Aller au contenu
← Back to the blog
AI · 6 min read · Updated 13 October 2026

AI in Swiss fiduciary firms: how to adopt it in 2026

How Swiss fiduciary firms can adopt AI in 2026: choose a first use case, protect client data and measure practical results without losing accounting control.

Équipe en réunion de travail

In 2026, a Swiss fiduciary firm can adopt AI by starting with a narrow process, such as preparing supplier invoices for bookkeeping, with human approval and explicit confidentiality rules. Success means reliable entries and genuinely avoided work, not a longer list of enabled features. Without an identifiable survey and verifiable methodology, no national adoption rate can be presented here as an established fact.

How widespread is AI adoption in Swiss fiduciary firms?

“Adoption” covers very different activities. Asking an assistant to rephrase an email is not equivalent to using it to prepare accounting entries from confidential client records.

When comparing products or assessing your own firm, distinguish three levels:

Level Example Main consideration
Occasional assistance Drafting a message without client data Check the generated content
Accounting preparation Extracting data and suggesting an account Check the source and accounting treatment
Workflow integration Sending a proposal to accounting software Control access, approvals and audit trails

The percentages and testimonials in the previous version of this article have not been retained because the available information does not allow them to be verified. This does not establish that no studies exist. It means an undocumented estimate is not a sound basis for an investment decision.

The more useful question is: which process can your team improve without weakening its controls?

Which process should a fiduciary firm start with?

Why test supplier invoices first?

Supplier invoices provide a manageable starting point if your firm handles them regularly. The original document gives reviewers an identifiable reference against which to compare extracted information.

A pilot can be limited to:

  • extracting the supplier, date, amount and currency;
  • identifying the VAT shown on the document;
  • suggesting an account and cost allocation;
  • flagging possible duplicates or missing details;
  • preparing a proposed entry for approval rather than posting it automatically.

Reading a Swiss QR-bill helps capture certain payment details. It does not establish that an expense is legitimate, the account allocation is correct or input VAT is deductible.

Which tasks should wait?

Avoid starting with complex tax decisions, unusual year-end adjustments or autonomous payments. Payroll also requires particular care: personal data, AVS/AHV, LPP/BVG and withholding tax rules involve much more than text extraction.

An assistant can prepare an answer or flag an anomaly. It should not become the default authority deciding the accounting or tax treatment.

How do you organise a useful, measurable pilot?

Define your testing method before choosing a tool. Otherwise, an impressive demonstration using a few clean documents may conceal everyday difficulties.

1. Set a clear scope

Choose one document type, suitable client assignments and a responsible person. Include ordinary invoices alongside credit notes, foreign currencies, poorly legible documents and known exceptions. Use fictional or properly anonymised data until the data protection arrangements have been assessed.

2. Involve the operational team

An experienced bookkeeper should participate in testing rather than discovering the product after purchase. Their job is to identify errors, assess their seriousness and record the effort required to correct them.

Also specify who can change settings, approve proposals and stop processing. Staff must be able to report mistakes without their feedback being treated as resistance to change.

3. Measure the entire job

Compare batches of similar complexity. Include preparation, checking, corrections and exception handling, not just extraction time.

Measure What it tells you
Proposals accepted without correction Practical usefulness of preparation
Errors by category Weaknesses involving amounts, accounts or VAT
Total processing effort Net effect, including review
Unresolved exceptions Work pushed back onto the team
Traceability Whether a decision can be reconstructed

Set your continuation criteria before testing. There is no universally acceptable error rate: an imperfect description and a payment to the wrong beneficiary have very different consequences.

What human review should remain?

During the pilot, check every proposal before using it in the accounts. Any subsequent reduction in checking must be justified by observed results and process risk, not simply by how long the tool has been used.

Retain explicit approval for sensitive transactions, including changed payment details, uncertain VAT treatment, unusual entries and contradictory documents.

