Accounting and tax firms: what AI can take on today

In an accounting firm, a large part of the day is not spent on accounting. It goes on collecting documents, sorting them, chasing clients for what is missing and rewriting letters that all look the same. With every new tax rule and reporting obligation, there is more of that work: more filings per client, more documents, more deadlines. This is the work AI can take on today, while the expertise stays where it belongs: with you and your accountants.

I look at it as a business owner, not as an accountant. I founded Oliverlist in 2016. Since 2020, our chain has been automated from importing contacts to issuing invoices, and AI has been part of our product since May 2023. Today it sorts prospect replies, writes call scripts and coaches the team. The lesson carries over to a firm: AI is useful on work that arrives in bulk and has to be sorted, completed and written up. It is far less useful when someone has to exercise judgment.

Here are four areas where it can help. They are possibilities, not client references: every firm has its own software, its own clients and its own habits.

Sorting client documents as they arrive

Clients send everything, any way they like: a PDF invoice by email, a photo of a receipt on WhatsApp, statements from two or three banks, a tenancy contract, documents in several languages, amounts in several currencies.

AI can read each document as it comes in and do the first sort:

• recognise the type: purchase or sales invoice, bank statement, expense claim, contract, letter from the tax authority;
• extract the fields you need: date, supplier, TRN, invoice number, amounts before and after VAT;
• file it under the right client and the right period;
• flag duplicates, unreadable scans and anything that matches nothing.

This is not accounting yet. It is the work that comes before it, the work that keeps your team busy before they even open the accounting software.

It is the same kind of problem as prospect replies at Oliverlist: a constant, mixed flow that has to be sorted quickly. In both cases one rule matters: give the AI the right to say "I don't know". A doubtful document should land in a "to check" pile, not in the wrong account. The accounting treatment stays with your accountant. AI suggests, a person approves.

Chasing missing documents

Working out what is missing often takes longer than asking for it. Yet the answer is already in the file: each bank transaction without a supporting document points to a missing document.

AI can match bank transactions against the documents received, list what is missing for each client (date, amount, bank description) and draft the reminder. Not one template sent to everyone, but a clear message telling each client exactly what to send and how, in the client's language. If it writes in a language your team does not read, have someone who does check it before it goes out.

Set the rhythm once: a first request, a reminder, then an alert to the accountant if nothing arrives. At first, every reminder is read before it is sent. Once tone and content are stable, simple reminders can go out on their own. Your team keeps the sensitive cases: the client who is always late, the one going through a difficult period, the one you should simply call.

First drafts of summaries and letters

Many documents start from a blank page even though they always follow the same outline: the year-end summary for the business owner, the note explaining why profit fell, the answer to a question that comes back every year, the letter to the tax authority, the notes from a client meeting.

AI can write the first draft from the figures in the file and the accountant's notes. The accountant reviews, corrects, adds the analysis and signs. You do not save time on thinking. You save it on drafting, by starting from a text to correct rather than from nothing.

At Oliverlist, this is how AI writes our team's call scripts: from what we already know, so nobody starts from scratch.

One rule prevents a lot of trouble: every figure in the text must come from the file, never from the AI. If an amount is missing from the data, the draft leaves a blank and flags it.

Getting ready for filing deadlines

Your calendar is known months ahead: VAT returns, corporate tax filings, licence renewals if you handle them, payroll if you run it for clients. What is less visible is how ready each client file is one week before the date.

AI can keep a list of upcoming deadlines for every client, check what is still missing for each one and send your team a short list every week: files running late, documents still expected, clients to chase first. It can also draft the message to the client: "Your return is due soon, here is what we still need from you."

Your team keeps what requires judgment: elections and options, payment plans, the conversation with a client who cannot pay on time.

Key takeaways

• AI takes on the flows: documents to sort, reminders to write, first drafts, deadline tracking. Judgment and signature stay with the accountant.
• Start with one flow, the one that weighs on your team every week, and measure how long it takes before you change anything.
• Give the AI an "I don't know" option: a doubtful document goes to a person, not to an account.
• At first, everything that goes to a client is reviewed. Reduce the review once errors become rare, not before.
• None of this requires a new accounting system: the work happens upstream of it and around it.