During year-end, a good part of an accounting assistant's day requires no expertise at all: asking for a missing bank statement for the third time, renaming PDFs, answering "did you get my invoice?". That is exactly the kind of work an AI agent handles well, as long as it is properly scoped.
If the topic is new to you, start with my article on AI agent use cases for business, which explains what an agent is. Here I focus on accounting firms, which carry a constraint most businesses don't face to the same degree: professional secrecy.
In short. Six realistic uses: collecting documents, sorting email, pre-filling invoice entry, drafting client replies, preparing engagement letters, summarizing meetings. In all six, AI prepares and a person approves. First steps: a business account, data reuse switched off, a data processing agreement and as little personal data as possible sent to the tool.
What does an AI agent change in an accounting firm?
An AI agent is a program that reads a request, fetches information from tools you define in advance and carries out a limited action: drafting a reply, filing a document, creating a task. Unlike a plain chat assistant, it is connected to your tools (email, document portal, accounting software) and acts within limits you set.
In a firm, that means less time on file logistics and as much time as before on accounting judgment and advice. AI replaces neither the staff accountant nor the partner who signs.
What are 6 practical uses of AI in an accounting firm?
1. Collect supporting documents and chase clients
The agent compares the list of documents expected for each file (bank statements, purchase invoices, expense reports) with what has been uploaded. It drafts a personalized reminder listing exactly what is missing, sent from the staff member's address, then updates the list when the document arrives.
FromYour accountant
Subject2026 year-end: 3 documents missing
Draft to approve
Hi Ms. Ross, to move forward with your year-end, I still need: the September bank statement, the invoice for the vehicle bought in June and the expense report for the March trade show. You can upload them to your usual portal. Thank you!
2. Sort the firm's shared inbox
The agent reads each incoming email, links it to the right client file, labels it (attachment, question, urgent, tax authority) and flags it to the right person. It doesn't reply on its own: it files and alerts.
3. Pre-fill invoice data entry
For a purchase invoice, the agent extracts the supplier, date and amounts and suggests an account based on the client's history. Many accounting platforms already do part of this: the agent is most useful on the cases they handle badly (poor scans, new suppliers) and for flagging doubts rather than resolving them. For pulling receipts into an accounting tool automatically, see automating bookkeeping with n8n, Pennylane and Qonto.
4. Draft replies to routine questions
"Where is my tax return?", "What's the deadline for…?", "Can you send me my financial statements?". The agent drafts an answer from the file and the firm's procedures; the staff member reviews, edits if needed and sends it. Substantive tax or legal questions are flagged and left to a person.
5. Prepare engagement letters and new client onboarding
From the intake form and the discovery call, the agent pre-fills the engagement letter on the firm's template, prepares the list of documents to request and creates the setup tasks. The partner checks the scope and signs. For following up on unsigned engagement letters, it works like automated quote follow-up.
6. Summarize a meeting and prepare the next one
After a review meeting, the agent turns the staff member's notes into a structured summary (decisions, documents to provide, next deadlines) and sets reminders. If you record meetings, tell the client and ask for their consent.
How risky is each use?
| Use | Data involved | Risk | Who approves |
|---|---|---|---|
| Document collection | List of documents, client name and email | Low | The staff member at first, then spot checks |
| Inbox sorting | Message content and attachments | Medium | Nobody is contacted without a person |
| Data entry pre-fill | Invoices, amounts, suppliers | Medium | The staff member, on every entry |
| Draft replies | Client file, correspondence | Medium to high | The staff member, before each send |
| Engagement letters | Client identity, scope of work | Medium | The partner, who signs |
| Meeting summaries | Notes, sometimes the owner's personal data | High if recorded | The staff member, before sharing |
Which confidentiality rules apply when accountants use AI?
The rules depend on your country and your professional body. I work mostly with French businesses, so here is the French and EU picture; if you practice elsewhere, check your own regulator's guidance, which often follows the same logic.
