Industries

AI for accountants: less keying, faster payables, faster month-end.

By Merryl DMello, founder of DMello Advisory. Eighteen years building and leading products at Intel, HP, Rackspace, BioNTech and Phoenix Contact.

In a small accounting practice or bookkeeping firm, AI in accounting comes down to three kinds of repeat work it handles reliably today: reading supplier invoices and drafting the posting (accounts payable, in the UK often called the purchase ledger), moving figures from one system into another, and assembling the standard month-end pack. It drafts. A qualified person checks and signs off. Judgement and advice stay with you.

We do not sell software. Software vendors rank their own products, and most pages on this topic are their lists. This page is vendor-neutral. We measure your week first, then build only what pays.

What can AI reliably do in an accounting firm today?

Reliably means three things: the task repeats, the steps can be written down, and a person checks the result. Then AI can read, sort and draft. These are the usual cases in a practice.

Invoices in. A supplier invoice arrives by email or upload. AI reads supplier, date, net amount, tax and total, proposes the coding and puts it in a review queue. You approve before anything posts. More on office admin in general: admin tasks.

Data entry. The same figure is often typed twice: bank lines, client spreadsheets, portal exports. A script can move it once and flag every mismatch for a person.

Month-end reporting. Pull the balances, fill the standard pack from the template, draft the variance notes. The reviewer reads, edits and signs.

Chasing clients. Missing receipts and unanswered questions need a reminder. AI writes the draft in your tone, and you send it. The inbox side is covered on email, calendar and filing.

Which tasks pay back first?

Start with what runs most often and needs the least judgement. Rank tasks by active minutes at the screen, times how often they run in a month. A task of five minutes done eight times a month ranks above a task of 15 minutes done once, even though the single task looks shorter.

How to work out your own number. Active minutes per run, times runs per month, times the number of people who do it. Divide by 60 for hours per month. Use timings from a real week, not estimates.

We do not quote a saving for your firm before it is measured. The Time Audit produces those minutes.

What is accounts payable (purchase ledger) automation, and where does it pay?

Accounts payable is the path of a supplier invoice from arrival to payment. The full walk-through, with what to measure and how to start, is on our page on accounts payable automation. The usual steps are capture, matching, coding, approval and the payment run. Invoice processing automation is the same path seen from the invoice.

What AI does well. It reads the invoice, whatever the layout. It matches supplier and amount against what you expect, proposes the coding and flags a possible duplicate. Each result goes to a review queue.

What stays with a person. Approval, and releasing the payment. An exception, such as an unknown supplier or an amount that does not match, goes to a person too.

Where it pays. It pays when many invoices arrive in many layouts and the same few people key them. It does not pay when you receive a handful of invoices a month from the same suppliers.

Whether to buy packaged payables software or build a small workflow on tools you own is a separate question. Measure the steps first. Then the choice is about your numbers, not a vendor list.

What AI agents can do in an accounting practice

An AI agent, in plain terms, is software that carries out a sequence of steps for you instead of answering one question at a time. In a small accounting practice the useful examples are short and boring. An agent reads a supplier invoice, checks it against the order, drafts the posting and queues it for review. Then it stops.

What suits an agent. Work with written steps, a clear start and end, and a named person who checks the result. Invoice handling is the clearest case. Chasing a client for a missing receipt and filling the standard month-end pack are two more. The path is described in accounts payable automation.

What does not suit an agent. Approving a posting, releasing a payment, advising a client and signing off. These stay with a qualified person. If you cannot write the steps of a task down, an agent has nothing to follow.

We do not promise a result for your practice. We measure the task first. The Time Audit shows which tasks have the steps and the volume for an agent to be worth building.

Which AI tools should a small accounting practice use?

There is no single best tool. The right one depends on which steps eat your time. The tool choice comes after you have measured which task costs the most hours, not before. A firm that loses its week to invoice capture needs something different from a firm that loses it to month-end commentary or to chasing clients for receipts.

So the order is: measure first, decide second, buy or build third. The Time Audit does the first two. It does not push a product, because we do not sell one.

Will AI replace accountants?

Not the people who do the judging. AI drafts, and a qualified person checks and signs off. What it takes over is repeat desk work: keying invoice data, moving figures between systems, assembling a standard pack.

Advice, judgement and responsibility for the numbers stay with you. How much of your week is repeat desk work is a measurement question. Count one real week and you will know for your practice.

