Most bookkeepers who run Xero have already set up bank rules. They match the regular payments, the recurring suppliers, the predictable subscriptions. On a clean week those rules clear most of the bank feed on their own. The trouble is the rest of it, and the handful of matches inside the tidy majority that were quietly wrong.
Bank rules are pattern matchers. They fire when a transaction looks like one they have seen before. They cannot tell you that a supplier invoice was paid twice, that a refund was coded as income, or that a $4,200 payment sitting in the feed belongs to a job you have not raised yet. That judgement is where reconciliation actually lives, and it is the part rules were never built to do.
Where rule-based reconciliation stops working
A bank rule works on a single transaction in isolation. It has no memory of last month, no view of your open invoices, and no sense of what a number should be. That blind spot shows up in predictable places:
Duplicate payments, where the same supplier is paid twice within a few days and both entries match an existing bill.
Miscoded refunds and reversals that land as revenue instead of reducing an expense.
Timing mismatches, where a customer pays a rounded amount that does not tie to any single invoice.
Personal spending run through the business account that a rule happily codes to a plausible expense category.
Bank fees and merchant charges that drift into the wrong account because the description changed slightly.
None of these are exotic. They are the everyday noise of a real Australian business banking feed, and every one of them survives a rules-only reconciliation because the rule did exactly what it was told.
What Claude does in a reconciliation
Claude reads the whole picture at once. Given read access to the Xero bank feed, the open invoices and bills, and the recent coding history, it works the way an experienced bookkeeper does: it looks for what does not fit. Rather than matching one transaction at a time, it compares the feed against everything else in the file and flags the entries that need a human decision.
In practice that means a short list at the end of each run. Not five hundred green ticks, but the eight or ten items that are genuinely ambiguous, each with a plain-English note explaining why Claude stopped on it and what the likely explanation is. A $3,180 payment that matches two different bills. A Stripe payout that nets three invoices and a fee. A duplicate to the same ABN two days apart. You approve, correct, or investigate, while the routine matches still flow through Xero's own rules untouched.
It explains its reasoning, every time
The difference that matters for a bookkeeping practice is auditability. Claude does not silently reconcile. It names the transaction, the candidates it considered, and the reason it chose to flag rather than match. That trail is what lets a senior reviewer sign off in minutes instead of re-checking the whole ledger, and it is what keeps the work defensible if the ATO ever asks.
A worked example: the $45,000 nobody caught
One Sydney trades business we looked at had a clean-looking Xero file. Bank rules were matching, the feed was reconciled to zero, and the monthly numbers looked fine. Running Claude across six months of history surfaced $45,000 of problems that rules had waved through: a supplier paid twice on a large materials order, three customer refunds booked as sales income, and a director's car expenses coded to plant and equipment. The GST position was wrong as a result, which meant the last two BAS lodgements were overstated.
The point of the example is not the size of the number. It is that none of it looked wrong to a rule. Every transaction matched a pattern. It took reading the transactions against each other to see the errors, and that reading is cheap to run and expensive to skip.
Setting it up without handing over your Xero login
You do not give Claude your Xero password, and you should be wary of any provider who asks for one. The reconciliation runs on read access to the data it needs, produces a review list, and leaves the actual posting to a person inside Xero. Nothing is changed in your accounts without a human approving it.
Read access to the bank feed, contacts, invoices and bills is enough for the review to work.
Claude proposes and your team disposes. Corrections are made by a person in Xero, not by an agent acting on its own.
The output is a document you can keep, so the reasoning behind every flag stays on file.
For a practice running dozens of client files, this is the difference between a reconciliation that clears the feed and one that stands up to review.
What it is worth
A bookkeeper spends roughly two to four hours a month per client chasing reconciliation exceptions by hand. At a modest $80 an hour that is $8,000 to $16,000 a year across ten clients, before you count the cost of an error that reaches a BAS. Cutting the manual exception hunt down to a reviewed short list does not remove the bookkeeper. It moves them off the tedious part and onto the judgement that clients actually pay for.
If you run Xero for your own business or for clients and want to see what a Claude reconciliation review turns up on a real file, we can walk you through it. Book a short call and bring one month of history you think is already clean.
