You're building a cash flow forecast in Excel. Your numbers are three days stale because you exported them from your accounting software last Tuesday, and half the formulas broke when a customer paid an invoice you hadn't accounted for. Now imagine that same spreadsheet updating itself, live, while you type a question into the sidebar and get an answer sourced straight from your books.
That's the pitch behind Xero's newest move: embedding its AI agent, JAX, directly into Microsoft 365. Announced in July 2026, the integration puts real-time accounting data inside Excel, Word, Outlook, and Teams — the tools small business owners and their bookkeepers already live in every day. It's a preview of where accounting software is headed, and it's worth understanding whether or not you use Xero.
What JAX Actually Is
JAX stands for "Just Ask Xero." It's Xero's AI financial agent, designed to sit on top of your accounting data and answer plain-language questions about it: who owes you money, what your top customers are, how this quarter's expenses compare to last quarter. Before this announcement, JAX lived inside Xero itself. The Microsoft 365 integration is what moves it into the tools you already have open all day.
What Changes Inside Excel, Outlook, and Teams
The integration works through Microsoft 365 Copilot, and it does two distinct things depending on where you're working:
In Copilot Chat, JAX answers questions in plain text and lets you confirm actions without leaving the app. Ask "which invoices are overdue?" and get a real answer, with a direct link back into Xero to fire off a payment reminder — no tab-switching, no re-authenticating, no digging through a report menu first.
In Excel, the effect is more concrete: live Xero data lands as structured tables you can actually build on. That means a multi-quarter profit-and-loss breakdown, or a list of overdue invoices complete with amounts and due dates, drops straight into a spreadsheet as real cells — not a static export. The manual ritual of exporting a CSV, cleaning up the formatting, and re-doing it next week because the numbers changed again is the specific friction this is meant to remove.
Xero says Outlook and Teams support is coming next, with the long-term goal of grounding Word documents and PowerPoint presentations in live Xero data too — so a monthly report or a board deck could theoretically stay in sync with the actual ledger instead of a snapshot someone remembered to update.
The feature enters public preview in August 2026. Xero states that financial data shared through the integration is used only for the active session and is not used to train Copilot's underlying AI models — an important distinction if you're evaluating whether to turn this on for a business with sensitive customer or payroll data.
Why This Matters Even If You Don't Use Xero
This isn't really a Xero story — it's a preview of where every major accounting platform is headed. Microsoft Copilot for Finance already integrates with several ERP and accounting systems, and Gartner has projected that the large majority of finance functions will run at least one AI-enabled tool by the end of 2026. QuickBooks has its own AI assistant with similar ambitions, and Intuit, Sage, and a wave of smaller fintech tools are racing to embed the same kind of conversational access into the apps people already have open. The pattern is consistent: accounting data is moving out of a single dashboard and into the general-purpose tools — spreadsheets, email, chat — where financial decisions actually get made.
For a small business owner, that shift has a real upside. Industry surveys put AI accounting tool adoption among small businesses at roughly a quarter today, growing fastest among companies with five to twenty employees — exactly the range where there's enough transaction volume to make manual reconciliation genuinely painful, but rarely enough headcount to have a dedicated controller. Bookkeepers and finance teams using AI copilots report reclaiming several hours a week when routine questions — "what's outstanding," "what did we spend on software last month" — get answered instantly instead of requiring a login and a report export. If you're spending real time each week reconciling exports or rebuilding the same Excel summary from scratch, tools like this are aimed squarely at you.
It's also worth noting where the value concentrates. The highest-ROI use cases so far are high-volume and low-judgment: chasing overdue invoices, categorizing routine expenses, summarizing this month versus last month. If your business processes a lot of invoices, juggles several bank accounts, or burns significant hours on reconciliation every month, that's where an embedded AI layer pays for itself fastest. If your bookkeeping is already simple — a handful of transactions a month — the novelty may outpace the actual time saved.
What to Watch Before You Turn It On
Convenience and correctness aren't the same thing, and embedding AI directly into your financial workflow raises a few questions worth asking before you flip the switch on any platform's version of this:
Where does the data actually go, and for how long? Xero's claim that data is session-only and excluded from model training is the right answer — but it's worth confirming for any tool you connect to your books, and re-checking if the vendor's terms change. Financial data is a higher-stakes leak than almost anything else in your business: exposed records can enable fraud and identity theft, not just embarrassment.
Who's allowed to ask what? An AI agent with broad read access to your accounting system is only as safe as the permissions around it. Make sure whoever sets this up scopes access the same way you'd scope a new employee's login — not everyone in Teams needs to be able to ask about payroll.
Where does it get things wrong? AI tools are noticeably more reliable on descriptive questions ("what's overdue") than on judgment calls (estimated tax payments, worker classification, which expenses are actually deductible). Treat anything in that second category as a draft answer that a human — you or your accountant — reviews before it becomes a decision.
Does convenience quietly increase your dependence on one vendor's cloud? The more of your financial reasoning happens inside a proprietary AI layer, the harder it becomes to leave, audit independently, or reconstruct your numbers without that vendor. That's a fair trade for a lot of businesses. It's worth being conscious of it as a trade, not a free upgrade.
A Quick Checklist Before You Adopt Any Embedded AI Tool
Whether it's JAX, Copilot for Finance, or whatever QuickBooks ships next, the evaluation questions are the same:
- Confirm the data-use terms in writing. "Session-only, not used for training" should be stated explicitly in the vendor's documentation, not implied. If it isn't written down, assume the opposite.
- Scope access before you turn it on. Decide who in your organization can query financial data through the AI layer, and set permissions accordingly — don't leave it wide open by default.
- Pilot it on read-only questions first. Let it answer "what's overdue" for a few weeks before you let it draft anything that touches tax filings, payroll, or contracts.
- Keep a human review step on anything judgment-based. Categorization suggestions, deduction calls, and worker-classification questions should land as a draft for your accountant, not a final answer.
- Ask what happens if you leave. Can you export your full transaction history in an open format, or does years of context live only inside the vendor's AI layer?
The Bigger Shift: Bookkeeping Is Becoming Ambient
Whatever platform you're on, the direction is clear: financial data is moving out of a single locked dashboard and into the everyday tools where decisions get made. That's a genuine improvement for a business owner who's tired of context-switching. But it also means the underlying question — where does my financial data actually live, and can I trust and verify it — matters more, not less, as more layers of AI sit between you and your ledger.
Keep Your Books Transparent as AI Gets More Involved
As accounting tools get more automated, it's worth knowing exactly what's happening to your financial data underneath the convenience. Beancount.io is built on plain-text accounting: your books live in version-controlled text files you can read, audit, and back up yourself — no proprietary database, no black-box AI layer standing between you and your numbers. Get started for free and keep full control of your financial data, however many AI assistants end up reading it.