Accounting firms are under twofold pressure. On the one hand, production work is becoming standardised and its prices are being driven down, while clients expect faster answers and more advice. On the other, recruitment remains difficult, and experienced staff spend a significant share of their time on data entry, reconciliation and checking tasks that do not draw on their expertise. Artificial intelligence (AI) arrives in this context with a simple promise: to free up time on production so that it can be reinvested in the review of the accounts, client relations and advisory work.
The promise is real, but it runs up against constraints specific to the profession. Professional secrecy applies to all the information entrusted by clients. Responsibility for the engagement rests with the firm and with the chartered accountant, and it cannot be shared with a tool. Due diligence obligations in respect of anti-money laundering require firms to know their clients and to document the diligence they carry out. A firm that adopts AI without taking these rules into account takes risks that its clients would not forgive.
This guide is intended for chartered accountant partners, engagement managers and firm managers who want to integrate AI into their organisation. It describes the uses to examine for each type of engagement, the rules to follow, how to organise the firm and train staff, and the pitfalls to avoid. The examples are generic: they name no firm, client, person or tool.
Why AI matters to accounting firms#
A firm's work rests on three raw materials: documents, data and judgement. Documents arrive in a variety of forms, often incomplete. Data is entered, reconciled, checked and turned into accounts. Judgement comes into play at every stage, to decide how a transaction should be treated, assess a risk or advise a client. Production tools have already automated part of data entry; language models add a new capability, that of reading, summarising and drafting text.
It is this capability that changes the game. A large part of a firm's time goes on text: client emails, requests for documents, review notes, summaries, answers to tax or employment questions, regulatory monitoring. This is where AI can help most, provided that what it produces is properly supervised.
The firm's engagements and their tasks#
A firm carries out several types of engagement, which do not offer the same possibilities. Bookkeeping and accounts production concentrate the repetitive tasks: collecting documents, data entry, bank reconciliations, matching (lettrage). The review of the accounts and the preparation of the annual accounts call more on judgement: consistency checks, analysis of the accounts, substantiation of balances, preparation of the tax return package. Payroll, employment and tax engagements demand constant monitoring and great rigour. Finally, advisory work draws its value from knowledge of the client and its sector.
Before choosing a tool, it is useful to measure, for each type of engagement, where time is actually spent. This measurement often reveals hidden tasks: chasing documents, reworking badly scanned documents, back-and-forth exchanges with a client on the same question. It is sometimes these tasks, more than data entry itself, that offer the clearest gains.
The uses to examine#
The uses below are possibilities to be assessed in the context of each firm. Each one assumes human oversight proportionate to the risk.
Production#
A tool can read the documents received, extract the relevant information, propose a posting, identify missing documents and prepare requests to the client. It can also flag anomalies: a duplicate invoice, an unusual amount, a new supplier. The member of staff validates the proposals and deals with the exceptions. With the gradual roll-out of business-to-business e-invoicing, a growing share of documents will arrive as structured data, which will make this processing easier.
Review#
During the review, a tool can compare balances from one financial year to the next, flag variances that need explaining, prepare working papers and document the checks performed. It can also propose an initial analysis of the accounts based on the year's data. The reviewer retains judgement over what is material and over how the points identified are dealt with.
Client relations#
An assistant can prepare answers to clients' routine questions, draft clear requests for documents, summarise an email exchange before a meeting or prepare meeting minutes. These uses improve the firm's responsiveness, provided that answers on tax, employment or legal matters are checked before being sent.
Regulatory monitoring#
Tax, employment and accounting regulations are constantly changing. A tool can help track newly published texts, prepare a summary of them and identify the clients affected by a change. The summary must always be checked against the source text, because a model can summarise a technical text inaccurately.
Advisory work#
To prepare for an advisory meeting, a tool can gather the figures from the client file, propose an analysis of the situation, prepare simple simulations and list the questions to cover. The advice itself, which commits the firm, remains the work of the chartered accountant.
Internal management of the firm#
AI can also serve the firm itself: tracking time and deadlines, preparing engagement letters from templates, documenting internal procedures, helping to onboard new staff. These less visible uses are often the simplest to put in place because they concern internal data.
Professional secrecy comes first#
The chartered accountant is bound by professional secrecy, under the rules governing the profession and the French Criminal Code. This secrecy covers all information entrusted by clients or learned in the course of the engagement. Entrusting this information to an AI tool means passing it on to a provider: the question is therefore far from theoretical.
Before using a tool on client data, the firm must know where the data is processed, who can access it at the provider, how long it is retained, and whether it is used to train models. It must make sure that the service's contractual terms are compatible with its obligations. Consumer tools used with a personal account, whose terms change and whose data may be reused, do not generally meet these conditions.
The General Data Protection Regulation (GDPR) applies on top of professional secrecy as soon as files contain personal data, which is the case for payroll, directors' personal tax returns or the files of sole traders. The firm, acting as controller or processor depending on the engagement, must put a contractual framework around its provider's involvement.
Responsibility, the engagement letter and professional judgement#
Responsibility for the engagement remains with the firm and the chartered accountant, whatever tool is used. A tool that proposes an incorrect posting, an incomplete summary or an inaccurate tax answer bears no liability of its own; the liability lies with the firm that let the error through. Human review is therefore not an optional precaution: it is a condition of practice.
The engagement letter may usefully mention the use of tools, including AI tools, in carrying out the engagement, together with the associated confidentiality safeguards. This transparency reassures clients and prevents misunderstandings. It is worth discussing this with the firm's professional indemnity insurer, to check that the intended use does not raise any difficulty.
Anti-money laundering: assisting without replacing#
Chartered accountants are subject to due diligence obligations in respect of anti-money laundering and counter-terrorist financing. They must identify their clients, assess the risks, exercise appropriate vigilance throughout the relationship and file suspicious activity reports with the competent financial intelligence unit.
