AI for Xero

No, and we would rather say so plainly than let the wording imply otherwise. We are an independent Australian product. We are not affiliated with, endorsed by, certified by or partnered with Xero Limited, and Xero is their trademark, not ours. Our software connects to a Xero organisation the way any third-party application does — through Xero’s published interfaces, with your explicit authorisation, and only to the extent that authorisation permits. We think the distinction matters because the alternative is a marketing habit that leaves firms genuinely confused about who is accountable when something goes wrong. If you want to know exactly what our software will and will not be able to do against your specific Xero setup and plan, we will confirm that with you during onboarding rather than guess at it on a web page.

The AI reads the file to understand it, and writes only what you have authorised it to write — with a human in the loop. Reading is the larger half: the chart of accounts, contacts, historical transactions, tax treatments, tracking categories and how each supplier has actually been handled in that organisation. That history is what makes coding accurate, and it is why a file with two years of consistent treatment gets better results than a blank one. On the writing side, the AI proposes transactions and a person reviews and approves them. We deliberately do not offer a mode where it silently posts a high volume of unreviewed entries into a live client file, because the recovery from that going wrong is worse than the labour it saves.

No, and a firm that expects that will be disappointed. Xero is the ledger — it is where the file lives, where the bank feeds land, where the reports come from and where your team works. Nothing about our product changes that, and we have no interest in becoming a second ledger. What we add is the layer that decides what a document is and how it should be treated before it becomes a transaction, and the layer that chases the client for what is missing. Xero has its own automation features and they keep improving; where they cover your need, use them. The honest test at the consultation is whether what you are struggling with is actually addressed by what you already own, and sometimes the answer is yes.

Access is granted by you, per Xero organisation, through Xero’s own authorisation flow — we never ask for a Xero username and password, and any vendor who does should be shown the door. You authorise each organisation you want connected, and you can disconnect it at any time from your side. Practically, for a firm this means the connection is per-client-file rather than a blanket key to everything you touch, which matters when you are handling other people’s financial records under the Privacy Act 1988. Access on our side is controlled per staff member and every action is logged — what was proposed, who reviewed it, what they changed.

Yes — tracking is treated as part of the coding decision rather than an afterthought. Where an organisation uses tracking categories, the AI codes against them using the same source it uses for everything else: how transactions of that kind have actually been tracked in that file. Same for splits across categories, and for the recurring patterns your team has already established. If a transaction does not fit a known tracking pattern, it is flagged rather than assigned to whatever seems closest — which is the behaviour that matters, because a wrong tracking allocation is invisible in the P&L and only surfaces when a client asks why a job looks unprofitable.

That is the case it is built for, and it is a meaningfully different problem from automating one business’s own file. Each organisation is a strictly isolated context with its own chart of accounts, its own supplier history and its own rules — a bill captured for one client cannot be posted to another. The practice view sits above them and shows where attention is needed: which files have exceptions waiting, which clients are not responding to document requests, where the queues are building. Rollout is normally staged rather than switched on across the whole book at once, starting with a handful of files that have clean history so you can see the behaviour honestly before extending it.