When AI automates accounting, what is an accounting firm still needed for?

Artificial intelligence is already automating part of the accounting work. It does not make the accounting firm useless: it shifts its value towards control, interpretation, advice, and responsibility.

This article is a bit special. It does not deal with a rate or a tax deadline. It raises a more uncomfortable question, including for us: when tools can read invoices, analyze bank statements, propose entries, and detect anomalies, what is the purpose of an accounting firm?

Two simplistic answers are circulating. The first claims that "ChatGPT will do all the accounting for free." The second asserts that "nothing will ever replace humans." Both avoid the real issue: artificial intelligence is a technological disruption that transforms tasks, skills, and the business model of accounting firms.

A technological revolution that transforms the profession

The International Labour Organization estimates that about one in four workers is in a profession with some degree of exposure to generative artificial intelligence. However, it believes that for most professions, the transformation of work is more likely than the complete disappearance of jobs.

The distinction is essential. Accounting involves collecting documents, reading them, posting, bank reconciliation, control, searching for missing items, closing, explaining results, and sometimes defending a position. These tasks are neither equally complex nor equally automatable.

It would therefore be naive to conclude that everything will remain as it was. Manual entry will decrease, some routine positions will be reduced or redesigned, and simple situations will be largely handled in self-service. However, the outlook remains positive: automation can free up time, speed up processes, and make data available sooner. But installing a license without clarifying the process sometimes amounts to automating chaos — quickly, indeed.

AI does not, on its own, maintain accounting.

ChatGPT can analyze structured files, summarize data, calculate, and produce tables or graphs. However, OpenAI recommends checking the code, results, and assumptions before relying on them, and notes that extracting exact values from scanned tables or complex documents may still be imperfect.

Automated accounting relies, in practice, on a broader chain: document recognition, accounting software or ERP, bank connection, allocation rules, reconciliation templates, approval circuits, access rights, and controls. The documentation for Odoo 19 already illustrates this reality: supplier invoices can be scanned to pre-fill fields, bank transactions imported automatically, and certain operations reconciled according to defined conditions.

Saying that "ChatGPT does accounting" therefore confuses the visible interface with the entire system that works behind it. Automation is not free either: licenses, integration, configuration, document organization, security, training, supervision, and exception handling. It does not eliminate the cost; it shifts it to design and control.

What tools already know how to do — and what still needs to be verified

Generic advice is also becoming automatable. Generating a first explanation costs little. Determining if it is correct for this company, at this specific moment, can constitute the bulk of the work.

The less comfortable truth: some tasks will disappear

A accounting firm cannot indefinitely defend an invoice based on the number of lines entered. Repetitive, codifiable, and easily controllable tasks — copying data, classifying a standard document, applying a stable rule, or reconciling two identical amounts — quickly lose value.

This evolution also requires rethinking training. Young employees often learned the trade through basic operations before tackling difficult cases. If these tasks disappear, the fundamentals must be conveyed differently: commented files, supervised controls, exception analysis, and reasoning reviews.

Finally, being human does not, in itself, constitute a value proposition. A friendly but slow, imprecise, or opaque accounting firm is not protected against automation. Human value lies in judgment, competence, integrity, useful contradiction, and responsibility.

The more data entry becomes automated, the more the exception becomes the profession.

A system processes what it receives. It does not always know what is missing, what has been poorly described, or what the client has not yet understood themselves.

Should an invoice of CHF 18,000 titled "IT development" be expensed or capitalized? Is a private payment made by the company an advance to the shareholder, compensation, or a business expense? Is a service billed abroad really exempt from Swiss VAT? Is a sum received in December a product, an advance, or a deposit?

These questions require facts that are sometimes absent from the documents: contract, actual usage, intention of the parties, chronology, or information given orally.

A system can therefore perfectly process the received documents and completely ignore those that remain in an email box, a car, or the memory of the manager. Speed does not guarantee completeness.

The human relationship is not a decorative supplement.

The human argument is often reduced to empathy. We need to be more precise. An artificial intelligence can already integrate risk tolerance, family goals, and several contradictory scenarios. It can even produce a thoughtful response.

The professional, however, retains a particular role when discovering information that has not yet been formulated: a hesitation, two incompatible objectives, a tax question that hides a cash flow problem, or a structural choice mainly related to a partner, family, or succession.

The choice between a sole proprietorship and an LLC thus involves responsibility, taxation, social insurance, private needs, growth, and succession. The choice between salary and dividend links corporate law, taxation, social security, retirement planning, and cash flow. AI can assist these decisions; it does not automatically reduce them to a questionnaire. And sometimes, the best advice is less about confirming than knowing how to say: "yes, but here’s what you are agreeing to."

Illustrative example: 88% automatic, 12% decisive

Let’s take a service LLC that records 500 transactions over a month. The system automatically recognizes and matches 440 regular operations. These figures are purely illustrative.