AI can help gather identification information, spot unusual transactions in the accounting data or document the due diligence carried out. It replaces neither the chartered accountant's risk assessment nor the decision to file a suspicious activity report. The firm's framework must remain explainable: in the event of an inspection, it must be possible to show how an alert was handled and by whom.
Organising the firm: usage rules, approved tools, an AI lead#
The use of AI in a firm cannot rest on individual initiative. Without rules, staff use the tools they know, often with personal accounts, on client data. Simple usage rules therefore need to be written down: which tools are approved, for which uses, with which data, and which sign-offs are mandatory before an output goes to a client.
Appointing an AI lead helps to keep these rules alive. This lead, who does not need to be a technical specialist, answers staff questions, records the uses that work, flags those that cause problems and keeps the list of approved tools up to date. In a firm with several partners, it helps for one partner to own the subject and settle the choices.
Training staff#
Staff training works best when it starts from their own tasks. A production accountant does not have the same needs as a reviewer or a partner in charge of advisory work. The most useful exercises are based on typical, anonymised client files and on everyday situations: dealing with an ambiguous document, preparing a request for documents, summarising an exchange, checking a tax answer.
Training must also cover the firm's rules and the limits of the tools. A member of staff must be able to recognise an answer that is plausible but wrong, check a summary against its source and refuse to use an unapproved tool on client data. These reflexes matter more than mastery of any particular feature.
Measuring the gain without overestimating it#
To know whether AI brings anything, you need to measure before deploying: time spent per type of task, response times to clients, the number of corrections at review. These measurements, however simple, make it possible to compare before and after using the same method.
The real gain must take account of checking time. A tool that quickly produces a posting that then has to be checked line by line adds nothing. You also need to ask what becomes of the time freed up: if it is reinvested in review, client relations or advisory work, the firm derives value from it; if it simply dissipates, the gain remains theoretical.
Talking about AI with clients#
A firm's clients hear about AI and ask questions: does the firm use it, is their data protected, will fees go down? It is better to prepare clear answers than to improvise them. Transparency about the tools used and the confidentiality safeguards strengthens trust.
Clients may also expect the firm to advise them on their own use of AI, particularly for their financial management. A firm that has structured its own use is better placed to support its clients, and this support can become an advisory engagement in its own right.
Pitfalls to avoid#
The first pitfall is letting each member of staff choose their own tools, with personal accounts, on client data. The second is sending a client a generated answer without having checked it. The third is automating production without revisiting the review checks that depend on it. The fourth is measuring the gain without counting checking time. Finally, the fifth is forgetting the due diligence obligations in the belief that a tool fulfils them on the firm's behalf.
Glossary#
- Professional secrecy: the obligation not to disclose information entrusted by clients or learned in the course of the engagement.
- Engagement letter: a document that defines the engagement entrusted to the firm, its terms and the responsibilities of each party.
- Review: the set of checks carried out on the accounts to ensure their consistency and substantiate the balances.
- Matching (lettrage): reconciling entries that offset each other, such as an invoice and its payment.
- Due diligence obligations: the diligence required to know clients and detect suspicious transactions in relation to money laundering.
- Controller: the person who determines the purposes and means of the processing of personal data.
- Usage rules: rules that state which tools are approved, with which data, and who validates their outputs.
- AI lead: the person responsible for keeping the usage rules alive, answering questions and monitoring uses within the firm.
Printable checklist#
This list helps to organise the use of AI in a firm. It replaces neither the advice of a lawyer nor the rules of your professional body.
Before choosing a tool#
- Time spent per type of task has been measured, at least roughly.
- Where the data is processed, how long it is retained and how the provider uses it are known.
- The contractual terms are compatible with professional secrecy and the GDPR.
- The professional indemnity insurer has been consulted if the intended use warrants it.
Before deploying#
- Usage rules are written down: approved tools, uses, data, mandatory sign-offs.
- An AI lead has been appointed, and a partner owns the subject.
- Engagement letters mention the use of tools and the associated safeguards, if the firm so decides.
- Staff have been trained on their own tasks and on the firm's rules.
Over time#
- Every generated tax, employment or legal answer is checked before being sent.
- Errors identified are logged and shared with the AI lead.
- The due diligence framework remains explainable: each alert has a record and a named owner.
- The time freed up is tracked, and how it is used is discussed among the partners.
Frequently asked questions#
Will AI replace accounting staff?#
It transforms their tasks more than it replaces them. Data entry and reconciliation shrink, while checking, analysis and client relations take up more room. The firms that train their staff in these new tasks get the most out of it.
Can a consumer tool be used with client data?#
Generally not, when it is used with a personal account whose terms guarantee neither where the data is processed nor that it will not be reused. The firm must check that the service's terms are compatible with professional secrecy and the GDPR.
Should clients be told that AI is being used?#
Transparency is recommended. Mentioning the use of tools and the confidentiality safeguards in the engagement letter, or in general information issued by the firm, prevents misunderstandings and strengthens trust.
Can AI fulfil the firm's due diligence obligations?#
No. It can help gather information and spot unusual transactions, but the risk assessment and the decision to file a suspicious activity report remain with the chartered accountant.
Where should a small firm start?#
With simple usage rules, an approved tool whose terms have been checked, and a limited first use, such as preparing requests for documents or summarising exchanges before a meeting. The rest is built on what has worked.
Conclusion#
AI can help an accounting firm shift its time from production to review, client relations and advisory work. It only does so lastingly if the firm stays in control: tools chosen for their confidentiality safeguards, written usage rules, staff trained to check, and a responsibility that clearly remains with the chartered accountant. The safest approach is to start with a limited use, measure what it changes, and then extend what has proved its worth.