Speak to us now:
1300 728 875
Your bank feed has matched almost everything. The GST report generated without a single error message, the BAS figures sat ready to lodge, and the whole thing took eleven minutes.
If it seems too good to be true, it probably is. AI can assist with BAS preparation, but it cannot independently guarantee that every transaction is complete, correctly coded and reported in the right period. It calculates what it has been given, but it doesn’t know what it has not been given, and it doesn’t know why any particular payment happened.
Artificial intelligence has moved into accounting software quickly, and most business owners now run at least part of their bookkeeping through it without ever deciding to. This article covers what AI handles well, where it gets things wrong, who carries responsibility for the figures, and how to tell whether your business is simple enough to lean on automation.
Partly. Accounting software can automate tasks that used to take a bookkeeper hours, and it handles most of the mechanical work feeding into a Business Activity Statement. It categorises transactions, applies rules, calculates GST totals, produces a draft statement, and often transmits that statement electronically to the ATO. If you use Xero, MYOB or QuickBooks, you have seen how much happens without you touching it.
But producing figures and confirming figures are different jobs. Software is very good at processing data and calculating totals. It helps review the underlying records but can’t do that alone. Confirming GST treatment, providing BAS advice and authorising lodgement involve interpretation, context and accountability, all of which remain human work and still need human review before anything is lodged.
AI cannot identify a transaction it was never given. AI cannot always determine why a transaction occurred, and AI cannot accept responsibility for the final statement. Your BAS reports obligations including GST, PAYG instalments and PAYG withholding.
Document capture tools read a photographed receipt, extract supplier, date, total and GST amount, and attach the source document to the transaction. A transaction with the tax invoice attached is far easier to substantiate later. It also lifts data quality across the whole file, since the source document travels with the entry.
The software matches bank feed entries to invoices, learns from how you coded similar items previously, suggests accounts and GST codes, and flags what it could not match. Most of this runs on machine learning that reads your usage patterns rather than any understanding of your business. This works when transactions are predictable and were treated correctly in the past. The system isn’t evaluating whether your historical coding was right; it’s copying it.
Software doesn’t fat-finger a number. AI can reduce individual data-entry mistakes, catch duplicate entries before they land, and deliver fewer errors across high-volume repetitive tasks. It can also repeat an incorrect rule across every similar transaction. One is a typo affecting one line; the other is a systematic error affecting a quarter of coding.
Automated feeds and frequent reconciliations let you see your GST position during the quarter rather than a week before lodgement. That real-time visibility makes cash flow forecasting far more useful, because you’re planning against figures that reflect this week rather than last quarter. Once the system is set up properly, the time savings compound across every reporting period. But real-time reporting is only valuable when the underlying records are complete. Current and wrong is worse than late and right, because it feels reliable.
Integrations with ecommerce platforms, payment gateways, CRM, payroll and inventory remove duplicate handling and pull your financial data into one place, while multiplying the places a mapping can be configured incorrectly.
Not sure your software has treated every transaction correctly? A registered BAS agent can review your reconciliations, GST coding and final figures before you lodge. Request a BAS review.
The Tax Practitioners Board (TPB) defines what does and does not constitute a “BAS service,” and is explicit that not everything involved in preparing an activity statement qualifies. Entering data, coding transactions according to instructions, processing payments and preparing bank reconciliations are not BAS services, because they do not require interpreting or applying a BAS provision. The TPB specifically notes this holds whether those processes are done manually or through an automated process.
Now look at what the TPB does classify as a BAS service. It covers determining the figures to go on an activity statement, confirming those figures, coding transactions where that requires interpreting a BAS provision, setting up software and determining default GST codes tailored to a business, and reconciling data to work out what belongs on the statement.
The regulatory boundary falls almost exactly where the automation boundary falls. The work AI does well is broadly the work the law treats as mechanical. The work the law treats as requiring professional judgement is precisely the work AI can’t do, because it’s defined by applying a rule to specific circumstances. Both lines are drawn in the same place for the same reason — some tasks are processing, and some are judgement. Legal compliance sits on the judgement side of that line.
When the software generates your BAS, those figures are the accumulated output of hundreds of earlier decisions. A transaction can be entered with perfect accuracy and still carry the wrong GST code. It can be correctly coded and land in the wrong quarter. Bank feeds show money moving, but they don’t explain the commercial purpose behind it, and treatment often depends on that. Australian tax rules turn on purpose and circumstance, not on what the bank statement says.
Preparing a BAS properly means checking bank reconciliations, GST and payroll reports, clearing accounts, merchant settlements, supporting invoices, prior-period adjustments, and anything unusual or high-value.
This is the biggest failure mode, and it’s invisible by design. A BAS built on incomplete records can be arithmetically flawless and still wrong.
