How to Categorize Transactions in Xero
A working sequence for coding bank feed lines in Xero, deciding when to match instead of code, and setting a review rule for automated suggestions before you accept them.
Categorization in Xero happens against the bank feed. Connect the bank accounts to Xero first, then give every imported line one of two outcomes: match it to a record your books already hold, or code it to an account. Reconciliation is the separate check that follows, comparing your own record of transactions against the bank's so the entries line up. Coding prepares the line, matching links it to something already recorded, and reconciling closes it out.
What a category decides
Xero's guidance on recording transactions covers recording, categorizing, and managing them together, which is a fair description of the job. A category is the account a transaction lands in, and that account decides where the amount shows up in the profit and loss or the balance sheet. Get it wrong and the reports are wrong, even when the bank balance agrees.
Reconciliation is the quality control on top. Xero describes it as comparing your record of transactions against the bank's, with your entries matching theirs. Doing that regularly catches errors early and supports accurate financial reports, which is why coding and reconciling belong to the same regular routine rather than a scramble at year end.
Start where the data enters: the bank feed
Nothing can be categorized until the transactions arrive. Xero's reconciliation guidance starts from connecting your bank accounts, and the connected feed is the queue you work through.
Before coding anything, read the line for what it tells you: date, amount, direction of the money, and whatever text the bank supplied. Bank descriptions are often abbreviated, or they carry a payment processor's name instead of the merchant's, so a line that looks unfamiliar may be an ordinary supplier routed through a different rail. When the description is not enough on its own, the answer usually sits in a document, a contract, or a short question to the client. Asking costs less than guessing and unwinding it two months later.
Choosing the account
Work from the chart of accounts you already have. An existing account that a reasonable reader would expect the cost to sit in beats a new one, because every extra account splits history and makes comparatives harder to read.
A few tests settle most cases. Does this cost behave like something already in the chart? Would a reader of the accounts be surprised to find it there? Do you need it separated on the face of the reports, or only for an internal question a filter can answer? A new account earns its place when the answer to the last question is the reports.
Ownership of the decision matters as much as the decision. Xero publishes guidance on recording, categorizing, and managing transactions, but the chart belongs to the business and its accountant, and the account you pick becomes precedent for every similar transaction after it.
Match an existing record, or code a fresh line
A payment against an invoice you already raised is not new information. The invoice or bill sits in the books already, so the bank line should be matched to it. Coding it to an account instead records the same money twice, once in the original document and once as a new transaction.
Reconciliation compares your records against the bank's, and duplicates are exactly what that comparison exposes, usually as a revenue or expense figure nobody can explain. If a line looks like a payment against something, look for the something first. Transfers between accounts owned by the same business follow the same logic: the money left one connected account and arrived in another, so it belongs in the books once as a movement rather than twice as income and cost.
Booke's Xero page describes AI Bookkeeper pairing payments with the corresponding bills and invoices as part of document matching and reconciliation prep. Partial payments, batched settlements, refunds, and foreign currency amounts are where matching gets harder and a person still has to look.
How does AI categorize transactions?
The two automated layers covered here both propose a category, and the difference between them matters.
Xero markets JAX, its AI finance partner, as categorizing and matching transactions for you once your bank accounts are connected, with availability stated as the Growing price plan and above. That plan limit matters when you work across client files, because the same workflow is not available everywhere.
Booke sits in a different position. QuickBooks Online or Xero stays the accounting system of record, and the AI bookkeeper for Xero works with the bank feeds already connected there rather than adding a second ledger. Its Xero page describes learning from up to two years of transaction history, picking up how categorization has been handled in that file, and applying those patterns to new transactions before reconciliation review. The automation is described as running daily to keep the file moving.
Confirm the JAX behaviour Xero markets, along with its stated availability on the Growing price plan and above, against Xero's current published pages before you rely on it in a client file. Booke routes low-confidence transactions to your team, and your team keeps control of the review decision.
Rules, AI suggestions, and manual coding compared
Four ways a category gets chosen when you work a Xero file, what each one suits, and the part of the decision that stays with a person in every case.
Approach: Deterministic rules; How the category is chosen: Repeats a choice you defined in advance and behaves identically every time; Where it fits: Work where the choice does not vary from one instance to the next; What stays with you: Defining which records are eligible, what evidence is required, and when a person reviews the result
Approach: Xero's JAX; How the category is chosen: Marketed by Xero as categorizing and matching transactions for you, on the Growing plan and above; Where it fits: Routine feed work inside Xero, where the plan allows it; What stays with you: Checking the outcome instead of assuming it
Approach: Booke AI Bookkeeper; How the category is chosen: Applies patterns learned from up to two years of transaction history before reconciliation review; Where it fits: Firms keeping QuickBooks Online or Xero as the system of record; What stays with you: Reviewing the low-confidence items routed to your team
Approach: Manual coding; How the category is chosen: Your own judgment on a single line; Where it fits: One-off, ambiguous, or poorly documented transactions; What stays with you: The whole decision
When a rule beats a suggestion
Booke's bookkeeping automation software page separates the parts that usually get bundled together: rules repeat deterministic choices, OCR pulls data out of documents, and AI recommends a category. A complete setup also says which records are eligible, what evidence is required, and when a person has to review the result.
