Which accounting firm tasks should be automated first?

Chasing missing documents: high volume, clear rule, low risk. Next come reading and filing documents, then suggesting ledger coding, always approved by a person.

Can AI post entries on its own?

That is not the setting we recommend. The AI suggests an entry, a team member checks and approves it. No line is posted without a person’s agreement.

How much time can a firm save?

It depends on your clients and tools. Published figures mostly come from vendors and are self-reported, for example up to 5 hours a week according to 61% of accountants surveyed by Qonto and Ifop. Measure one week before and after.

Do we need to change accounting software to automate?

No. Automation plugs into the mailbox, accounting software and shared folders you already use. Changing tools adds risk without being necessary.

Accounting firms: the tasks that really automate

Which accounting firm tasks can really be automated?

Short answer: in an accounting firm, what automates well is high-volume work with a stable rule: chasing missing documents, reading and filing supporting documents, suggesting a ledger account, matching the obvious bank lines. What does not automate is judgment: settling an ambiguous case, deciding an entry, advising the client. The AI prepares, a person on your team approves.

A fiduciary’s staff time is expert time. Yet a large share of the week goes to gestures that need no expertise: chasing a receipt, opening a document sent over WhatsApp, retyping an amount. This article sorts those tasks on two criteria, how much time they take and how reliably an AI handles them, and says where to stop.

Where the time goes: the five heaviest tasks

Most published figures on this topic come from software vendors, meaning people who sell the solution. Read them as orders of magnitude, not measurements. For instance, the Qonto x Ifop 2026 barometer, run among more than 300 French accountants, reports that 61% of them expect to save up to 5 hours a week with AI, and 72% name time saved on repetitive tasks as the first concrete benefit. These are self-reported estimates, not a stopwatch.

The same pattern keeps coming back when we talk to firms. Here are the five items that consume the most, in the order we tackle them.

Collecting documents. Receipts are missing, you chase, then you chase the chase. It is the most visible task at year-end close.
Reading and filing documents. One invoice arrives by e-mail, another as a photo, a third as a scanned letter. Each one is opened, read, named and filed.
Data entry and coding. Copying the supplier, the amount, the VAT rate, picking the account.
Bank reconciliation. Matching each statement line to a document.
Routine answers. “Where is my file?”, “Did you get my invoice?” Each question interrupts someone mid-task.

The sort: what automates, what gets prepared, what stays human

These tasks are not equal. The table below ranks them by realistic automation reliability and by the role that stays with your team.

Task	What the AI does	Typical reliability	Your team’s role
Chasing missing documents	Finds transactions without a receipt, drafts and sends the reminder	High: simple rule, clear history	Approves the list of clients contacted
Filing documents	Reads amount, date, supplier, VAT, files it in the right folder	High on standard documents, medium on unusual ones	Settles ambiguous cases
Coding suggestions	Suggests account and rate from the client’s history	Medium: depends on a consistent chart of accounts	Checks every suggestion before posting
Bank reconciliation	Matches the obvious lines (amount, date, label)	High on obvious cases, low on grouped payments	Handles the remaining lines
Routine answers	Drafts a reply from the file status	Medium	Sends or edits
Advice, judgment, edge cases	Nothing, or context preparation at most	Not applicable	Fully human

To size the imperfection, one French-language vendor states that about 15% of documents, ambiguous or unusual, still need a manual check, and about 10% of reconciliations are not automatic. It gives neither sample nor method, so keep the logic rather than the percentages. A serious automation does not aim for 100%. It aims for a sort: the simple passes alone, the doubt goes up to a person.

First task to automate: chasing missing documents

If you can pick only one, pick document reminders. It combines three rare qualities: high volume, a clear rule (a transaction without a document) and low risk (an awkward message can be fixed, a wrong entry costs more).

The principle is simple. The tool compares the period’s transactions with the documents received, finds those with no receipt, then drafts a reminder suited to the client: shorter for a responsive client, more detailed for one who forgets. Your team sees the list before anything is sent. On the studio’s page for fiduciaries, an illustrative example, not a client measurement, contrasts about 3 hours a week per employee with 20 minutes of review. We do not present that as a guaranteed result: the time actually saved depends on your clients and your tools, and is measured before and after.

Second and third tasks: reading documents, suggesting coding

Document reading is improving fast but stays uneven. A standard-format invoice is read correctly. A crumpled receipt, a handwritten note or a bilingual invoice less so. The useful approach is to let through on their own the documents whose extraction is consistent (the total equals the sum of the lines, the date is plausible, the supplier is known), and to send the rest to a person.

Coding calls for more caution. A supplier’s name does not always qualify a service, and the VAT rate can depend on the final use of the item or service. A generative model can also give a wrong justification with confidence. The right setting is a suggestion, never a posted entry: no line is booked without a team member’s approval. The tool learns from your corrections, which gradually reduces the number of cases to redo.

What does not automate, and why that is good news

Three families of tasks stay human, and it is better to say so up front.

Judgment on edge cases. An unusual transaction, a client whose situation changes, a document that does not fit: the AI can prepare the context, but the decision belongs to your team.
Advice. It is what your clients pay for. Saving time on data entry only makes sense if that time goes to advice, not to a heavier caseload.
Responsibility for what is signed. Anything that binds the firm is reviewed and approved by a person. The automation described here organizes the work, it does not replace that review.

This article covers work organization only. It says nothing about the obligations of your profession, which are a matter for your professional bodies and advisers.

A simple calculation to see if it is worth it

Take a deliberately hypothetical example, to be replaced with your own numbers. One employee spends 3 hours a week chasing receipts, at an internal cost of CHF 80 an hour. That is CHF 240 a week, or about CHF 11,000 over 46 working weeks. If automation brings that down to 20 minutes of review, the difference runs to thousands of francs per employee per year, before even counting team morale.

The calculation has two limits. The hourly rate depends on your firm, and the time actually saved will never match the example. That is why we recommend measuring one real week before deciding, rather than trusting a brochure figure, ours included. If the yearly gap is smaller than the cost of the project, do not start the project.

When automating is not the right choice

Automation does not suit every firm, or every moment.

Your documents already arrive clean. If clients use a portal that already files everything, automatic reading adds little.
Your chart of accounts is inconsistent. Automation plugged into poorly kept references amplifies errors instead of reducing them. Clean up first.
Volume is too low. For a few dozen documents a month, a good manual routine costs less than a project.
You are in the middle of a year-end close. Do not change tools at peak time. Start afterwards, with the calm needed to test.

How to start without taking risks

Measure for one week. Note the time spent on collection, filing and entry, per employee. Without a baseline, no gain can be shown.
Pick a single process. Chasing missing documents is the safest candidate.
Plug it into your current tools. Mailbox, accounting software, shared folders. Changing tools multiplies the risks.
Test on a limited scope. A few clients, with every message approved.
Measure again, then extend. Compare with the starting week. Roll out only if the gain is real.

This is the approach we use at the studio, on a fixed fee agreed before we start. You can see the processes the studio automates and the plans and prices on the site. If AI is not the right lever for you, we say so after the diagnostic.