Software can’t know about cash sales never recorded, a second bank account never connected, missing receipts, unentered supplier invoices, a duplicated invoice, an unrecorded refund, incorrect opening balances, payroll sitting in another system, or personal spending from the business account with no explanation attached. If an invoice never made it into the file, the software produces a perfectly calculated BAS from incomplete information, and every reconciliation looks green.
Under the ATO’s safe harbour provisions, a business engaging a registered agent may avoid certain administrative penalties, but only where it provided all relevant information. The ATO’s own explanatory material uses this exact scenario, and the owner who forgot to hand over his cash receipts remained liable. Incomplete records break the protection, and they break the automation.
AI recognises the supplier; it doesn’t know the purpose of that particular purchase.
The same hardware store appears three times on your statement. One is trading stock, one a repair, one a capital asset. Same merchant, same card, three different treatments. Consider also a vehicle expense with mixed business and private use, a deposit that’s a loan rather than sales income, a refund relating to an earlier period, owner drawings, or purchases connected to GST-free or input-taxed sales. The bank description carries almost none of the information needed to decide.
In our experience, software sees movement but not meaning. It can flag that sales are down 30%. It can’t tell you whether that reflects a seasonal downturn, a customer paying late, changed invoicing timing, a project finishing, or a restructure.
A human reviewer asks a different class of question. Why did this move? Does the timing make sense? Is this consistent with what actually happened operationally? Does it belong in this period at all?
Purchases and sales can land in different periods. Your GST accounting method (cash or accruals) changes when amounts are reported. A receipt date, invoice date and payment date are three different things and aren’t interchangeable. Automated matching confirms two records correspond. It doesn’t confirm the entry belongs in the period you’re about to lodge.
Bank rules and AI suggestions both lean on historical treatment. If that treatment was wrong, automation faithfully reproduces the error, and a new transaction can resemble an old one without sharing its purpose. A single misconfigured rule can apply the wrong GST code hundreds of times, flowing into your BAS, management reports and year-end accounts simultaneously. Errors at that scale distort financial reports you rely on to make decisions, not just the figures you send the ATO.
It’s far cheaper to configure automation correctly at the start than to unwind months of repeated errors afterwards. Correcting the setup is a conversation; correcting a year of coding is a project.
Public chatbots are a separate category from the AI inside your accounting software. An AI chatbot such as ChatGPT or Google Gemini is genuinely useful for drafting marketing copy or explaining a concept in plain English. Tax advice specific to your business is a different task.
A generative response about GST treatment can be fluent and authoritative in tone while being incomplete, outdated, or built on an untrue assumption about your circumstances. AI outputs shift with the wording of the question, so rephrasing it may get you a different answer. What the AI produces depends entirely on your user inputs, and the tool can’t inspect your accounting file or signal how uncertain it is. Treat AI-generated content as a thinking partner, not a substitute for tailored advice.
Software vendors don’t lodge your BAS. You do. A robot cannot make decisions; therefore, it cannot be held accountable.
Using accounting software doesn’t transfer your record-keeping obligations to the provider. The business remains responsible for keeping proper records, for the accuracy of what is reported, and for the statement it authorises. Nothing in Australian law reduces your exposure because a tool generated the number.
However, a provision recognises registered agents.
Under the safe harbour provisions in the Taxation Administration Act 1953, a business engaging a registered BAS or tax agent may not be liable for certain administrative penalties (specifically for a false or misleading statement resulting in a shortfall, or for failing to lodge on time) where the error resulted from the agent failing to take reasonable care. The conditions are strict:
The practical implication is straightforward. There is no equivalent safe harbour for using software. A registered BAS agent is a person the penalty framework recognises; an automation rule is not.
None of this means an agent makes penalties disappear. It means the law treats regulated professional judgement differently from a tool. If you rely on automation alone, you carry the full weight of the outcome yourself, which means business owners need to understand what their AI use is actually doing to the numbers.
If something has already gone wrong, it’s usually fixable. Depending on the type, size and timing of the error, the ATO allows many mistakes to be corrected in a later BAS, while others require revising the original. Time and value limits apply and vary with your GST turnover. Once the ATO has notified you of a review or audit covering that period, you generally can’t simply correct it in your next BAS.
A registered BAS agent’s role in an AI-assisted process is to review the data, confirm GST treatment, check reconciliations, calculate liabilities and entitlements, prepare or lodge the statement, and represent you with the ATO.
You are not legally required to engage a BAS agent to lodge your own BAS. Owners can and do prepare their own. The registration requirement applies to anyone providing BAS services for a fee, which is why it matters that anyone you pay appears on the TPB register. For your own lodgement, what matters is understanding what has and has not been checked before you sign.