Bookkeeping remains subject to the applicable requirements of the Swiss Code of Obligations, Articles 957 onwards, available through Fedlex, the federal legislation platform. Software does not replace those requirements or the responsibilities of the business and the professionals involved in their respective roles.

Maintain a usable audit trail: source document, proposal, correction, approving person and final entry. A confidence score generated by a system is not, on its own, evidence of accounting accuracy.

How can you check whether a tool suits Swiss requirements?

A general-purpose assistant can help with drafting and summarising. For accounting preparation, however, it must be tested against your actual rules, data and exceptions.

Swiss VAT rates are 8.1%, 2.6% and 3.8%, effective since 1 January 2024, according to the Federal Tax Administration, FTA. Recognising a rate is not enough: the nature of the transaction, VAT accounting method and conditions for input tax deduction also matter.

Check specifically whether the solution supports:

  • each client's chart of accounts and VAT codes;
  • credit notes, currencies and rounding;
  • transferring proposals without creating duplicate entries;
  • exporting data and change histories;
  • consistent behaviour after a model update.

For payroll matters, consult the official information on ahv-iv.ch. Swissdec certification covers a defined software and transmission scope. It does not certify the accuracy of every answer produced by an AI assistant.

How should you protect client data?

Review contracts and data flows before uploading real documents. Even an ordinary invoice can contain personal data and confidential commercial information.

The Federal Act on Data Protection, FADP, and guidance from the Federal Data Protection and Information Commissioner, FDPIC, are the official references to consult.

Ask the provider:

  • Where are documents, backups and logs processed?
  • Which subcontractors and staff can access them?
  • Is content used for model training, and what contractual guarantees apply?
  • What are the retention periods and deletion procedures?
  • How is data encrypted in transit and at rest?
  • How are incidents, access rights and international transfers managed?

Swiss hosting alone does not guarantee compliance. Conversely, processing abroad is not automatically prohibited: the applicable legal and contractual conditions need to be assessed.

Do not treat “end-to-end encryption” as a universal shortcut either. Many services need to process readable data within their execution environment. Understand precisely where this happens and what safeguards apply there.

What changes for staff and SME clients?

The aim is to shift effort from repeated data entry towards checking, analysis and useful client discussions. That requires training on the tool's limitations, escalation procedures and data protection.

It does not justify guaranteeing that no jobs will disappear or claiming an automatic recruitment advantage. Employment effects depend on the firm's organisation and management decisions.

For SME clients, assess practical outcomes: more complete records, reports delivered when agreed, explained corrections and clear responsibilities. Automated extraction will not necessarily detect an invoice that was never submitted.

For SMEs that prefer to delegate, Sequence Accounting provides a fixed-price accounting service: clients upload their documents and the team handles the bookkeeping. This illustrates an important distinction: decide who will do the work before selecting the technology that supports it.

What should you decide after the pilot?

Expand gradually if quality, confidentiality and total effort are satisfactory. Narrow or adjust the scope if exceptions consume the expected benefit. Stop testing if access controls, traceability or critical errors remain unresolved.

In 2026, useful AI adoption in a fiduciary firm depends less on promised autonomy than on a documented, controlled and reversible process.

Frequently asked questions

What percentage of Swiss fiduciary firms use AI in 2026?+

No verified national percentage is stated here. Before citing a figure, check the original publication, sample and definition of use: occasional assistance is not the same as integration into accounting workflows.

What is the best first AI use case for a fiduciary firm?+

Preparing supplier invoices offers a bounded test: extract details, suggest account allocations and flag possible duplicates, with human approval before posting.

Can client documents be uploaded to an AI service?+

Only after checking the legal and contractual framework, access arrangements, subcontractors, retention and any international transfers. Use fictional or properly anonymised data for initial testing.

Can AI replace accounting review?+

AI can prepare work and flag anomalies, but it does not remove applicable responsibilities. Review should reflect risk and demonstrated results, with explicit approval for sensitive transactions.

Less time on admin, more on your business.