When is it not worth it?

It does not pay when volume is low. A firm with few clients or few invoices spends more time on setup than it saves.

It does not pay when each client's books are handled in a different way and the process changes every time. There is no pattern to automate.

It does not pay without a reviewer. Every draft needs a named person who approves it. Without one, the pile of unchecked drafts grows.

And it does not pay while the data question is open. If client data cannot go to a service until the terms are clear, wait until they are.

What does it cost?

The first step costs nothing. In the free 20-minute diagnosis we check whether and where AI pays off in your business. After that come fixed prices, net + VAT:

The Time Audit is credited 100% against a Build Sprint ordered within 30 days. We work with companies in any country, remote-first. Prices are net. VAT or other taxes are applied according to the rules of your country.

What about client confidentiality and data protection?

Client ledgers hold personal and financial data. Before an AI service sees any of it, read the provider's terms. Find out where the data is processed and whether it is used to train models. Get a written data processing agreement. If the answer is not in writing, rule that provider out.

The checks, step by step, are on AI and data privacy. The rules that apply depend on your country and on your professional body. This page is not legal or tax advice.

How do we start?

With the Time Audit. It measures every recurring task in your practice and writes a decision on each row: automate, drop or keep. You see in black and white which task goes first. If your firm already pays for Microsoft 365, part of the build can sit in tools you own, as described on Power Automate and Copilot Studio.

Before it comes the free 20-minute diagnosis. We go through your routines and decide whether the Time Audit is the right tool. If it is not, I say so.

Book the 20-minute diagnosis

Free, 20 minutes, no sales pitch.

What usually comes next

Once invoices and reporting run with less typing, the next question is price. Which fixed-fee client work costs more to deliver than it earns? The Cost Audit prices each recurring delivery on fully-loaded hourly costs and marks the rows that run at a loss. It also gives you a repricing line you can send to the client.

Frequently asked questions

What can AI reliably do in an accounting firm?

Repeat work with a written process and a human check: reading supplier invoices and drafting the coding, moving figures between systems, assembling the month-end pack, and drafting client reminders. Judgement and sign-off stay with a qualified person.

What is accounts payable automation?

It automates the path of a supplier invoice from arrival to payment: capture, matching, coding, approval and payment run. AI handles capture, matching and a proposed coding. A person approves and releases the payment.

Which accounting tasks pay back first?

Those that run often and need little judgement: invoice capture, re-keying between systems and month-end pack assembly. Rank them by active minutes times frequency per month.

Which AI tools are best for accountants?

It depends on which steps eat your time. We do not sell software and do not rank products. Measure your week first, then decide whether to buy a package or build a small workflow.

When is AI not worth it for a bookkeeping firm?

When volume is low, when every client's process differs each time, when nobody reviews the drafts, or when the data protection terms are not yet clear.

Will AI replace accountants?

Not the judging. AI drafts and a qualified person checks and signs off. It takes over repeat desk work such as keying invoices and assembling standard packs. Advice and responsibility for the numbers stay with you.

What are AI agents for accountants?

Software that carries out a sequence of steps, for example: read the invoice, check it against the order, draft the posting, queue it for review. A person approves. See accounts payable automation.

What does it cost?

The 20-minute diagnosis is free. The Time Audit is €1,490 ($1,700) net + VAT, the Cost Audit €1,790 ($2,040) net + VAT and the Build Sprint €4,900 ($5,570) net + VAT. All are fixed prices, delivered within 5 working days. Details are on the Time Audit page.

Is client data safe in an AI tool?

It depends on the provider. Check where the data is processed, whether it trains models, and get a written data processing agreement before any client data goes in. This is not legal advice.

How do we start?

With the Time Audit: every recurring task measured, with a decision per row. The free 20-minute diagnosis comes first and settles whether it fits.

About the author. Merryl DMello, founder of DMello Advisory. Eighteen years building and leading products at Intel, HP, Rackspace, BioNTech and Phoenix Contact.

At HP, Merryl automated seven manual steps and cleared a three-year backlog in five weeks, a 90% reduction in effort. At BioNTech, Merryl built a reporting dashboard for 35+ stakeholders that saved 10 hours a month per stakeholder. This is earlier employment, not DMello Advisory client work.

Every delivery from our development team is reviewed personally by Merryl DMello before it reaches you.

Related tasks

Keep reading

Last updated:

Other services: The Cost Audit The Company Day