Which admin tasks can AI take on in a Swiss Treuhand office?

AI is good at the collecting, sorting and retyping that fills a fiduciary office's year: requesting and chasing client documents, pulling data out of invoices and receipts, triaging client email, tracking deadlines and finding internal know-how. It should not own the figures, the tax judgement or the advice, which stay with a qualified person who checks what AI prepares.

Zürich firms are not short of goodwill towards AI. In the ZKB's KMU-ZH Monitor 2026, the most cited obstacle was lack of know-how (40%), and mid-sized firms raised data protection and security concerns more often than small ones (46% against 38%). [3] This article is meant to close the know-how gap for a Treuhand office of 5 to 50 people.

1. Requesting and chasing client documents

Every year the same exercise: a list of what each client must send, a request, and three reminders. AI can keep a checklist per client, draft the request in your tone, notice what has arrived and draft the next reminder for what has not. A person sends it, or approves a rule to send routine reminders automatically. This is usually the quickest win, because it touches no figures at all.

2. Reading invoices, receipts and statements

Documents arrive as PDFs, phone photos and email attachments. AI can read them, extract supplier, date, amount, VAT and description, and prepare the data in your accounting software's import format. The important design choice is the review step: anything the model is unsure about is flagged, and someone checks before it is booked. Done that way, staff stop typing and start checking.

3. Triaging client email

A shared inbox in a Treuhand office mixes urgent questions, documents, newsletters and "just checking" emails. AI can sort them by client and topic, attach incoming documents to the right place, and draft replies to routine questions such as what is still missing or when a return will be filed. Anything that needs a professional view goes to the right person, with the context assembled.

4. Deadlines and reminders

VAT periods, tax returns, payroll changes and year-end closings live in a mix of spreadsheets and memory. A simple workflow can keep one list per client, remind the responsible person ahead of time, and flag where documents are still missing. It is unglamorous and saves a surprising amount of stress in March.

5. Finding internal know-how

Juniors spend time looking for last year's working papers, the right checklist or how the firm handled a particular case. A knowledge assistant over your own templates and procedures answers in plain language and links to the source. It must respect who may see which client file.

What to keep with people

• Tax and accounting judgement. AI can find the relevant note; a qualified person decides.
• Signing off figures that go into accounts, returns or reports.
• Advice to clients and anything that changes the relationship.
• Rare, one-off work. If a task happens twice a year, automating it rarely pays.

The data protection points to settle first

The FDPIC has confirmed that the revised Federal Act on Data Protection applies directly to AI-supported processing. [4] For a fiduciary, three points follow:

• The AI service is a processor. You may hand it processing only by contract, only if no duty of confidentiality prohibits it, and you must satisfy yourself that it can guarantee data security. It may only pass the work to a third party with your prior approval. [5] In practice: a business plan with a data processing agreement, never personal accounts.
• Do not assume you are exempt from keeping a record of processing activities. The Act lets the Federal Council exempt firms with fewer than 250 employees only where processing carries a negligible risk to the people concerned. [6] Client financial data is hard to call negligible, so add the AI service to your record.
• Plan for breaches. A data security breach likely to lead to a high risk must be reported to the FDPIC as soon as possible. [7] Know which AI services hold what, so you can answer quickly if something goes wrong.

This is not legal advice; check your own position with your adviser.

Buy before you build

If you are a TREUHAND|SUISSE member, look at TreuhandGPT first. The association launched it with the Zürich company Connect AI in December 2025, giving more than 2,000 member offices access to an AI platform operated on Swiss infrastructure. [1, 2] A shared platform will not know your inbox, templates or client list, which is where custom workflows earn their place. But it may cover the drafting and research side, and that changes what is worth building.

How I would start

1. Write down the hours a week each of the five jobs above takes, roughly. One morning's work.
2. Pick the one with the most hours and the least judgement. It is usually document requests or invoice extraction.
3. Settle the data protection points above before anything touches client data.
4. Run it for a month with a person checking everything, then decide whether to extend.

That is what the AI Opportunity Audit does, in one to two weeks and for CHF 750 to 1,500. More on how I work with fiduciaries and accountants.