That separation gives a usable rule of thumb. Where the input is stable and the answer never varies, a deterministic rule is the better tool, because it behaves the same way every month and it is easy to audit. Where the input varies, a suggestion helps more than a rigid instruction, because it adapts instead of forcing every line into a shape that stopped fitting. The failure modes differ too. A bad rule is silent and repeats itself. A weak suggestion is visible, provided somebody is looking.
Confirm in Xero's current documentation what automation controls exist in the file you are working in before building a client process around them.
Rules repeat a defined choice; Booke routes low-confidence recommendations to your team for review.
Reviewing low-confidence suggestions
Booke's Xero page states that low-confidence transactions are brought to the team for review and that the team keeps control of review decisions. Booke does not publish a numeric confidence threshold, so treat the routing as a working boundary and never quote a figure to a client.
A review that means something asks a few questions. Does the account match what the underlying document says happened? Is this a payment against a record already in the ledger? Does this counterparty have history in the file, and does the line agree with it? When the answer needs information you do not have, the item stays open and the question goes to the client rather than into a plausible-looking account.
With the browser extension installed, AI Bookkeeper shows read-only statuses beside bank transactions, so you can see what state an item is in while you work. The statuses report; they do not act. Booke's own copy is direct about the limit: repeatable work can be handled within a defined scope, unresolved work stays visible to people, and AI bookkeeping does not remove every accounting decision or move professional responsibility onto software.
Finding and correcting bad categories
Miscoded transactions usually show up in the reports before anyone finds them in the feed: an expense account that jumped for no reason, a holding account carrying a balance it should not, one supplier sitting under two different accounts in the same quarter. Reviewing your records against the bank statements on a regular rhythm catches this early.
Corrections inside a period that has been reconciled, reported, or filed need more care than a quick edit. Note what changed and why, then check whether anything downstream depends on the number you are about to move: a sales tax return, a management report, a covenant calculation.
Corrections also carry information. Booke's Xero page states that each approval and correction is used to improve future work, so a fix you make today feeds the patterns applied later instead of being discarded.
The same job in QuickBooks Online
Firms running mixed client stacks should not assume the workflows transfer. Booke's automation page puts it plainly: the two platforms follow the same controlled automation principles, but their supported reconciliation scope is different. The accounting judgment behind a category carries over; the supported scope does not, so evaluate each workflow on its own terms. If you also work in QuickBooks Online, our guide to categorizing transactions in QuickBooks Online covers that side.
A month-end pass that holds up
Run the same sequence every month:
Confirm the connected bank accounts have brought in a complete feed for the period.
Clear matched items first, so payments against existing invoices and bills stop competing for attention.
Code the remaining lines, using an existing account unless the reports genuinely need a new one.
Work the low-confidence queue on its own, with the documents open.
Collect what is still unanswered and send it to the client as one query rather than five.
Reconcile the accounts against the statements and investigate any difference before you close the period.
Write down the decisions you made on unusual items, so next month's version of you is not re-deciding them.
Regular reconciliation is what makes the reports trustworthy, and the queue of unresolved items is the part you should not clear under time pressure. If you are weighing whether to hand the repetitive share of this to software, AI bookkeeping software sets out the operating model and where the review boundary sits.
What to re-check before relying on this
Platform features and plan availability move. The Xero behaviour described here, including JAX and its stated availability on the Growing price plan and above, was last checked on 4 September 2026 against Xero's published pages, and Booke's described product behaviour was last checked on the same date. Before you build a client process on any of it, confirm the current position with the vendor, particularly anything tied to a pricing plan.
Common questions about categorizing in Xero
Is Xero's AI reconciliation available on every plan?
No. Xero states that JAX, its AI finance partner, is available on the Growing price plan and above. On files below that plan, categorization and matching stay with whoever is doing the reconciliation, plus any third-party tooling connected to the file.
What is the difference between matching a transaction and coding it?
Matching links a bank line to a record the books already hold, such as an invoice or a bill. Coding creates the accounting entry for a line that has no existing record. Reconciliation then compares your records against the bank's, so a line that was coded when it should have been matched tends to surface there as a duplicate.
How should low-confidence suggestions be reviewed?
Treat them as a queue, not a rubber stamp. Booke's Xero page describes low-confidence transactions being brought to the team for review, with the team keeping control of the decision. Work them with the supporting document open, and leave anything you cannot support unresolved until the client answers.
How do bank rules and AI recommendations differ?
A rule repeats a choice you defined in advance and behaves identically every time. AI recommends a category and can adapt when the input varies. Rules are easier to audit; suggestions cope better with messy inputs. Both still need a person to confirm the outcome, and a complete setup also defines which records are eligible and what evidence is required.
Does Booke AI replace Xero?
No. Booke's published position is that QuickBooks Online or Xero remains the accounting system of record, with AI Bookkeeper working from the bank feeds already connected there while people keep final review and close responsibility.
Do my corrections change future suggestions?
Booke's Xero page states that each approval and correction is used to improve future work. Treat a correction as input to the pattern rather than an immediate switch, and keep reviewing similar items after you make it.
Explore Booke AI workflows
Sources
