Nobody wants to automate bookkeeping more than I do.

You might assume that, because we offer bookkeeping as a service, we’ve got an interest in protecting that work from automation.

We don’t.

Bookkeeping takes up a huge amount of my team’s time. Collecting data, posting data and matching data – and that’s just to reconcile the bank. After that we’ve still got to post all the adjustments needed to make the numbers meaningful, including prepayments, accruals and deferred income.

For context, in the quarter to June 26, our team spent 31% (almost 1 in 3 hours) of their available time doing work that could potentially be done by a machine.

So whenever we can automate repetitive work safely, we want to. And when we find genuine efficiencies, we want to pass those on to our clients. They’re right to look to us to continually find faster and smarter ways of working.

So what changes when bookkeeping becomes automated?

I want our team to be plumbers connecting the financial systems of a business. Their job should be designing the pipework, checking for leaks and fixing blockages, not manually carrying buckets of data from one place to another.

When we’re spending less time pushing data around, we’ve got more time to do the things that you tell us you need from us. Accountability, challenge, insight and supporting you while you navigating complicated situations and transactions.

I’d love for our client managers to start every month with up to date, accurate financial information already in front of them. That should be the starting point of their work, not the majority of their work.

So, when new technology promises to automate bookkeeping, I’m every bit as excited as agency owners are. Maybe more.

But I’m approaching adoption with a healthy dose of caution.

Xerocon 2026 gave us a glimpse of the future.

I’ll be honest in that I had a bit of a heads-up on what was coming as I’m a member of Xero’s XPAC (Xero Partner Advisory Council). The capabilities that AI is bringing makes me excited and nervous in equal measure.

Xero presented a vision of a finance function in which AI doesn’t just enter information. It collects documents, reconciles transactions, chases debtors, connects systems, spots potential problems and suggests what businesses should do next.

These 3 announcements are the most significant if you’re running an agency.

1. JAX

JAX is Xero’s AI tool. It can already auto-reconcile your bank account by matching transactions to bank feeds in real time.

Jax will also be able to automate debtor chasing in a more sophisticated way than is possible now. It’ll be able to consolidate several outstanding invoices into one reminder, as well as adjusting its chasing approach based on a customer’s past payment behaviour. These are things that Chaser already can do, but there are benefits of having this functionality sitting within Xero.

Then there’s Smart Document Capture, this is like Dext but not quite as sophisticated yet. Xero subscribers already have free access to Hubdoc, so I’m slightly fuzzy on exactly how Smart Document Capture and Hubdoc will sit alongside each other but I’m sure it will become clear in time.

2. XeroForce

This is Xero’s custom AI agent builder. It can be used to connect Xero to third party tools. Early access is already available.

As with anything, what you get out depends on the quality of what you put in so, as far as Xeroforce goes, it’s very exciting for accountants but I suspect Agencies won’t leverage it unless they have a fairly sophisticated inhouse finance team.

3. Insights

Jax promises to start flagging insights, like where cashflow might be getting squeezed, and sharing ways to address problems. This might include things like delaying non-critical supplier payments or arranging funding.

That could be incredibly valuable. But be cautious incase insights are based on inaccurate data.

If the Xero is not fully or correctly reconciled there’s a real risk that the advice Jax surfaces is wrong

The dangers of relying on AI for your bookkeeping

While AI has the potential to save you a lot of manual work. Implementing it without the right controls in place can cause you real world problems.

1. AI will probably get it right. But not definitely

AI doesn’t know what it doesn’t know. It’s probabilistic meaning It looks at the available information and selects the answer it considers most likely.

A bank reconciliation cannot be probably correct. It’s either correct or it isn’t.

A payment is either matched to the right invoice or it isn’t. A cost either has the correct VAT treatment or it doesn’t.

If the answer isn’t clear, the transaction needs to be flagged (or left to a human to reconcile).

A system that is right 95% of the time sounds good. But the problems caused by the 5% of transactions that are incorrectly processed can be painful. We know because we experienced them.

When Jax auto-reconcile was launched we trialled it to see how it compared to the tool we’re already using. 

Most of the transactions were reconciled correctly, but some receipts were matched against the wrong sales invoices. 

This meant that some clients started being chased for invoices they’d already paid. Equally problematically, other clients that did owe money weren’t being chased.

We quickly turned Jax off but will revisit it again when it’s developed some more.

2. Insights based on inaccurate data

Imagine that JAX flags an upcoming cash shortage and recommends delaying supplier payments or arranging funding. That could be very helpful if it is right.

But what if a major customer receipt hasn’t been reconciled yet? What if the bank feed has glitched and not pulled through some receipts? What if invoices that will never be collected remain sitting in debtors?

The Insights may be surfaced perfectly, just based on incorrect accounting data.

3. The Human reality

The part of bookkeeping that often takes the most time is chasing and chasing and chasing for missing paperwork.

AI can help with the chasing. But it cannot magic a bundle of receipts out of your pocket and into Xero.

Someone still has to take a photograph and submit it to Dext (or whichever document capture tool you use).

Also, AI doesn’t know your intentions. It doesn’t know what’s going on in your life or what your goals are so when it’s offering advice to you it doesn’t necessarily have the context it needs to tailor that advice to you.

How to safely use AI in your finance function

My message to agency owners is to embrace this technology but don’t rush to remove the controls around it.

Give AI the right work, be aware of its limitations, and put guardrails in place to protect your data (and your business).

Let AI collate, read and process data but accept that there will be errors. Then keep a trusted team of humans on hand to oversee the systems, handle exceptions and use the information to challenge and advise you.

Xero calls this approach ‘Accountable Intelligence’. 

It’s based on the principle of keeping a human in the loop. AI removes manual work without removing human oversight and accountability.

We’re on the same page

We want the same thing you do: to spend less time and money moving data around, and spend more time helping to improve your business.

Over time, the balance should change. Humans should not need to check every transaction forever. The technology should process what can be determined safely and human attention can be directed to the things that genuinely need judgement.

That is the hybrid finance team we want to be building.

You can read more about the announcements in Xero’s Xerocon London 2026 update.