AI tools for Australian accounting firms in 2026

This is written from the perspective of firms in the 15 to 60 staff range running modern cloud stacks. It deliberately includes tools we do not sell and would not build, because the point is to give you a map rather than a pitch.

Prices and features move quickly in this category. Verify current terms with the vendor before committing.

The five jobs, and what fits each

Most firms need something in at least three of these. Very few need something in all six.

Ledger and compliance platforms

Xero has invested heavily here. JAX, launched in 2026, is positioned as an AI agent that answers questions about financial data in plain language and handles routine tasks like quote and invoice creation, working across the Xero platform and messaging channels. Underneath that, the account coding suggestions have been in the product for years and get more accurate with transaction history.

MYOB has moved more conservatively on generative AI, with the emphasis on coding suggestions learned from historical patterns rather than a conversational agent. For practices with a mixed client base this can be the more predictable option.

QuickBooks offers automatic transaction categorisation and anomaly detection.

What these are good at: volume coding, reconciliation suggestions, anomaly flagging. Genuine time savings on bookkeeping-heavy client bases.

Where they fall short: they operate inside their own boundary. If your workflow spans the ledger, your practice management system and your document store, the native AI does not cross those lines. That gap is where custom automation earns its place.

What to verify: all three maintain approved ATO integration status for lodgement. Any new tool touching lodgement needs to as well.

Document capture

Dext is the incumbent for receipt and invoice capture in the Australian market, extracting supplier, date, tax amount and totals from documents submitted by app, email or bank feed, then publishing to Xero, MYOB or QuickBooks. Its AI Assist feature suggests categories and tax treatments based on historical behaviour. Hubdoc is bundled with Xero and covers similar ground more simply.

What these are good at: structured, printed documents at volume. This is the most mature AI category in the stack and the one where accuracy claims are closest to reality.

Where they fall short: handwritten records and poor-quality photographs still need review. Accuracy on a clean supplier invoice and accuracy on a crumpled fuel receipt photographed in a car park are not the same number, whatever the marketing says.

Practice management

Karbon has the most developed AI feature set among practice management platforms used by Australian mid-size firms: email thread summarisation, draft generation, client summaries and task prioritisation, with firm-level admin controls to enable or disable features. It runs on Azure OpenAI with data retained within Karbon.

FYI is purpose-built for Australian firms on the Xero stack and is strong on document and process automation.

XPM and MYOB Practice are the workflow and tax layer for their respective stacks.

What these are good at: the correspondence and coordination overhead that surrounds compliance work. In most firms this is a bigger time sink than partners realise, because it is distributed across everyone rather than concentrated.

Where they fall short: they help you manage the work. They do not do the work. A tool that summarises an email thread about a workpaper does not prepare the workpaper.

Reporting and forecasting

Fathom, Spotlight Reporting and Float turn ledger data into management reports, KPI dashboards and cash flow forecasts. Xero's own predictive cash flow extends to 180 days on some plans.

What these are good at: standardising output across a client base, which is where the leverage is. If every client's management pack is bespoke, no tool will save you much.

Where they fall short: the analysis is only as good as the ledger. Forecasting on an unreconciled file produces confident nonsense.

Tax research

This is the category where general-purpose AI is most dangerous and where Australian-specific tooling matters most.

A small number of Australian-built tools now target tax research directly, including SavvyWise, founded by people out of an established Perth practice. The distinguishing feature of anything worth using here is that it cites Australian primary source material you can check.

Under Code items 9 and 10 the verification is your responsibility regardless. See what TPB(GS) 55/2026 means for your firm.

The general-purpose layer

Microsoft 365 Copilot, ChatGPT and Claude sit across everything. Most firms are already using at least one, often without a decision having been made.

These are good at drafting, explaining, summarising and restructuring text. They are poor at Australian tax technical work, because they are trained largely on international material and will produce fluent, confident and incorrect answers about Division 7A.

We compare the three in detail in Copilot vs ChatGPT vs Claude for accounting firms.

The compliance point matters more here than anywhere else in the stack, because these tools are easiest for staff to adopt without approval. See can Australian tax agents use ChatGPT with client data.

Custom automation: when off-the-shelf runs out

Everything above solves a problem inside one system. The workflows that consume the most time in a compliance practice usually span several.

Preparing an individual return workpaper touches the client's source documents, the ledger, ATO pre-fill, your workpaper template and your document management system. No vendor owns that whole path, which is why it stays manual in most firms even when every individual system has AI features.

This is the gap custom automation fills, and it is what we build. It is worth being honest about when it does not make sense: below roughly 15 staff, or where a workflow runs a few dozen times a year, configured off-the-shelf tools will beat a custom build on economics every time.

How to evaluate any AI tool

Six questions. Ask them of every vendor, including us, and get the answers in writing.

1. Where is client data stored and processed? Country matters for the Privacy Act and for client conversations.
2. Is customer content used to train models? Ask about your specific plan, not the marketing page. Consumer and business tiers of the same product often differ.
3. What is the retention period, and is it configurable?
4. Who inside the vendor can access data, and under what circumstances?
5. What happens to data on termination?
6. What certifications does the vendor hold? SOC 2 Type II is the common baseline.

TPB(GS) 55/2026 expects you to complete an appropriate review of commercial AI tools to confirm information will be kept secure and that Privacy Act requirements are met. These six questions are that review.

Tools built for the Australian accounting market generally answer these more cleanly than general-purpose platforms, because they were designed with local privacy law in view. That is a reason to prefer them, not a reason to skip the diligence.

What we would actually recommend

If you are starting from a standard cloud stack and want the shortest path to time back:

1. Get document capture right. It is the most mature category and the least risky. If you are not running Dext or Hubdoc properly, that is the cheapest win available.
2. Turn on and configure what is already in your existing platforms. Most firms are paying for AI features in Xero, MYOB or Karbon that nobody has switched on or trained the team to use.
3. Decide the general-purpose question deliberately. Pick one platform, put it on a business tier, write the policy, and train people. Doing this is much better than letting it happen on personal accounts.
4. Only then, look at custom automation for the cross-system workflow that is costing you most. By this point you will know what that is, because you will have exhausted the easy answers.

Doing these in the other order is how firms end up with a bespoke build that duplicates a feature they already own.

If you want that sequence worked out against your actual stack and volumes, our automation audit does it in a fixed scope.