There’s a real difference between an approved integration inside your accounting platform and pasting a bank statement into a public AI platform.
Your financial records contain a lot of information about other people, including customer contact details, employee and payroll records, bank account details, supplier information and commercially sensitive margin data. Once customer data leaves your accounting file, you lose some control over it, and sensitive data is difficult to claw back.
The Office of the Australian Information Commissioner (OAIC) advises organisations to assess their privacy obligations when using commercially available AI products and recommends, as best practice, that organisations do not enter personal information, particularly sensitive information, into publicly available generative AI tools. Privacy laws apply to information you put into an AI system as much as to information you store, and obligations under the Privacy Act don’t pause because a third party is doing the processing.
The Australian Cyber Security Centre (ACSC), in guidance developed for small business with COSBOA and New Zealand’s NCSC, identifies the main risks of cloud-based AI tools as data leaks, privacy breaches, unreliable outputs and dependence on third-party providers, and recommends reviewing the data handling, access and privacy policies of any AI vendor. It references a 2025 incident in which a contractor uploaded personal information, including health records, into an AI system, resulting in a notifiable data breach.
Before connecting a tool, get answers to these:
A short written AI policy is worth the hour it takes. Set clear rules about which tools are approved, what confidential data must never be pasted into them, and who signs off on a new integration. Good AI governance in a small business is mostly a matter of staff members knowing where the line sits.
Cloud AI isn’t inherently unsafe. For most small businesses, a reputable cloud accounting file is much safer than a laptop under the counter. You should have answers to these questions before your payroll data goes anywhere new.
There’s no one-size-fits-all answer. Complexity decides it, and size or appetite for technology matter far less.
Automation can handle a large share when you have low transaction volume, one main bank account, repetitive transactions, no employees, no inventory, consistent GST treatment, complete digital records, genuinely separated personal spending, and regular reconciliations. Lower complexity doesn’t mean no review, but it does mean a shorter review.
A hybrid process tends to suit businesses with employees and payroll, growing transaction counts, several revenue streams, ecommerce or gateway integrations, some inventory and occasional capital purchases, where the owner is comfortable reviewing reports, and a bookkeeper is available for periodic checks. Many businesses land here as they grow.
Closer oversight is warranted where you have multiple entities, multiple accounts or currencies, significant inventory adjustments, complicated payroll, mixed or input-taxed GST treatments, imports or exports, mixed personal and business spending, incomplete records, overdue statements, previous BAS errors, or ATO correspondence.
| Business situation | Recommended approach |
| Micro-business with simple, repetitive transactions | AI-supported bookkeeping with scheduled human checks |
| Small business with payroll or several integrations | Hybrid process with regular bookkeeper or BAS agent review |
| Growing business with inventory, multiple revenue streams or complex GST | Ongoing human oversight supported by automation |
| Business with overdue records or previous BAS errors | Professional clean-up before expanding automation |
This is a practical framework, not a legal classification. The direction of travel holds regardless. As complexity rises, the proportion of your BAS that can safely be automated falls.
AI-led suits simple businesses with predictable transactions, clean records and an owner comfortable reviewing software output. You get fast repetitive processing and frequent reporting, at the cost of context. You rely heavily on correct initial setup, and a wrong rule replicates silently.
Human-led suits complex or unusual transactions, payroll and inventory, and owners without time to review software output. You get contextual judgement, strategic advice and someone who asks questions, at a higher labour cost that is genuinely unnecessary for routine high-volume processing.
Hybrid is the practical answer for most small businesses. AI handles repetitive processing, and the owner or bookkeeper reviews exceptions. A registered BAS agent checks reconciliations and final figures before lodgement. When an error surfaces, the underlying rule gets corrected, not just the number. Newer features like predictive analytics and fraud detection are genuinely useful under appropriate oversight, and unreliable without it.
AI and human bookkeepers aren’t competing options for most businesses. The menial work is gone, and that’s a good thing, because it frees the bookkeeper to focus on what drives profitability.
| Safe to automate | Automate, then review | Requires human judgement |
| Bank feed imports | Suggested account categories | GST treatment of mixed-use purchases |
| Receipt and invoice capture | Recurring supplier coding | Capital vs expense decisions |
| Attaching source documents | New supplier transactions | Period allocation and adjustments |
| Matching invoices to payments | High-value transactions | Whether records are complete |
| Duplicate detection | Payment gateway settlements | Loan vs income classification |
| Report generation | Payroll and STP totals | Final review and authorisation |
Base accounting software rarely does everything on its own. Receipt capture is often an add-on, as is advanced payroll, inventory, or the automation tier that actually delivers the features you saw demonstrated. Each integration is another subscription, and costs creep up until the stack is genuinely expensive.
