Xero-Claude Partnership: Can It Replace a CFO?

Originally published on LinkedIn.

Last week, Xero and Anthropic announced a multi-year partnership to embed Claude directly into Xero's platform. I read the announcement with genuine interest. I use Claude daily in my work, building complex financial models and scenario analyses for clients. It is a remarkable tool. I have a lot of respect for what Anthropic has built.

But one line in the official press release stopped me.

Chris Ciauri, Managing Director of International at Anthropic, said: "This provides small businesses and their advisors with the kind of financial intelligence that used to require a dedicated analyst or CFO."

I think that framing sets the wrong expectation for small businesses, and I want to explain why, because the risk is not immediately obvious unless you have spent years doing what a CFO and qualified financial analyst actually does, rather than building the technology that sits around it.

Let me start with the AI itself

I use Claude to build financial models. Not occasionally. Daily. For real clients, with real numbers, informing real decisions.

And I can tell you with complete confidence: it makes mistakes. Not always. Not even frequently on straightforward tasks. But on the kind of nuanced, multi-layered financial analysis that a business owner needs to make a significant decision, errors occur. A misapplied assumption in a cash flow model. An incorrect treatment of a timing difference. A margin calculation that looks right but uses the wrong denominator.

These are not obvious errors. They are the kind that a qualified finance professional catches because they know what the answer should look like before they see it. A business owner relying on a chatbot has no such reference point. They see a confident, well-formatted answer and they act on it.

That is a problem the technology cannot solve by itself.

The deeper problem: most people building these tools do not fully understand what a CFO does

I say this without any disrespect to the engineers and product teams behind these announcements. They are solving genuinely hard problems. But there is a pattern in the AI and fintech space of senior technology executives making statements about replacing financial analysts and CFOs that reveal a fundamental misunderstanding of what those roles actually involve.

A CFO is not an answering machine for financial questions. The job is not to retrieve data. It is to make judgements, under uncertainty, with incomplete information, in the context of a specific business, its people, its market, and its history.

The claim that AI can replicate that tells me the person making it has a clear picture of what finance produces, but a much hazier picture of what it requires.

What we actually find when we start working with a client

At CompassPoint, we work with businesses as their CFO function. These are not disorganised operations. Many of them already have accountants and bookkeepers in place. They have been using Xero or a similar platform for years. On paper, their financial data exists.

In practice, almost every single engagement begins the same way. The first few weeks are spent cleaning the accounting data, making it compliant with IFRS, restructuring the chart of accounts, and correcting errors that have compounded silently over time. Costs in the wrong categories. Revenue recognised in the wrong periods. Personal expenses sitting inside the business. No consistent treatment of intercompany transactions. Accruals either missing or applied incorrectly.

This is not a reflection of negligence. It is a reflection of the fact that bookkeeping and accounting are specialisms, and most small business owners are running their business, not studying for their ACCA qualification.

Now ask yourself: if this is the data quality we encounter in businesses with existing finance support, what is the data quality sitting inside Xero for the millions of small businesses who are managing their own books?

An AI tool does not audit your data before answering your questions. It reads what is there and responds with confidence. Garbage in, confident garbage out.

And then comes the judgement problem

Even if the data were perfect, which it is not, the most important part of a CFO's value cannot be automated. Not yet. Not meaningfully.

The real work begins after the numbers are corrected. A business owner looks at their management accounts and asks: can I afford to hire? How much should I be spending on marketing? Should I take a dividend this year or reinvest? Should I expand into a new market or consolidate what I have?

These are not questions you answer by retrieving a figure from a database. You answer them by understanding the business, its cash position, its risk appetite, its growth trajectory, the owner's personal financial situation, the market conditions it is operating in, and a dozen other contextual factors that exist nowhere in Xero.

That is the judgement layer. It is built from years of experience, professional qualification, and deep familiarity with a specific business. No amount of reasoning capability in a language model substitutes for it.

The autopilot analogy

Autopilot exists in every modern commercial aircraft. It is sophisticated, reliable, and handles the majority of a flight's technical demands with more consistency than any human could manage alone.

But there is a trained pilot in that cockpit. A qualified professional who understands what the instruments are telling them, can recognise when something is wrong, and is accountable for every passenger on board.

Nobody is seriously suggesting we remove the pilot because the autopilot is impressive.

The Xero and Anthropic integration is, in the right hands, a useful co-pilot for financial management. But a co-pilot requires a pilot. And right now, the announcement is being framed as though the plane can fly itself.

What this means for small business owners

The Xero and Anthropic partnership is commercially interesting and technically real. Used by a business with clean, well-structured data and a qualified finance professional reviewing the outputs, it will add genuine value.

Used by a founder with inconsistent bookkeeping, no qualified oversight, and a significant financial decision to make, it is a liability dressed as an asset. The technology will answer the question. It will just be answering it with the wrong data, and nobody in the room will know.

What a CFO brings to a business cannot be downloaded or integrated. It is built through years of training, pattern recognition developed across many businesses, and the professional accountability that comes with a qualification. No product announcement changes that.

Before any AI tool can tell you where your business is going, someone qualified needs to ensure it accurately knows where it has been.