AI Can Do Your Bookkeeping. But Who Knows Whether It Is Right?

Software has made bookkeeping easier. It hasn’t made accounting knowledge automatic — and the first months of MTD for Income Tax are showing us the difference.

• Software vs. knowledge
• “The software accepted it”
• Capability without responsibility
• The £800 accountant
• MTD as a live experiment
• Will AI take the jobs?
• Where juniors learn judgement
• If you do your own books

There is a question I think accountants are going to hear increasingly often:

“Why should I pay an accountant when software and AI can do it for me?”

It is a fair question. Accounting software is becoming easier to use. Bank feeds import transactions automatically. Software suggests categories. Invoices can be scanned rather than entered manually. AI can analyse data, produce reports and explain financial information within seconds. And now Making Tax Digital for Income Tax is encouraging hundreds of thousands of sole traders and landlords to maintain their financial records digitally throughout the year. So perhaps accountants and bookkeepers should be worried. Except that after seeing what has happened during the first few months of MTD, I am beginning to think the opposite may be true.

Software has made bookkeeping easier. It hasn’t made accounting knowledge automatic.

Since MTD for Income Tax began in April 2026, my team and I have had an interesting opportunity to see records prepared by business owners who have decided to do more of their own bookkeeping. Some have done it extremely well. Others have demonstrated something that I think will become increasingly important as financial technology develops:

We have seen income classified incorrectly. Personal expenditure treated as business expenditure. Business transactions omitted. Payments recorded without understanding what they actually represent. And transactions confidently allocated to categories simply because the software suggested them.

The screen looks tidy.

The bank reconciles.

There may even be a green tick beside everything.

But that does not necessarily mean the accounting treatment is correct.

“The software accepted it” is not the same as “the accountant agrees with it”

This distinction is going to matter enormously. Software is very good at processing information. AI is becoming extraordinarily good at interpreting information. But neither necessarily understands the complete circumstances behind a transaction unless somebody provides that context. Consider something as simple as £2,000 arriving in a business bank account.

The bank feed knows that £2,000 arrived. Increasingly sophisticated AI may make a very intelligent guess about what it represents. But the accounting and tax treatment depends on why the money arrived.

That is not merely data processing.

It is judgement.

AI has capability. It doesn’t have responsibility.

I recently read an excellent article by FCCA Zain ul Abideen called Capability Without Consequence. His argument is that finance professionals may be asking the wrong question when they ask: “Can we trust AI?” Instead, he suggests thinking of AI almost as a member of the finance team: extraordinarily capable, but with no accountability for what ultimately leaves the office. His point particularly resonated with me. AI cannot attend an HMRC enquiry and explain why a transaction was treated in a particular way. It cannot take professional responsibility for advice. It cannot know everything that the business owner forgot to tell it.

That is why professional judgement has not suddenly become redundant simply because the technology has become more capable. In fact, it may become more valuable.

The £800 accountant versus the spreadsheet

This brings me back to an argument I have seen recently online. Imagine a business owner pays an accountant £800 for their annual accounts. They receive what appears to be a fairly ordinary set of accounts containing numbers that could seemingly have been produced from Xero, a spreadsheet or perhaps, increasingly, by AI. It is understandable for the owner to think:

“What exactly did I pay £800 for?”

If all the accountant did was copy numbers into a set of accounts, I would agree with them. Technology should make that work cheaper and faster. But that isn’t what a good accountant is being paid for. You are not paying £800 simply because somebody knows how to turn your bookkeeping into a PDF.

You are paying someone to ask:

• Does the income look complete?
• Have the expenses been treated correctly?
• Are any of these costs private?
• Should this transaction be capital rather than revenue?
• Does the VAT treatment make sense?
• Has something been posted twice?
• Is something missing?
• Does this figure agree with what we know about the business?
• Is there a tax adjustment that the bookkeeping software cannot know about?
• Does the overall story told by these accounts make sense?

A spreadsheet can calculate.

Accounting software can organise.

AI can increasingly analyse.

But somebody still needs to know when the answer does not make sense.

MTD may demonstrate this much faster than AI does

Making Tax Digital provides an interesting real-world experiment. For the 2026/27 tax year, qualifying businesses are required to maintain digital records and provide quarterly summaries of income and expenditure through compatible software. HMRC describes those quarterly updates specifically as summaries rather than tax returns, and no accounting or tax adjustments need to be made before the quarterly update is submitted. That distinction is important. The simplicity of submitting a quarterly update could easily create a false sense that everything underneath it has been checked. It hasn’t necessarily. HMRC itself says taxpayers should correct their digital records as soon as they become aware of an error, and that the records ultimately need to be finalised before the tax return is submitted.

And with more than 864,000 sole traders and landlords initially within the first MTD Income Tax population, we are now conducting this experiment on a very large scale.

So will AI and automation take bookkeeping jobs?

Probably some of them. I don’t think the accounting profession should pretend otherwise. Work that once involved manually entering hundreds of transactions is increasingly being automated. And that is a good thing. Previous generations experienced exactly the same process in manufacturing, banking, administration and countless other industries. Technology removed certain jobs. But it also created work that nobody had previously imagined. I suspect accounting will be no different.

We may need fewer people entering transactions.

We may need more people reviewing exceptions.

We may spend less time producing numbers and more time interpreting them.

We may spend less time bookkeeping and more time correcting bookkeeping.

And accountants may increasingly be asked not: “Can you prepare these accounts?” but:

“Can you tell me whether the accounts my software produced are actually right?”

There is another consequence accountants should be thinking about

There is a more uncomfortable side to all of this. If junior accountants stop doing the routine work because AI does it for them, where will they learn the judgement required to check AI later in their careers? This is one of the strongest points in Zain ul Abideen’s argument. Experienced accountants know something looks wrong because they have prepared accounts manually, investigated strange balances, made mistakes, corrected them and seen hundreds or thousands of businesses over their careers. AI may give a newly qualified accountant access to extraordinarily sophisticated capability. But access to capability is not the same as having experience.

That means accountancy firms have another challenge ahead of us. We need to embrace AI while still teaching people accounting. At Business Help UK Group, that is increasingly how I see our own future. I don’t want our accountants spending their careers doing work that technology can perform more efficiently. I want technology doing the repetitive work so our people have more time to investigate, question, understand businesses and advise clients. But I also don’t want accountants blindly accepting something because an AI system or accounting platform says it is correct.

And what about business owners doing their own bookkeeping?

I certainly don’t believe every sole trader needs to employ a bookkeeper. For many straightforward businesses, modern accounting software makes maintaining your own records perfectly realistic. And AI will make that easier still. But there is an important distinction between being able to operate the software and understanding what the transactions mean for your accounts and tax. You may be perfectly capable of doing both. The danger comes when software makes the first one so easy that we assume we have automatically acquired the second. That may turn out to be one of the biggest unintended consequences of the AI accounting revolution.