For Australian accounting firms
Four unrelated products go by that name, only one of them touches the work that makes your season long, and the marketing for all four reads the same. This is a guide to telling them apart, the questions that separate the serious tools from the demos, and the things none of them can do.
Written by the team behind BeforeMay, which builds one of them. Where that matters we say so.
A practitioner comparing three vendors is usually comparing three categories, which is why the comparison never resolves. They do different jobs, they are bought by different people, and two of them are not sold to you at all.
Answer a question about the law
You describe a client situation and get back an answer with citations into legislation and ATO guidance, often as a draft letter. It replaces the hour you spend reading, not the day you spend preparing. Judge one on whether every assertion carries a citation you can open, and on what it does when the law does not settle the question.
Sold to the taxpayer, not to you
A consumer product: connect a bank feed, classify some deductions, lodge in fifteen minutes. Most of the visible marketing in this category is here, which is why searching for it turns up so little that is about a practice. If your clients are using one, the relevant question is what arrives at your office afterwards.
Sit inside the ledger
Coding suggestions on bank lines, inside or beside the accounting file. Mature, useful, and a different job — it makes the ledger tidier, and the work that follows the ledger is untouched.
The work between the documents and the signature
Read the source documents, treat every figure, and assemble the schedules — the working paper — that a return is signed off from. This is the one that touches the hours that actually hurt, and the one with the least written about it. It is also what BeforeMay does.
The rest of this page is about the fourth, because it is the one with the least written about it and the one where the evaluation is hardest.
Reading an amount off a statement stopped being interesting a while ago. Any competent tool will get the merchant, the date and the amount off a photograph of a receipt taken at an angle in a car park. That is table stakes, and a demo built around it is a demo built around the easy half.
The hard part is what happens to the figure afterwards. Which schedule it belongs on. Whether it is deductible at all, and on what basis. What to do when two documents disagree, or when the one that would settle it was never sent. Whether the total in the schedule can be walked back to the page it came from eighteen months later, by somebody who was not there.
That last one is the whole game. A working paper comes back from review not because the arithmetic is wrong but because the partner cannot get from a figure to its source without standing up and finding you — which is what review is actually checking. A tool that produces beautiful totals and no trail has moved that conversation later, not removed it.
Every vendor answers “does it use AI” the same way. These are the ones where the answers diverge, roughly in order of how expensive it is to discover the answer after you have bought.
Not "is there a list of sources somewhere". Click a number in a schedule and land on the document, the right page of it, without getting up. A tool that produces correct totals and cannot show its working has moved the checking rather than removed it, and the person who discovers that is whoever signs.
This is the question that separates the serious ones. A missing purchase contract, a cost base nobody has, a statement that stops in March — the honest behaviour is to say so, in the cell, and leave the schedule provisional. The dangerous behaviour is a plausible number. Ask to be shown a case with a document deliberately withheld.
You will disagree with it. Sometimes it will be wrong and sometimes the client will have told you something the documents do not say. If the answer to disagreeing is to re-run the whole thing and hope, it is not a working paper, it is a report.
Whatever a tool raised, what you did with each item, who changed what, and when. You need that for your own quality management, and you need it on the day a client asks how a figure was arrived at eighteen months ago.
Not "is it secure" — every vendor says yes. Ask where the files are stored, which steps run where, and which single step, if any, leaves the country. A vendor who can answer that crisply has thought about it. One who reaches for "enterprise-grade" has not. Ours is on the FAQ, with the one thing that leaves named rather than buried.
A phone photograph of a receipt taken at an angle. A locked bank PDF. Sixty-one pages of which eleven are the same terms and conditions. A .eml with the real statement inside it. Demos run on clean PDFs; seasons do not.
Individual returns are where every tool is strongest, because the shape is predictable. Companies, trusts and partnerships are where the same product is often quietly much weaker. Ask for a completed file of the entity type you actually have the most of.
Per seat, per return, per document, per page. The three are wildly different for a practice whose volume triples for four months. Price it against your July, not your February.
Worth reading before the part about us, because a vendor who will not write this section has left you to discover it in March.