Among the remaining 60 transactions are a private trip paid by the company, a foreign software subscription, a January invoice concerning December, an IT project that may need to be activated, and a customer payment lower than the invoice.

Automation creates efficiency in the 440 operations. But the quality of the accounts can hinge on a few atypical situations: VAT, profit, partner current account, or margin may depend on it. Exception management is therefore not an ungrateful remnant; it becomes the center of value.

Automating does not automatically transfer responsibility

For a corporation, the board of directors retains the responsibility to organize accounting, financial control, and, when necessary, financial planning. These responsibilities are non-transferable and inalienable. The managers of a limited liability company have comparable duties. Delegating the bookkeeping to a accounting firm or using an automated tool does not eliminate the duty of organization and oversight.

Conversely, the accounting firm is not a simple conduit without responsibility. Depending on the mandate, it must execute the agreed services diligently, apply the announced controls, and report any anomalies within its scope.

In case of an error, the client may argue that the accounting firm promised complete coverage. The accounting firm may respond that the data was incomplete, that the alert was ignored, or that the control was outside the mandate. The publisher will invoke the limits of the service and the expected validation from the user. The outcome will depend on the contract, the information provided, the accesses, the validations, the warnings, and the causal link.

The best protection is therefore not a statement claiming that "AI takes care of it." It is a documented process indicating who does what, who validates what, and how an exception is handled.

Seven questions a small business should ask its accounting firm

1. What operations are actually automated?
2. What human controls are performed, how often, and by whom?
3. How are exceptions and missing data detected?
4. Where is the data processed and how is access protected?
5. How are sensitive decisions documented?
6. Who validates the final result and who responds when an error occurs?
7. Does the productivity gain also improve the price, speed, or quality of the service?

The question is not whether to choose between a "human" accounting firm and a "technological" accounting firm. It is whether the organization properly combines both.

What will the accounting firm of tomorrow be for?

The accounting firm of tomorrow will likely do less data entry and more design, control, and interpretation. It will need to organize flows, supervise automations, control exceptions, translate numbers into consequences, connect accounting, taxation, law, HR, and financing, and then coordinate specialists when the case exceeds its scope.

It will also need to train its employees in judgment. A machine can propose an entry; the professional must understand why it is correct, under what circumstances it is no longer correct, and how to explain it.

It is not necessarily the fiduciary that disappears. It is part of its work, as well as its old business model.

How Delta Conseil SA can support you

At Delta, we have the habit — sometimes uncomfortable — of questioning our own methods. Artificial intelligence forces us to ask a simple question: among the services we provide, which ones actually create value for the customer?

A customer should not pay a person to copy information that a system can reliably extract. Automation should be used when it improves timeliness, traceability, or quality. The time saved should then be used to control, understand, anticipate, and advise — not just to produce more lines faster.

Delta Conseil SA can analyze the administrative and accounting processes of a small or medium-sized enterprise, reduce double entries, clarify roles and validations, organize document flows, and integrate controls into accounting, taxation, and management. When the file requires specialized expertise, we coordinate the appropriate intervention.

Frequently Asked Questions

Can ChatGPT independently establish the accounting of a small business?

It can analyze data, propose rankings, perform calculations, and prepare entries. However, complete accounting requires the collection of documents, a recording system, rules, validations, and the handling of exceptions.

Does a small sole proprietorship still need an accountant?

Not necessarily all the time. A freelancer with few operations, no staff, and no complex VAT can manage a large part of their administration with good tools. Periodic validation becomes useful when the fiscal, social, or documentary stakes increase.

Will AI lower accounting fees?

It should reduce the time spent on certain standardized tasks. However, the final price will depend on the cost of the tools, integration, control, complexity, and level of advice. The model primarily based on data entry is the most exposed.

Who is responsible when an entry proposed by AI is incorrect?

It depends on the mandate, the role of each participant, the agreed controls, and the information provided. The manager retains their organizational duties; the accountant may be liable for the services under their mandate; the user must verify the results according to the tool's conditions.

How to test the reliability of accounting automation?

It must be confronted with real cases, measure exceptions, verify access, document rules, impose validation for sensitive operations, and regularly control results. A high rate of automation is only valuable if significant errors are detected.

Official and reference sources

• International Labour Organization — Generative AI and Jobs: A Refined Global Index of Occupational Exposure, May 20, 2025.
• SME Portal of the Confederation — Generative AI: Catalyst for Growth and Transformation.
• OpenAI Help Center — Data analysis with ChatGPT.
• Odoo 19 — Digitization of Supplier Invoices.
• Odoo 19 — Bank Synchronization, bank reconciliation and reconciliation models.
• Swiss Code of Obligations, notably art. 398, 716a and 810.
• SME Portal of the Confederation — Board of Directors' Obligations.
• Delta Conseil SA — Generative AI and Data Protection: What Can a Swiss SME Share with ChatGPT?