Xero says AI success depends on financial control

Image credit: Getty Images

Xero says giving accountants and small businesses control over AI decisions will be essential if the technology is to handle sensitive financial work, including the creation of full end-of-year accounts.

Akankshu Dhawan, Xero’s SVP of Product & Tech, discussed the company’s approach at Xerocon 2026 in London, where Xero announced a series of new tools, products and services. He said the company’s development process is shaped heavily by feedback from customers, particularly as accounting software moves from manual workflows to increasingly automated ones.

Building financial tools around customer control

Dhawan said listening to customers is central to product development, even as Xero pushes into new technology. Some users may be comfortable adopting AI quickly, while others have moved from paper-based accounting to computers and are now being asked to work with AI systems.

Recommended reading

Forrester: Sovereignty is now an AI design requirement

“I think that’s the best way to build products. Sometimes you do need to push the envelope sometimes — as the world is changing so fast, technology is changing so fast, not everybody is going to be comfortable with it, but that’s also an opportunity to listen.”

Accounting and bookkeeping are natural areas for automation because they involve large volumes of structured data, repetitive procedures and complex filing requirements. Automating those tasks could free accountants and bookkeepers to focus on other work, but mistakes can be especially damaging for small businesses and startups.

Xero’s answer is to make the controls around its AI systems visible and usable, rather than allowing automation to operate without review.

“Our job is to help bring as much calm to this. As long as accountants feel they are in control, and we can give them the right set of controls, that can help them navigate this with as little anxiety as possible.”

How Xero’s JAX platform is intended to work

That approach is built into JAX, Xero’s AI platform. The company describes JAX as providing “accountable intelligence” and plans to infuse AI across its products through the Xero operating system.

Dhawan said Xero’s advantage comes from operating in a sector with a defined “source of truth”: financial information that customers can inspect, validate and correct. That creates a feedback loop in which users can confirm inputs and provide additional data before an AI-generated result is accepted.

“It was very obvious to me any industry where there is a source of truth, and you can build a full loop…where customers can validate the input and give you the data…as long as you can do that well, you will have the ability to disrupt.”

Xero has announced that JAX will be able to create full end-of-year accounts, extending the platform into work where errors could have serious financial consequences. Dhawan acknowledged that relying too heavily on AI creates risk if the accompanying safeguards are not designed properly.

The intended workflow is not for JAX to make an irreversible decision in the background. Instead, the system should present likely alternatives or flag an issue for review, allowing the accountant or business owner to approve the result.

“We don’t want our AI to just send you a notification one day saying 'I’ve sent your taxes to HMRC' without you having even looked at it…you need to strike the right balance.”

Training data and the human review step

Xero says its AI is trained on 20 years of data from more than five million customers. Dhawan said that scale should help JAX identify unusual or troublesome variants and narrow down possible answers.

In practice, he described a system that might tell a user that a transaction or accounting treatment is likely to be one of two options, then ask the user to check. That preserves a human decision point while reducing the amount of manual investigation required.

“Because we have this…we can give you an idea of 'it’s either this, or this' — now check. It’s much more productive, and you still save time, and you still keep the human in the loop.”

Dhawan said Xero’s promise that users will be “always in control” is not an optional feature but a basic expectation for accounting software. The company’s relationship with accountants and bookkeepers, he argued, depends on whether they can trust the advice presented to them and to their small-business clients, regardless of whether AI produced it.

“Honestly, that’s table stakes in our industry — I’m often blown away by promises that companies sometimes make.”

For Xero, the commercial case is tied to that trust: AI can create value for accountants, bookkeepers and their clients only if faster automation does not remove their ability to inspect and approve critical financial work. JAX’s planned end-of-year accounts capability will put that principle to a significant test.