In France, accountants are bound by professional secrecy, and article 226-13 of the French Criminal Code punishes disclosing secret information you hold through your profession with one year of imprisonment and a 15,000 euro fine. Sending a client file to a badly configured AI tool raises exactly that question.
The GDPR applies too, since client files contain personal data. In its Q&A on using generative AI systems (July 2024), the French data protection authority (CNIL) gives guidance that fits a firm well:
- Prefer a solution you control. When personal data is sent to the tool, the CNIL generally considers an on-premise deployment safer; the cloud remains possible with strict contractual safeguards.
- A data processing agreement with the vendor that sets out responsibilities and who can access the data.
- No personal accounts. Dedicated business addresses, never a staff member's free subscription.
- Data reuse switched off where possible, so your files don't train the vendor's model.
- As little personal data as possible. The agent doesn't need the owner's social security number to ask for a bank statement.
- Safeguards for transfers outside the EU if the vendor processes data elsewhere. Self-hosting your automations in Europe is one option, covered in n8n and GDPR.
- Check every output. The CNIL advises never reusing the system's answers as they are.
- An internal policy listing allowed and forbidden uses, plus staff training.
Finally, if an agent replies directly to clients (in a portal or by email), the EU AI Act requires telling people they are interacting with AI, unless it's obvious. The text published in 2024 says this obligation applies from August 2, 2026.
Where should an accounting firm start with AI?
- List the repetitive tasks of a typical week and the time they take. The tasks to automate checklist is a good starting point.
- Pick one low-risk task. Document collection is often the best candidate: little sensitive data, a visible gain.
- Take stock of the tools you already have (accounting platform, document portal, email) and what they can already do.
- Test on dummy data, then on a few willing clients, with a person approving every output.
- Measure the time spent before and after for a month, then decide whether to extend it.
If you want help scoping and building that first agent, it's the core of my custom AI agents for business work: we start with a free 30-minute call, then a free mockup built on your case, before any quote.
What should an accounting firm not hand over to AI?
- Tax or legal advice. A language model can be confidently wrong, quote a repealed rule or invent a threshold. Substantive answers are written or reviewed by a professional.
- Filings and submissions. Nothing goes to a tax authority without human approval.
- Unchecked entries. A suggested account stays a suggestion.
- Sensitive files (disputes, estates, a business in difficulty): keep them out of the agent's scope.
- A consumer tool on a personal account. It's the setup the CNIL advises against, and the most common one in practice.
Sources checked on October 17, 2026
- Legifrance, French Criminal Code, article 226-13 (breach of professional secrecy, in French)
- CNIL, Q&A on using a generative AI system (July 18, 2024, in French)
- CNIL, AI and GDPR compliance (in French)
- EUR-Lex, Regulation (EU) 2024/1689 on artificial intelligence, articles 50 and 113 (published text; check the consolidated version)
Frequently asked questions
Can an accountant use ChatGPT with client files?
Not on a personal account and not without precautions. You need a business plan with a data processing agreement, data reuse switched off, as little personal data as possible and a person checking every answer, as the French data protection authority recommends.
Will AI replace accounting staff?
It mainly replaces tasks: reminders, sorting, renaming, first drafts. Accounting judgment, the client relationship and responsibility for the signature stay with people. A well-scoped agent frees up time for advisory work.
Which task should an accounting firm automate first?
Collecting supporting documents and chasing clients for them: the task is repetitive, involves little sensitive data and the gain shows up in the very first closing period.
Does professional secrecy rule out AI in an accounting firm?
No, but it means you must know where the data goes and who can access it. A tool configured not to reuse data, covered by a contract, with limited access and human approval, lets you work within that framework.
Do clients need to be told that an AI agent is replying?
Yes, in the EU. The AI Act requires telling people they are interacting with AI unless it's obvious, and the GDPR requires informing them about how their data is processed. A clear notice at the start of the exchange is enough in most cases.