Most builders and trades businesses do their books at the wrong time of day. Nine at night, kitchen table, a month of receipts in the ute console, a supplier statement that does not agree with anything in Xero, and a growing suspicion that the last three jobs were less profitable than they felt.
AI has genuinely changed part of that. Not the marketing version - the actual version, where the data entry mostly disappears and you get your evenings back. But it has changed one part and not another, and in construction the part it cannot do is the part that costs the most: knowing which job a cost belongs to.
This is a practical guide to what to automate in Xero, what to hand to a tool like Claude, and where to keep your hands firmly on the wheel.
Start With What Xero Now Does By Itself
Xero's AI assistant, JAX, now reconciles high-confidence bank transactions automatically instead of queuing them up for you to click through. Xero's stated target is to auto-reconcile more than 80% of statement lines in real time, and it reports over 100 million transactions auto-reconciled since launch.
It learns from your own history - how you coded that merchant last month and the month before - plus anonymised patterns from similar transactions across Xero.
Worth knowing before you switch it on:
- Available on the Grow plan and above in Australia
- You can enable it per bank account, so trial it on one account rather than the lot
- A dedicated Reconciled page shows what it matched, and anything can be adjusted or reversed
- Low-confidence lines still come to you rather than being guessed
The construction caveat: automatic reconciliation is very good at deciding a $340 payment is a fuel expense. It has no idea whether that fuel belongs to the Scarborough job or the Joondalup one. If your job costing depends on how bank lines get coded, review the Reconciled page weekly for the first month rather than assuming it has your logic.
Xero's smart document capture also pulls data straight off supplier invoices without a separate syncing app in between, which removes most of the typing on merchant accounts.
Where a Tool Like Claude Earns Its Keep
Xero's AI is good at the work inside Xero. It is far less useful for the mess that exists before anything reaches Xero, and for the questions you actually want answered once it is in there. That gap is where a general-purpose assistant helps, and Claude handles long documents, tables and careful instructions well - including saying "this line is ambiguous" rather than inventing an answer.
1. Supplier statement reconciliation
The monthly ritual: a statement from the building supplier listing sixty-odd invoices and credits, against what is actually sitting in your payables. Missing invoices, a credit for returned materials that never came through, a delivery docket billed twice.
Paste both lists in and ask for a reconciliation. This is pattern-matching work, it is genuinely tedious, and AI is good at it.
2. Turning documents into importable data
PDF statements from an equipment finance provider with no bank feed. Twelve months of a director's card. A supplier who sends a spreadsheet with merged cells and a total halfway down. Ask for the output in the exact shape Xero expects and review the CSV before importing.
3. Reading your own numbers without exporting anything
Xero publishes an official MCP server on its own GitHub, which lets an AI assistant connect to your Xero organisation and read live data - profit and loss, aged receivables and payables, invoices, contacts, cash position. Instead of exporting a report and squinting at it, you ask a question and get an answer from the actual ledger.
For a builder, the useful questions are about cash timing: which builders or developers have quietly slipped from paying in 30 days to paying in 60, what is sitting over 60 days right now, what does the next fortnight look like against committed costs.
4. First-pass review before the BAS or the TPAR
Ask it to flag duplicates, suppliers coded to different accounts in different months, GST-registered suppliers with no GST claimed, and anything materially out of line with the same quarter last year. It is a first pass, not a review - but it catches the obvious things before your accountant bills you to find them.
Timely: your taxable payments annual report is due 28 August. If you paid contractors for building and construction services this year, that report is due within days. It is a good first test of AI-assisted review - ask it to cross-check your contractor payment list against the suppliers you have actually recorded, and flag anyone paid on invoice who is missing an ABN.
Prompts That Actually Work
Vague prompts produce vague output. These are specific enough to be useful - change the details to suit your business.
Supplier statement reconciliation. "Here is a supplier statement and my payables listing for the same supplier. List: invoices on the statement but not in my ledger; invoices in my ledger but not on the statement; and any amount differences over $1. Present as three tables with invoice numbers and dates. Do not adjust anything, just report."
Statement to CSV. "Here is a PDF statement. Extract every transaction into a CSV with exactly these columns: Date (DD/MM/YYYY), Description, Amount (negative for money out). Do not summarise, do not skip lines, do not round. If any line is unclear, list it separately at the end instead of guessing."
Job cost sense-check. "Here are the costs coded to job [X] and the original estimate by category. Show me actual versus estimate per category, the dollar and percentage variance, and the three categories most over. Then list any cost that looks unusual for this type of job. Flag only - do not reallocate anything."
Pre-BAS check. "Here is my transaction listing for the quarter. Flag: exact duplicates; near-duplicates within 3 days; any supplier coded to more than one account this quarter; expenses over $1,000 with no GST claimed; and any account more than 40% above the same quarter last year. Table, most material first. Do not correct anything."
Chasing progress claims. "Using my aged receivables, list everyone over 45 days, largest first. For the top five, draft a short follow-up email referencing the specific invoice numbers, dates and job names. Firm but professional, no threats, under 120 words each."
The pattern to copy: state the exact output format, say what not to do, and tell it to separate out anything it is unsure about instead of guessing. Nearly every bad outcome comes from a prompt that left it free to fill a gap.
