, the sheet and the deck from a single brief.
The checkpoint: read every sentence of the commentary. Plausible prose about a client’s finances is the most dangerous output in this entire article, because it is the one a client will act on.
Everything above assumes a second set of eyes, so the software has to support one. In practice that means four unglamorous things: comments threaded on the cell or the paragraph in question, track changes with author and time so a review is attributable, version history so you can show what the file looked like before the adjustment, and real-time co-editing so a reviewer is not working on an emailed copy that has already diverged.
If you run repeated jobs — a monthly pack for the same twelve clients — this is also where automation should stop and ask. Autonomous Workers can run a scheduled job where every step leaves a real artefact behind, with dry-run previews and approval gates, which is the only shape of automation that belongs anywhere near client reporting: it prepares, it does not send. The autonomous worker explainer covers how the gates work.
Short list, no hedging.
Your professional and ethical obligations are unchanged by the tool you used. If you want the general version of this decision before you change anything structural, what to automate first and the AI-readiness checklist are the upstream pages.
Oppermind is one subscription with the pieces this month-end needed already inside it: a spreadsheet editor with around 397 functions including XLOOKUP, FILTER, LET and LAMBDA, pivot tables, slicers, Solver, Goal Seek, conditional formatting, charts and real .xlsx round-trip; a document editor with real .docx round-trip, track changes with author and time, comments, footnotes and citations; slides with .pptx export for the board pack; and a full PDF editor that annotates, OCRs, redacts, builds fillable forms and signs. Real-time co-editing with live cursors and version history sit under all of it. Unlimited documents, spreadsheets, decks and designs on every plan.
If you would rather not switch, the same AI is available as a downloadable task pane inside Word, Excel and PowerPoint — a manual install, not a store listing. The Office alternative article covers that decision properly.
Free with no card, Starter A$9.95 a month, Pro A$29.99 a month, Pro Plus A$59.99 a month, and Business A$59.99 per seat a month. For what everything else costs, see the 2026 price guide.
Whatever you’re here to make, make more of it.
A workspace is not accounting software. The ledger, the bank feeds, the BAS and the lodgements belong in purpose-built systems, and nothing here replaces them or the professional judgement applied through them. What it replaces is the layer above: the file preparation, the working papers and the client-facing pack.
And if your practice cannot yet describe, in writing, what client information may be put into which tools, that is the piece of work to do before any of this — not after. Our privacy policy sets out how Oppermind handles the data you put into it; what you are permitted to put in is governed by your own obligations to your clients.
It can do the parts around bookkeeping — normalising a messy client file, building the working paper, drafting the commentary, producing the report pack. The ledger itself belongs in purpose-built accounting software, and the classifications, positions and sign-offs belong to you. Treat AI as the thing that prepares work for review, never as the thing that finalises it.
That is a question about your confidentiality obligations and your engagement terms, not about any one product, and it deserves a real answer before you paste anything. Decide what may leave your systems, write it down, and apply it consistently. A practice that has not made that decision explicitly has made it implicitly.
The messy client file. It is high-volume, low-judgement, immediately verifiable, and it never touches a position you would have to defend. Start there for a month before you go near anything client-facing.
Possibly, and that is a legitimate answer. Oppermind’s sheet reads and writes real .xlsx and carries around 397 functions, but if your practice depends on VBA, Power Query or a data model, those stay in Excel — and the same AI is available as a downloadable task pane inside it.
Free with no card, then Starter A$9.95, Pro A$29.99, Pro Plus A$59.99 a month, and Business A$59.99 per seat a month, all in Australian dollars. Every plan includes unlimited documents, spreadsheets, decks and designs.
This article is general information about software and practice workflow. It is not accounting, audit, tax, legal or financial advice, it does not take account of your circumstances, and nothing in it overrides your professional, ethical or regulatory obligations, your engagement terms, or your duties of client confidentiality. You remain responsible for all work carried out in your name, however it was prepared. Oppermind prices are current as at 9 September 2026, are in Australian dollars, and are subject to the plan terms at checkout. Oppermind is not affiliated with, endorsed by, or sponsored by CPA Australia, Microsoft or any other organisation or product named here; all product names and trade marks are the property of their respective owners.
A real spreadsheet, a real document, a real PDF editor and a review trail — in one subscription.