The return is yours, the judgement is yours, and the obligation to review does not move because the first draft was fast.
Nothing infers a document that never arrived. The gap has to be visible and chased, which is a workflow problem before it is an AI one.
Where reasonable practitioners differ, a tool that sounds certain is doing harm. The useful behaviour is to set out the positions and stop.
The junior work was never valuable for its own sake, but it is where judgement came from. Removing it is a decision about your firm in five years, not just about this season.
The fourth one is the one nobody puts in a comparison table, and it is the one a principal should think hardest about. Ticking and casting was never the valuable part — it was the apprenticeship, and whatever removes the work has to be paired with a deliberate answer about where judgement comes from instead.
The obligation is yours, not your vendor’s, and it has two halves. A client is entitled to know who handles their information and where. And where AI is used, the steps taken to verify and review its output should be documented — which most firms will end up building by hand.
So the question to put to any vendor is not whether they are secure. It is whether they can tell you, without hedging, which steps run where and which one leaves the country, and whether their product hands you the review record or leaves you to keep it yourself.
We answer both properly, with the guidance cited and linked, on the FAQ — including the awkward parts. The provider list with the location of each one is in the Privacy Policy.
We are the fourth category and nothing else. BeforeMay takes the documents for a case, reads them, treats each figure, and builds the working paper the return is signed off from — as a spreadsheet you can open, edit and keep, not a report you have to accept. Every figure is linked to the page it came from, and a figure no document supports is written as outstanding rather than estimated.
Individual returns are the strongest path. Companies, trusts and partnerships each have their own working-paper format rather than an individual’s relabelled. Documents are stored in Australia and most of the processing runs on our own services in Sydney; the AI analysis step runs overseas, which we would rather put in a sentence here than leave you to find.
The evaluation we ask people to run is one case: take a return you have already finished, put the same documents in, and compare. The first three are free.
Four different kinds of product share the name: research assistants that answer questions about the law, consumer lodgement apps sold to taxpayers, bookkeeping classifiers that code ledger lines, and working-paper preparation that reads source documents and assembles the schedules a return is signed off from. They overlap almost not at all, so the first job in any comparison is working out which one is in front of you.
It depends entirely on how much of the tool knows about Australia. A general model asked a question about a trust distribution will answer in a voice of complete confidence using rules from somewhere else. What matters is whether the schedules, the treatment of each figure and the working-paper format are built for Australian returns, or adapted from somewhere they were not.
Some of it, and the difference shows up in the same two places every time: whether the schedules are the ones an Australian return is actually signed off from, and what the tool does with an entity type that does not exist in the market it was built for. A product adapted from elsewhere tends to be excellent on an individual’s deductions and thin the moment a trust appears. Ask to see a completed file for the entity type you have most of, not a feature list.
No, and a product claiming otherwise is describing a different regulatory world from the one you work in. The return is signed by a registered agent who has reviewed it. What changes is how much of the preparation you personally do before that point.
The short version is that using it is not the issue; being able to show how you verified and reviewed the output is. We answer that properly, with the guidance cited and linked, on the FAQ — including where the obligation sits between a vendor and a practitioner, which is the part most vendors are vague about.
Stored in Australia; most of the processing runs on our own services in Sydney; one step — the AI analysis — runs overseas, and we would rather name it than have you find it later. The full list, with the location of every provider, is in the Privacy Policy, and the plain-English version is the first answer on the FAQ.
Yes — each has its own working-paper format rather than an individual's with different labels on it. Individual returns remain the strongest path, and we would say so in a demo.
Priced in returns, not seats: the first three returns are free, and after that it is per return or a monthly plan. The numbers are on the pricing page and are readable without an account, because a price you have to book a call for is a price.
One case. Pick a return you have already finished, put the same documents in, and compare the working paper to the one you produced yourself. That comparison answers more than any demo, and it is the only evaluation we ask anyone to run.
The longer answers — training data, tax file numbers, consent, what happens when it gets something wrong — are on the FAQ. Rates and thresholds are in the calculators, and the rest of the writing is on the blog.