Ever since Bexio acquired the AI company Kontera in October 2025, the same question has been hanging over many Swiss fiduciary offices: if the large platform provider will soon post documents fully automatically, is the trustee still needed? The honest answer is more uncomfortable than either camp believes. Yes, a substantial part of classic bookkeeping work is disappearing right now. And no, this does not make the trustee obsolete - their job is merely shifting. Those who understand this early will gain. Those who bet on the hype or on being displaced will lose.
This market overview sets out soberly which providers in Switzerland actually deliver something today, where the limits lie and which statutory obligations remain in place regardless of any software.
What AI in accounting can really do in 2026 - and what it cannot
Today’s technology is good at a clearly defined area: reading, classifying and reconciling documents. An AI agent recognises the content of an invoice, suggests a posting, books it and reconciles it against the bank or credit card transaction. This is not a thing of the future but is already in productive use at several Swiss providers.
What AI cannot do: judge whether a matter is correctly reflected for tax purposes, whether an accrual makes economic sense, whether a provision is appropriate, or whether the client would save money with a different structure. This is precisely where the trustee’s value lies - and that value has grown rather than shrunk, because the routine is being taken off their hands.
The useful rule of thumb for 2026:
- Capturing and reconciling (reading documents, posting suggestions, bank reconciliation): AI is strong here, often with a high hit rate.
- Checking and plausibility testing (does the suggestion hold up in context?): AI supports, the human decides.
- Advising and taking responsibility (financial statements, taxes, strategy): clearly stays with the human.
Market overview: these providers are genuinely active in Switzerland
The market in 2026 is hard to read, because every provider puts “AI” on its banner. Here are the verifiable facts on the most frequently named Swiss solutions - without promises we cannot check.
| Provider | What is verifiable | Status |
|---|---|---|
| Bexio / Kontera | Bexio (a subsidiary of Mobiliar) reported more than 90,000 SME clients at the time of the acquisition in October 2025; the mark of more than 100,000 clients was only reported in February 2026. Bexio acquired the AI company Kontera in October 2025. The Kontera AI was used by more than 2,500 companies and trustees; it reads documents, posts and reconciles. Data on Swiss servers. | Integration in phases from 2026, new packages from March 2026 |
| Arcarius | Swiss start-up, founded in late 2024 (CEO Isis Kratz, formerly ZKB; co-founder Robin Striebel, a trustee). AI agent for accounting questions and draft financial statements (“closing module”). Linked with Bexio at launch. | Live, connection to further systems announced |
| Infinity (infinity.swiss) | Swiss fintech by NextBusiness AG (Zurich Oberland), on the market since 2022. AI-supported accounting with document posting and e-banking reconciliation, also for fiduciary firms. | Live |
| TREUHAND|SUISSE GPT | Launched by the Treuhand Suisse association together with the Swiss company Connect AI. Industry-specific AI assistant for association members, based on Swiss expertise, with data held in Switzerland. Not a posting engine but a knowledge and answer tool. | Available to members |
An important point for context: these providers solve different problems. Bexio/Kontera, Arcarius and Infinity target operational bookkeeping (document in, posting out). TREUHAND|SUISSE GPT is a knowledge assistant for technical questions, not a posting engine. Anyone looking for “AI for fiduciary services” must first decide which of the two problems they want to solve.
The obligations no AI takes off your hands
This is where it gets serious, and it is precisely what gets lost in the hype. Whatever software you use, the legal responsibility stays with the company and with the appointed trustee. Three facts that apply regardless of any AI:
- Ten-year retention. Under Art. 958f CO, business records, supporting documents and business and audit reports must be retained for ten years. The period begins at the end of the financial year in which the last entry was made. It cannot be shortened by contract.
- Unalterability of archiving. Anyone who archives electronically must guarantee unalterability under the Ordinance on Business Records (GeBueV) - changes must be verifiable, and the archive must be capable of being made legible at any time. An AI that posts nicely but does not deliver an audit-proof archive does not solve this problem.
- Criminal responsibility. A breach of the bookkeeping obligation carries consequences. Art. 166 of the Swiss Criminal Code (SCC) makes the failure to keep accounts an offence (in the event of bankruptcy or a certificate of shortfall, up to three years’ imprisonment or a monetary penalty), and Art. 325 SCC covers the improper keeping of business records. This responsibility cannot be delegated to software.
On top of this comes the professional level: under the professional code of conduct, the trustee carries out their work on their own, full responsibility, and Treuhand Suisse requires its corporate members (fiduciary firms) to provide proof of professional liability insurance (minimum cover usually CHF 500,000). An AI has no liability. If it posts incorrectly and no one checks, it is ultimately the human who is liable.
What changes concretely for fiduciary offices and SMEs
For SME owners this means: pure data capture is becoming cheaper and faster - that is real and should be noticeable in your collaboration with the trustee. Anyone still charging per document captured is working on a shrinking business model. Anyone selling advice, tax optimisation and reliability is becoming more valuable.
For fiduciary offices the wave means a clear fork in the road:
- Automate the routine instead of defending it as billable hours. The tools for this are here.
- Position checking and control as a distinct, visible service - precisely the layer the AI needs in order to be trustworthy.
- Expand advisory work, because that is where the time freed up yields the highest return.
The pattern is not “human versus machine” but “human with machine versus human without machine”.
Our honest position: where we have evidence - and where we do not
At Vollmer Labs we do not talk about AI agents in accounting from theory. The decisive building blocks - automatically reading, posting and reconciling documents and deliberately bringing in a human for checking - run in production according to our own account: our CPA accounting agents work in our practice every day at a US fiduciary firm and take on real, recurring posting work there. This is not a prototype or a demo video, but day-to-day operation.
We are equally honest about what we cannot yet claim: a completed reference case with a Swiss fiduciary office using these agents end-to-end under local law (CO, GeBueV, VAT) is not something we have at this point. The building blocks are proven; the Swiss implementation in the fiduciary context is the next step, not the past. The fact that we also serve quite different industries with productive platforms - for example ballistic.club (live) and the RFQ tool rfqbuddy.com (in development), plus jeffri.ch for kitchen studios in a Swiss pilot - shows one thing above all: the agent technology is the same, only the technical rules change.
It is precisely these technical rules that are the sticking point in Swiss fiduciary services. An AI that does not cleanly reflect the ten-year retention obligation, the GeBueV requirements for unalterability and the VAT logic (mandatory from CHF 100,000 annual turnover) is not progress but a liability risk with a nice surface. The trustee therefore does not become obsolete. They become the human who vouches for the machine running within the right framework.