AI for Fiduciary and Tax Advisory Firms: Everyday Use
Where AI helps in a Swiss fiduciary or tax office, from engagement letters and VAT to onboarding, what it must not do, and how to start small.
In practice8 min read
A Swiss fiduciary (Treuhand) office builds up a large volume of documents over the years: engagement letters, annual accounts, tax returns, tax assessments, objections, VAT returns, file notes, emails with cantonal tax offices. The knowledge is all there. Finding it takes time, especially when the person who handled the file is on holiday or has left.
The biggest benefit of AI in fiduciary and tax work is in daily routine: finding documents faster, summarising files and drafting from your own records.
Where AI helps in a fiduciary office
The use cases below assume the AI can access your document store and cites its source with every answer. A general chatbot with no access to your files cannot do most of them.
Engagement letters
Every client has their own engagement letter, often amended over the years. The AI answers questions like these in seconds and points to the clause:
- "Is payroll included in the engagement with Muster GmbH, or billed separately?"
- "Which engagements have a notice period of less than three months?"
- "Where have we agreed a liability exclusion for tax advice?"
VAT
Many questions come up again and again: which method does a client use, which net tax rate applies, how was an unusual case handled last year? The standard Swiss VAT rate has been 8.1 per cent since 2024, which every professional knows. What they do not carry in their heads are the details of each client.
- "Does Bäckerei Huber file under the net tax rate method, and since when?"
- "How did we declare own use for the property renovation in 2024?"
- "Is there a note on the VAT audit for this client?"
Tax assessments and correspondence with the tax office
A tax assessment differs from the return that was filed. Before preparing an objection, the clerk wants to know how the tax office argued in similar cases and how the firm replied back then.
- "Summarise the correspondence with the tax office on this file since 2022."
- "Have we filed objections over disallowed property maintenance costs before? With what outcome?"
- "Draft a reply to the tax office based on our objection from last year."
The draft is a starting point. The responsible person checks the technical arguments, amounts and deadlines.
Year-end closing
From January to June, annual accounts pile up. Questions about previous years cost time, especially when a client has changed staff or advisers.
- "Which provisions has Keller AG booked in the last three annual accounts, and on what grounds?"
- "What does the file note on the 2025 closing meeting say?"
- "Which open points did we note for the following year at the last closing?"
Onboarding new staff
New colleagues ask a lot of questions, and experienced staff are least available in peak season. With an AI that knows your internal guidelines, checklists and sample files, newcomers can find much of it themselves:
- "What is our process when a new client brings in a tax return?"
- "Where is the template for a power of attorney towards the tax office?"
- "Which documents do we request for a property statement?"
The same approach helps when experienced staff leave. Their knowledge stays in the files they leave behind.
What AI must not do in fiduciary and tax work
No autonomous filing. An AI does not submit tax returns, VAT returns or objections. It prepares; a person checks and signs.
No unchecked amounts or deadlines. Language models make mistakes, even when they sound convincing. Every figure and every deadline is checked against the original source. That is why citations to document and page matter: they make the check quick.
No legal answers from general knowledge. Without access to current circulars and guidance notes, an AI answers questions on cantonal tax practice unreliably. Add the relevant publications from the Federal Tax Administration (ESTV) and your canton to the knowledge base, or check there.
No shifting of responsibility. The duty of care towards the client stays with the firm, whichever tool wrote the draft.
Confidentiality and client data
Fiduciaries and tax advisers work with exactly the data the Data Protection Act protects most closely: income, assets, family circumstances, sometimes medical costs. Under Art. 62 FADP, anyone who intentionally discloses such secret personal data learned in their profession commits an offence, with fines of up to CHF 250,000. Firms acting as statutory auditors are additionally bound by professional secrecy under Art. 321 of the Criminal Code. On top of that come the confidentiality clauses in your engagement letters.
In practice, client data does not belong in a private ChatGPT account. Whether an enterprise subscription, local AI or hybrid AI suits your office depends on your clients and contracts. For a fuller picture, including a sample staff policy, read ChatGPT and data protection in Switzerland.
For fiduciary offices, a setup where client files stay on your own server is the obvious choice. Either the language model runs there too (local), or only the question and a few short excerpts go to a cloud model (hybrid).
Start small
An AI project in a fiduciary office does not have to start with every file. One possible sequence:
- Pick one area, for example internal guidelines and checklists, or engagement letters. These are manageable volumes, and answers are easy to check.
- Choose two or three test users, ideally one experienced professional and one new hire. One spots wrong answers; the other shows which questions people really ask.
- Test for four weeks. Collect real questions from daily work and note whether the answer was right and the source fitted.
- Settle access rights before you expand. Payroll data, personnel files and especially sensitive clients should only be visible to authorised staff.
- Then add correspondence, annual accounts and tax files step by step.
Plan the start outside peak season. Nobody wants to try a new system in March.
KADARAG in a fiduciary office
KADARAG is an AI assistant for your document store that runs on your own server. It searches engagement letters, correspondence and annual accounts, answers with citations to document and page, and takes access rights from your existing file system. You can run it fully local, or hybrid without a graphics card of your own.
Next step
Write down ten questions your office asked this week whose answer is somewhere in your files. That list is the best test for any AI solution, ours included. If you would like to go through it with us, book an intro call.