Switzerland has one of the most complex tax systems in the world. Not because the individual provisions are particularly difficult, but because the system operates on three levels: federal, cantonal, and municipal. 26 cantons mean 26 tax laws, 26 tax ordinances, different tax rates, different deductions, different administrative practices, and different deadlines. On top of that come the federal legislation, the circulars of the Federal Tax Administration (FTA), and the case law of 26 cantonal administrative courts and the Federal Supreme Court.
For fiduciaries, tax research is daily business. And it is time-intensive. Not because the questions are difficult, but because the information is scattered across dozens of sources.
The Problem: Fragmented Information
A typical example from the daily work of a fiduciary firm. A client domiciled in the Canton of Zurich owns a property in the Canton of Bern and runs a sole proprietorship that also serves clients in the Canton of Aargau. The tax return involves three cantonal tax laws, the intercantonal tax allocation pursuant to FTA Circular No. 12, property taxation at the location of the property, and profit taxation at the place of registered office.
To answer these questions, the fiduciary must consult at least the following sources:
- The Federal Act on Direct Federal Tax (DFTA)
- The Tax Harmonisation Act (THA)
- The Tax Act of the Canton of Zurich
- The Tax Act of the Canton of Bern
- The Tax Act of the Canton of Aargau
- The respective tax ordinances
- Relevant circulars of the FTA
- Administrative practices of the cantonal tax authorities
- Federal Supreme Court rulings on intercantonal double taxation
Manual research takes hours. The fiduciary navigates between different databases, cantonal websites, PDF collections, and legal journals. Each source has a different format, a different search function, and a different update cycle.
Why General AI Falls Short Here
The obvious idea: why not just ask ChatGPT or a similar tool? The answer is as simple as it is sobering.
Hallucinations. General language models invent legal provisions that do not exist. They confuse cantonal regulations. They cite circulars with wrong numbers. They state tax rates that are outdated or entirely fabricated. For a non-binding initial assessment, that might be tolerable. For professional tax advisory, it is unacceptable.
Outdated data. Language models have a training data cutoff. Tax law changes annually. Tax rates are adjusted, deductions are modified, new circulars are published, the Federal Supreme Court issues landmark rulings. A model trained on 2024 data does not know the tax changes of 2025 and 2026.
No source references. Even if the answer happens to be correct, the fiduciary cannot use it without verifying the source. Without a reference to the specific statute, the relevant circular, or the pertinent Federal Supreme Court ruling, the answer is worthless. Fiduciary work requires source-based argumentation.
No cantonal differentiation. General models treat “Swiss tax law” as a monolithic construct. Reality is the opposite. The differences between cantons are substantial and practically relevant.
How AI-Assisted Tax Research Works
The solution lies not in general AI but in domain-specific AI built on verified legal sources. The principle is called Retrieval-Augmented Generation (RAG) and works in three steps.
Step 1: Verified data foundation. Instead of relying on training data, the system works with a curated knowledge base. This contains the current versions of all relevant laws, ordinances, circulars, and court decisions. The data comes from official sources: Fedlex for federal law, cantonal legislation portals for cantonal law, the Federal Supreme Court for case law. The database is continuously updated.
Step 2: Intelligent search. When the fiduciary asks a question, the system searches the knowledge base for relevant provisions. The search combines semantic similarity (the system understands the meaning of the question) with exact text search (the system finds specific article references, case numbers, and technical terms). Results from all relevant legal areas and cantons are brought together.
Step 3: Source-based answer. The language model formulates an answer based exclusively on the retrieved sources. Every statement is accompanied by a reference to the specific legal source. The fiduciary can access the original source with a single click and verify the accuracy of the answer.
Practical Use Cases
The following scenarios show how AI-assisted tax research concretely improves the daily work of fiduciaries.
Intercantonal tax allocation. A client has permanent establishments in three cantons. The fiduciary enters the basic data and receives within minutes an overview of the relevant provisions for each canton, the applicable allocation factors, relevant Federal Supreme Court rulings on allocation practice, and references to FTA circulars. Manual research that would take two to three hours is done in ten minutes.
Real estate capital gains tax. A client sells a property in the Canton of Lucerne. The calculation of the real estate capital gains tax differs considerably from canton to canton: monistic or dualistic system, calculation method, holding period deductions, replacement acquisition. The system delivers the specific provisions of the Canton of Lucerne, including the current tax rates, the applicable calculation formulas, and relevant rulings from the cantonal court.
Corporate tax reform. An SME is considering a restructuring. Merger Act, tax-neutral restructuring, loss carryforward, participation deduction: the questions are complex and involve federal law, cantonal law, and FTA practice. The system searches all relevant sources simultaneously and presents the provisions in a structured manner, with cross-references between the different legal areas.
Ongoing legal changes. At the beginning of 2026, new cantonal tax provisions came into force. The system automatically captured the changes and can reference the new versions in queries while simultaneously flagging which provisions have changed.
What a Good System Must Deliver
Not every AI tool is suitable for tax research. The following criteria distinguish professional tools from gimmicks.
Complete coverage. All 26 cantonal tax laws and ordinances, the DFTA, the THA, FTA circulars, case law from the Federal Supreme Court and cantonal administrative courts. Gaps in the data foundation make the system unreliable.
Timeliness. Tax law changes annually. A system updated monthly is borderline for professional use. Weekly or daily updates are the standard that professionals should expect.
Source attribution. Every statement must be traceable to a specific legal source. Article number, statute title, version. Without source references, the answer is worthless for professional work.
Cantonal differentiation. The system must recognise and accurately reflect cantonal differences. An answer that applies “in Switzerland” is insufficient. The fiduciary needs the answer for the specific canton.
Multilingual capability. Federal law exists in German, French, and Italian. Cantonal laws in their respective official language. A good system must search across languages and deliver results in the desired language.
Data protection. Fiduciary data is confidential. The system must process and store data in Switzerland. Transmitting client data to servers abroad is not an option for fiduciaries.
Efficiency Gains and Quality Improvement
AI-assisted tax research does not replace the fiduciary’s expertise. It replaces the time-intensive manual research. The fiduciary remains responsible for interpretation, strategic advice, and client communication. But the research that today consumes a large portion of working hours is reduced from hours to minutes.
In concrete terms, this means: more mandates with the same headcount, faster answers to clients, less risk from overlooked legal changes, and more time for the actual advisory work.
Enclava provides Swiss legal data as a verified, source-attributed AI knowledge base. With SwissLaw, the data foundation comprises over 27,000 laws and 1.1 million court decisions across all legal areas and cantons. Fully hosted in Switzerland, continuously updated, with complete source attribution.
Learn more about how Enclava can accelerate your tax research: enclava.ch or contact us at [email protected].