AI bookkeeping robot working on invoices, expenses and financial reports.

Can AI Do Your Bookkeeping? Where It Helps, and Where You Still Need a Human

AI can already play an impressive supporting role on the manual side of bookkeeping. It can read a receipt, match a bank transaction to an invoice or bill, and follow up a late payer without anyone lifting a finger. What it cannot do is understand your business well enough to make a judgement call, catch a mistake that looks entirely plausible to the untrained eye, or take responsibility if assumptions turn out to be wrong.

Carolyn says:

“Relying on software to chase up a late payer is a risky choice. Getting it wrong can cause irreparable damage to the client relationship. If your business is built on the principle of “Stack them high, sell them cheap” then you probably won’t mind. But if you care about that relationship, I would never recommend delegating something that important to such a blunt tool.”

So, the honest answer is this: AI is a genuinely useful part of bookkeeping, but it is not a replacement for a trained bookkeeper. Used well, it speeds things up and cuts out a lot of tedious admin. Used instead of a person who understands your accounts, it just moves the risk further down the line, where it costs more to sort out.

What’s Already Doing the Work in Your Software

It helps to know what is actually happening behind the scenes, because not everything sold as “AI” works the same way, and the differences affect how much you should trust the output without checking it.

Optical character recognition, or OCR, reads a document and lifts out the text. Apron’s invoice capture works this way, whether a bill arrives by upload, email, app or WhatsApp. It reads what is printed on the document and pulls out the supplier, the date and the amount. This is generally reliable for extracting figures, although it can struggle with dates in US format, for example. It is getting better all the time because human beings are correcting its mistakes, and it is learning. But on its own it has no understanding of what those figures mean for your accounts.

Rules based automation applies logic that a person set up in advance: if the payment is to a particular supplier, code it to a particular category. Apron’s automated reconciliation works this way, as do the bank rules built into Xero and FreeAgent. They are fast and consistent, but only as good as the rule behind them, and they will repeat the same miscoding every month without ever flagging that something is wrong. Also, bear in mind suppliers like Amazon that you might buy a wide range of different items from. Or closer to home, our landlord invoices us each month but the bill includes our rent, electricity, water and insurance. Line item capture does exist, but it still has a way to go before we will use it universally.

Personal experience:

Within our own accounts, we have quite a few annual bills that are paid in equal instalments across the year; Xero tends to post the payments straight to the cost account rather than setting it against the outstanding bill effectively double counting the expense. This incorrectly reduces profit and what an owner can draw out as a dividend. It also means you will underestimate your tax liability.

Machine learning goes a step further. Rather than following only an explicit rule, the software looks at how similar documents and transactions have been handled before and predicts the most likely answer, getting more accurate the more data it sees. Apron’s William AI and Xero’s auto reconcile work this way. It is trained on a firm’s own guidelines to start with, then learns from historical decisions over time. Bookkeepers will often spend months “training” the software but frequently correct the same errors every month.

What Can’t the Software Handle?

There is also a category of decisions that software simply cannot make, because they depend on context it does not have.

Deciding What Is Actually Allowable

Software can code a transaction, but it cannot reliably tell you whether a particular cost was a genuine business expense or a personal one, or whether a purchase should be capitalised rather than expensed.

Spotting a Mistake That Looks Entirely Normal

A sole trader who has taken money out of the business and labelled it “salary” will sail straight through most software checks, because the transaction itself looks unremarkable. It takes someone who understands that sole traders have drawings, not a salary, to catch it before it distorts the year-end figures. Sole traders can employ people too, so they could legitimately have salary costs. How is the software meant to know the difference?

Applying Rules to Your Specific Situation

VAT alone has five different rates, plus exemptions, place of supply rules and special treatment for international trade. Software can tell you what a rule says in general terms. It cannot tell you how that rule applies to your particular transaction with your particular supplier.

Here is a real life example from one of our clients who operates within the Construction Industry Scheme where context is essential.

Scenario

Construction Industry Scheme (“CIS”) status

Decommissioning an air conditioning unit only

Normally outside the scope of CIS

Decommissioning a unit and installing a new one

Normally inside the scope of CIS

Decommissioning a unit and replacing it with an existing, already owned unit

Can fall inside or outside CIS depending on how much work is required to move the replacement unit

Working Out the Tax Consequences

A director’s loan, a benefit in kind, a repair that is actually a capital improvement. These carry real tax consequences and need someone who can join the dots between the bookkeeping and your wider tax position.

