By Andrew Erkins | 23 February 2026
Xero has changed more in the past 18 months than in the previous five years combined. If you're using it to run your business finances and you haven't looked up from the bank reconciliation screen lately, you're missing what's happening underneath you
We manage Xero files for hundreds of Australian businesses. We're inside the product every day, across every industry we serve, and the shift we've seen since late 2024 is unlike anything Xero has done before. They've built an AI superagent, acquired an analytics platform, partnered with OpenAI, and at a February 2026 investor briefing, their CEO described the company's ambition as evolving from a system of record to a system of action to a system of decision-making. That's not the kind of language you use when you're tweaking features. That's a company rebuilding what it is
Here's what's actually live right now, what's coming, and what it means if you're a business owner trying to get more out of your numbers. If you want the broader picture on how AI is reshaping accounting across Australia, we've written about that separately
Auto bank reconciliation changes everything about the daily grind
JAX - Just Ask Xero - launched at Xerocon Brisbane in September 2025. Xero calls it a "financial superagent," which sounds like marketing until you see what it actually does to a bank reconciliation workflow
The auto bank reconciliation feature went into global beta in November 2025. At Xero's investor demo in February 2026, they showed a business with 118 transactions to reconcile. JAX handled 113 of them automatically. Three were left for human review. Xero claims accuracy above 97%
That is an investor demo, though, and demo environments tend to feature clean data. We are testing auto-reconciliation across a snapshot of real client files right now to understand how it handles the kinds of transactions that actually trip bookkeepers up - split payments, foreign currency receipts, contra entries, manual journal adjustments. Early signs are that it handles straightforward bank feeds well, but it needs training and oversight on anything more complex. The gap between demo conditions and a live file with five years of messy history is real
Xero cites early users reporting savings of four to seven hours a week on reconciliation. Those numbers come from Xero's own case studies, and they are plausible for businesses with clean, straightforward transaction flows. For more complex files, we expect the time savings to be real but less dramatic - the AI handles the routine matches quickly, but someone still needs to review the exceptions, train the system on business-specific patterns, and catch the ones it gets wrong. We are tracking this across our own client files to see where the real numbers land
What's clever about how Xero built this is the layered approach. JAX doesn't just do simple matching. It runs four mechanisms at once: your existing bank rules, predicted matches against invoices and bills already in Xero, patterns learned from your own reconciliation history, and - this is the interesting one - anonymised patterns from across all 4.5 million Xero subscribers. So even a brand new Xero file benefits from the collective intelligence of the entire platform. The more you use it, the sharper it gets for your business specifically, but it doesn't start from zero
The controls are sensible too. You opt in at the bank account level, so you can turn it on for your main transaction account and leave it off for accounts you want to reconcile manually. There's a new "Reconciled" page that shows exactly what JAX did and why - which rule it applied, which document it matched, whether it used your history or a platform-wide prediction. You can undo anything with a click
Xero's CEO mentioned during the rollout that some business owners enjoy the "dopamine hit" of reconciling transactions themselves, which is why auto-reconciliation is opt-in at the account level. You can turn it on for your main transaction account and leave it off for accounts you want to review manually
JAX also has something called Assure built in, which cross-checks its outputs against Xero's accounting logic before applying them. Low-confidence suggestions get flagged for human review instead of being auto-applied. That matters more than it sounds. A misclassified expense in January becomes an incorrect Business Activity Statement (BAS) in March. Errors in reconciliation don't just sit there - they compound through everything downstream. Assure is Xero's attempt to solve the trust problem that keeps businesses nervous about letting AI touch their books
If you're interested in how this kind of automation fits alongside human expertise rather than replacing it, we've written about what actually happens when you automate your accounts
The OpenAI partnership brings context from outside your ledger
This one flew under the radar for a lot of business owners. Xero has partnered with OpenAI to give JAX the ability to pull information from outside your Xero file - tax regulations, market data, interest rates, industry benchmarks - and combine it with your financial data in a single answer
The demo Xero showed is worth describing, with the caveat that it was a controlled demonstration. A business owner asked JAX whether they could afford to finance a $20,000 van purchase. JAX looked at the business's cash position and profit trends inside Xero, then went out via OpenAI to research current vehicle financing rates and terms, and came back with a combined recommendation
That's the kind of question business owners actually ask. Can I afford this? What happens to my cash if I take on this lease? Should I hire now or wait? The answers always require both your numbers and the world around them. If JAX delivers on this consistently in production, it would be genuinely useful. The question is whether the external data it pulls in is accurate, current, and relevant to Australian lending markets specifically - not just a generic answer assembled from whatever OpenAI's training data contains. We are watching this one closely before recommending clients rely on it for financial decisions
The financial insights agent built on this partnership is seeing over 90% accuracy in testing. It can break down revenue by category, analyse expense trends, and run scenario models using conversational language. And for accountants, Xero is building a separate agent through its upcoming Partner Hub that can query across an entire client base and generate meeting briefs automatically. We're watching that one closely
Syft Analytics is now part of Xero - and most people haven't turned it on
In September 2024, Xero acquired Syft Analytics for USD $70 million. By January 2026, it went live globally as "Xero Analytics (powered by Syft)" - built into every business plan at no extra cost
Syft was already the most popular reporting app in the Xero ecosystem. It won Best Management Reporting App for Xero Users in the State of Stack 2026 report. Now it's not an add-on. It's native. And from our conversations with clients, most Xero users either do not know Syft is there or have not activated it Invite us as your Xero Advisor