Where It Breaks - the Construction-Specific List
This is the part worth reading twice, because these are the items that cost real money on a building business.
- Job cost allocation. AI cannot know that half a load of gyprock went to one site and half to another, or that a delivery invoiced in June was for a job that started in July. Get this wrong and your job profitability is fiction - which means your next quote is built on fiction.
- Work in progress and progress claims. Revenue recognition across a build, retentions held, claims raised versus work performed. These are accounting judgements about timing, not data entry, and they determine your reported profit.
- Retentions. Frequently miscoded as a bad debt or netted off a claim. They are neither.
- GST on property. The margin scheme, going concern, new residential premises, and mixed developments. AI will produce a confident, plausible, wrong answer here faster than anywhere else.
- Contractor classification for TPAR and super. Whether someone is a subcontractor or an employee for super purposes turns on the substance of the arrangement. Getting it wrong is a backdated liability across everyone you engaged the same way.
- Capital versus repair on plant. Rebuilt the excavator or repaired it? Immediately deductible or depreciated? Real dollars, and a judgement call.
- Business versus private. The ute, the phone, the trailer, the tools that go home. Only you know the actual split.
Watch out: AI does not know what it does not know. It will not stop and tell you a transaction needs professional judgement unless you have explicitly asked it to flag uncertainty. Build that instruction into every prompt you use.
The Record-Keeping Rules Do Not Change
Whatever you automate, the ATO still requires you to keep all records relating to the business with business and private portions distinguished; keep them so the information is not altered or damaged; keep them for five years, generally from when the record was prepared or the transaction completed; be able to produce them on request and explain your record-keeping system; and keep them in English or easily convertible to English.
Three consequences worth spelling out. AI-extracted data is not the record - the invoice or delivery docket is, so keep it. Electronic records must be indexed and extractable into a standard format like CSV; a chat history is not a filing system. And the integrity rule means you should be able to trace a line in your ledger back to the document behind it, so if AI sits in the middle of that chain, keep both ends.
A Routine That Survives a Busy Month
- Weekly, 15 minutes. Review Xero's Reconciled page for anything auto-matched to the wrong job or account. Clear the low-confidence queue. Photograph the week's dockets.
- Monthly, 30 minutes. Reconcile the major supplier statements. Run the job cost sense-check on any job that has moved significantly. Read the comparison to the same month last year.
- Quarterly, before the BAS. Run the pre-BAS check, fix what it flags, then have a human look at the GST treatment on anything to do with property or unusual supplies. This is the point where an error becomes a lodged error.
Before You Upload Anything
- Check the terms of whichever tool you use - business and enterprise plans generally do not use your data for model training, consumer plans may differ
- Strip client names and account numbers where the task does not need them
- Never paste TFNs, bank account numbers or passwords
- If you hold personal information about clients or workers, make sure your privacy policy reflects the tools you actually use
Frequently Asked Questions
Can AI do the bookkeeping for a construction business?
It can do most of the data entry, statement reconciliation and formatting, and Xero's automatic bank reconciliation now handles a large share of routine bank lines by itself. It cannot allocate costs to the right job, deal with work in progress or retentions, or decide GST treatment on property transactions. In construction the input work automates well and the judgement work does not.
Will AI code my costs to the right job?
Not reliably. Job allocation depends on knowledge that does not appear on the invoice - which site a delivery went to, which job a subcontractor was on that week, how a shared load was split. AI can suggest allocations based on past patterns and flag anything unusual, but the allocation itself needs someone who knows what happened on site.
Can Claude connect directly to Xero?
Yes. Xero publishes an official MCP server on its own GitHub that lets an AI assistant connect to your Xero organisation and read live data such as profit and loss, aged receivables and payables, invoices and contacts. That means you can ask questions in plain English and get answers from your actual ledger rather than a stale export.
Do I still need to keep dockets and invoices if AI has extracted the data?
Yes. The ATO requires the records themselves to be kept for generally five years, stored so the information is not altered or damaged, and produced on request. Extracted data is a convenience, not a substitute for the source document.
Will AI replace my accountant?
It replaces a large part of the bookkeeping input work, which is good news if you have been paying someone to do data entry. It does not replace advice on structure, job costing, tax planning or the judgement calls that determine what you actually pay. Cleaner data makes that conversation more useful, because it starts from numbers you can trust.
The Short Version
The builders getting real value from this are not the ones automating the most. They automated the boring 80% - the matching, the typing, the statement reconciliations - and spent some of the time they saved actually looking at job profitability.
The risk is not that AI is bad at bookkeeping. It is that it is good enough to produce a file that looks finished, in a business where nobody checked whether the costs landed on the right job. Automate the input, keep a human on the treatment, and you get both the evenings back and numbers you can quote from.
If you want a hand setting that split up - or a review of your Xero file, job costing and GST treatment before it hits a BAS or a return - that is the work we do for Perth builders, developers and trades. Book a free consultation and we will show you where the gaps are.
Disclaimer: This article provides general information only and does not constitute professional tax, legal or financial advice. Every business situation is different. Please consult a qualified adviser before making decisions based on this content.