Taking Responsibility for the Answer

If HMRC opens a compliance check, “the software categorised it that way” is not a defence. Someone still needs to stand behind the numbers.

Anything repetitive and rules based is a strong fit for AI, provided someone still checks it regularly. It’s like having a trainee that needs supervision and mentoring. Anything that depends on context, intent or the specific facts of your business needs a person from the start.

The Right Way to Bring AI Into Your Bookkeeping

AI is a good tool in the hands of a trained user. It is a recipe for disaster in the hands of someone who does not understand whether it has got something right or wrong.

The businesses getting the most out of this are not the ones handing bookkeeping over to an app and hoping for the best. They are the ones using AI powered software with a bookkeeper who understands both the tool and their accounts, so the output gets checked rather than just accepted.

If you are not sure whether your current setup strikes the right balance, our Good Bookkeeping blog sets out what reliable bookkeeping actually requires, AI assisted or not.

For more on choosing between the two main platforms, our Xero and FreeAgent comparison covers what to weigh up.

Absolutely not. It can take over the repetitive parts of the job, but the judgement calls, the tax implications and the accountability still need to sit with a person.

Not on its own. What matters for MTD is that your records are kept digitally and submitted through compatible software. AI features within that software do not change the compliance requirement either way. Accurate mapping is essential for MTD submissions which requiressomeone who understands double entry bookkeeping, and how HMRC group expense categories.

No, absolutely not. Without a human checking it, the sequence can carry on chasing a customer who has already paid or who has a genuine dispute, which causes more friction than the late payment itself. There is also a time lag between the transaction in the bank and the feed into your software meaning that you could be chasing a customer for a payment they’ve already made.

Our experience is that none of the current software applications are good enough yet to handle this hands-free. At best they work as a starting point. The output is only as reliable as the records behind it, and most forecasting tools still need a fair amount of human interactionbefore the numbers are fit to act on. This is an area where we expect to see massive improvements in the near future as it’s something that most business owners would value and benefit from.

Ask what gets checked, and by whom. If nobody is reviewing what the software has coded or predicted before it feeds into your accounts, you should find someone who does.

Carolyn’s soapbox:

We see loads of VAT returns for new clients where the bookkeeper or accountant has literally just pressed a button. They clearly haven’t done any worthwhile work, or they wouldn’t have filed the return with so many obvious mistakes.

If your bookkeeper or accountant is not actively checking the numbers, then you’d be better off saving your money and doing it yourself. You (and not them) are the one who will be in trouble with HMRC for any mistakes regardless of who filed it.

If you’d like a copy of the detailed proven processes we apply at Composure for bookkeeping and VAT returns, please use our contact form and we’ll be only too happy to send you a copy https://composureaccounting.co.uk/contact/

Our Ethos

We believe in employing local people. We do not outsource or offshore outside the UK.

We believe in training the next generation by teaching them the fundamentals, not just how to enter data into a software application.

We believe in the value of having a relationship with our clients and understanding them and their businesses. They are not a faceless source of revenue.

We want to do our work in the most cost effective way possible so we can deliver excellent service to our clients in the most affordable way we can.

AI is only one of the tools we use. It is not the only tool we use.

If you would like us to look at how AI is being used in your own bookkeeping, and where a human still needs to be checking it, get in touch and we will talk it through.

Book a Discovery Call Subscribe to our Monthly Newsletter

We will only use your details to send you this series and occasional updates from Composure Accounting. You can unsubscribe at any time. For more information see our Privacy Policy.

IMPORTANT NOTICE — NOT TAX OR PROFESSIONAL ADVICE

This article is published by Composure Accounting & Taxation Limited for general informational purposes only. It does not constitute tax advice, legal advice, financial advice, or any other form of professional advice, and should not be relied upon as such. The information contained in this article reflects the law and government guidance as understood at the date of publication and is subject to change.

Every business’s position is different and depends on its specific circumstances. Nothing in this article should be treated as a recommendation to take or refrain from taking any particular course of action. You should always seek independent professional advice, including specialist employment law advice where appropriate, tailored to your own situation before making any decisions.

Composure Accounting & Taxation Limited accepts no liability for any loss or damage arising from reliance on the contents of this article. The publication of this article does not create a client relationship between the reader and Composure Accounting & Taxation Limited.

Composure Accounting & Taxation Limited is registered in England and Wales. Registered office: Wildens, Coneyhurst Road, Billingshurst, West Sussex, RH14 9DE