Then add the costs that never appear on an invoice, including setup and migration, staff time learning the system, someone reviewing exceptions weekly, and corrective bookkeeping when automation has been wrong for a quarter. Proper training takes hours you have to find, and configuring integrations often needs technical expertise a small team doesn’t have in-house. A tool that saves ten minutes but generates an hour of checking has not delivered efficiency, it has moved the work and added a bill.
Compare software and add-on costs, setup, time saved, time spent reviewing, professional checking and likely correction costs. Factor in whether you get local support in your own timezone when something breaks two days before lodgement. Human bookkeeping isn’t automatically cheaper either, so add it up properly.
Use automation without losing oversight. Darcy can configure your bookkeeping process, review the transactions that need judgement, and prepare or lodge your BAS. Speak to a BAS agent.
Chart of accounts, GST codes, bank feeds, opening balances, payroll settings, integration mappings, user permissions. This is where professional help is cheapest. Determining default GST codes tailored to your business is itself a BAS service under TPB guidance. Our accounting software setup service exists for this.
Recurring expenses, subscriptions, standard customer payments. Don’t automate unusual, high-value or mixed-use transactions early. Those need eyes on them.
Low-confidence matches, new suppliers, high-value items, duplicates, unmatched entries, manual journals, unusual GST codes, and large movements against the previous period.
Bank accounts, credit cards, payment gateways, clearing accounts, payroll accounts and loans. Clearing accounts and gateway settlements are where errors hide most comfortably. If reconciliation is the bottleneck, our data entry and bank reconciliation service handles it.
Confirm the transaction belongs in this period, the invoice is available, the business purpose is clear, private use has been considered, and the GST treatment makes sense.
Whoever signs should be able to explain why GST payable or refundable moved, whether accounts are reconciled, and whether payroll totals align with STP reporting. If you can’t explain the movement, don’t lodge it yet.
Correcting the BAS figure alone guarantees the error returns next quarter. Correct the transaction, the GST code, the bank rule, the integration mapping and the review procedure.
Anything unticked isn’t a crisis, but check it before you lodge.
Speak to a registered BAS agent if:
Darcy’s registered BAS agents and degree-qualified accountants help with one-off BAS reviews, preparation and lodgement, ongoing bookkeeping, catch-up bookkeeping, software setup, and designing hybrid AI-and-human workflows. We work with Australian businesses across Xero, MYOB, QuickBooks, Reckon and Saasu, Australia-wide. You can verify any agent’s registration on the TPB public register, including ours.
AI makes BAS preparation faster, and for many small businesses that’s a real gain. It works best where information is complete, transactions are simple, and treatment is repetitive.
What it cannot do is understand your business. It cannot identify a transaction it never received, cannot always tell you why a payment happened, and cannot take responsibility for the statement you authorise. A BAS can be generated perfectly and still rest on incorrect assumptions. Because the output looks identical either way, you still need to check the figures.
The key points are simple enough. Australian tax rules turn on context that automation can’t see, no AI output is perfectly accurate on its own, and most business owners get better results from automation paired with review than from either extreme. Your complexity determines how much oversight you need. For most small businesses, the workable answer is automation supported by professional review.
Not confident your software has treated every transaction correctly? Have Darcy review your BAS and bookkeeping process before your next lodgement. Call 1300 728 875 or get help with your BAS.
AI-supported software can process transactions, calculate GST totals, produce a draft BAS and often transmit it to the ATO. It can’t confirm the underlying records are complete, the coding is correct, or amounts fall in the right period. Someone still needs to review and authorise the statement.
It depends on the inputs: complete records, correct configuration, appropriate GST codes, and the business’s complexity. A BAS can be mathematically flawless and still wrong, because it only reflects the information the system was given and the rules it was told to apply.
No. You’re not required to use a BAS agent simply because you lodge a BAS, and owners can lodge their own. An agent becomes valuable where transactions are complex, or you want final figures independently checked. Anyone providing BAS services for a fee generally does need TPB registration.
You are. Software doesn’t transfer your record-keeping obligations or your responsibility for accuracy. Errors may be correctable in a later BAS or may require revising the original. There’s no safe harbour for software. It applies to registered agents, in defined circumstances, and only where you provided all relevant information.
It can suggest one based on previous transactions, and will be right most of the time on routine items. It can be wrong where the business purpose, supplier relationship, or private-use component differs from what it learned, and it will apply that wrong code consistently until someone changes the rule.
That depends on the tool’s privacy, access, retention and security controls. The OAIC recommends organisations avoid entering personal information into publicly available generative AI tools. An approved integration inside your accounting platform is a different proposition from a public chatbot, but either way, understand where the data goes and whether it’s used for training before uploading customer, employee or banking details.
"*" indicates required fields