AI contract review means a system reads a draft, flags risks, measures it against a standard and points out missing clauses. For Swiss teams, the selection turns less on feature count than on three points: professional secrecy, coverage of Swiss contract law, and whether the tool works inside the Word document.
Contract review is not the same as general legal AI
A contract review tool works on the specific document; a general legal AI system answers legal questions. The distinction sounds academic, but it decides which tools belong on the shortlist at all.
Anyone reviewing thirty contracts a week needs risk detection, standard comparison and version diffing. Anyone asking how the Federal Supreme Court assesses a clause needs research in case law. Both are legal work, but a tool that is strong at one can be useless at the other.
This selection covers the first case only. A side-by-side of platforms across both jobs is in the independent comparison of the ten best legal AI tools.
Exclusion criteria come before the feature list
With client data, the legal position decides the field of possible providers before a single feature is compared. Reversing that order costs weeks evaluating tools that are ruled out anyway.
Professional secrecy under Art. 321 of the Swiss Criminal Code extends by its first paragraph to auxiliary persons. Anyone sending a client's contract drafts through an external system makes its operator an auxiliary person. On top of that comes the duty under Art. 13(2) BGFA to ensure that auxiliary persons observe the secret. This is not a formality that a privacy policy disposes of, but a question the provider has to answer contractually.
In data protection terms, processing on behalf under Art. 9 revDSG applies, and as soon as data leaves Switzerland, the requirements for disclosure abroad under Art. 16 revDSG. Three questions settle the point in one conversation:
Where does the data sit, and where does model inference run? Those are two different places, and the second is rarely volunteered.
Has abuse monitoring at the model provider been waived, so that no human review of inputs takes place?
Does the provider contractually accept being bound by professional secrecy as an auxiliary person?
What review depth can be measured against
Review depth shows not in the number of features but in which clauses a system recognises as problematic. Four capabilities differ markedly between providers: risk detection with a severity rating, comparison against an in-house standard, detection of missing clauses, and consistency across contract versions.
For Swiss contracts, though, the real difference lies elsewhere. Tools developed largely on Anglo-American contract work assess clauses by standards that do not apply here. Three examples that surface immediately in a test:
Limitations of liability. Under Art. 100(1) of the Swiss Code of Obligations, an exclusion of liability for unlawful intent and gross negligence is void. A sweeping liability waiver of the kind common in international templates is therefore ineffective to that extent. A tool that marks such a clause as favourable is reviewing under the wrong law.
Warranty periods. In sales law, warranty claims lapse two years after delivery under Art. 210 of the Code of Obligations. Adopting a template with a different period should be a deliberate choice, not an oversight.
Attribution of auxiliary persons. Under Art. 101 of the Code of Obligations, a party who delegates performance to an auxiliary person is liable for that person's conduct. Contracting out of that attribution runs into the limits of Art. 100 again.
Anyone comparing tools should put exactly these clauses into the test corpus. They separate candidates faster than any feature list.
Testing with your own contracts
A sound evaluation takes two to four weeks and works with real, closed contracts whose weak points are already known. Demo data only shows what the provider wants to show.
An approach that works:
Pick three to five closed contracts where it is clear in hindsight which points were problematic.
Run the same documents through every tool without naming the known weak points in advance.
Record two numbers: how many of the known points were found, and how many alerts were false positives.
Deliberately alter one clause so that it becomes ineffective under Swiss law, and check whether the tool notices.
The last step is the most informative and is almost always skipped. As a reference point for scale: at CASUS, an average contract review returns around twenty findings and takes one to three minutes. A list of three findings is too coarse; one with two hundred is unusable, because it does not shorten the read-through but relocates it.
Where the workflow breaks
Practical value is decided by where the result lands. A finding that sits in a web interface and has to be carried back into the document by hand costs more time on a thirty-page contract with forty adjustments than the review saved.
At CASUS, around eighty per cent of usage happens in the Word add-in rather than the web application. That is a clear indication of where legal work actually takes place, and a criterion that rarely appears in feature comparisons.
Marcel Held, Chief Operating Officer at Domenig Partner, describes the same point from the adoption side: "New software often fails because people do not adopt it. CASUS was quickly embraced by our team because it fits into existing workflows rather than replacing them."
Where CASUS fits in
CASUS is a platform that brings document work and legal sources into the same flow, not a pure contract review tool.
Three modules are relevant for contract review: Risk Review analyses risks from each party's perspective and gives every finding a severity, Benchmark measures a document against a stored playbook and reports the match as a percentage, and Proofread checks cross-references, definitions and numbering. Changes are applied in Word with the formatting intact. For larger contract portfolios, the AI Data Room extracts defined fields from many documents into a table.
Data is stored in Switzerland, model inference runs in Europe, and CASUS staff are bound by professional secrecy as auxiliary persons under Art. 321 of the Swiss Criminal Code. The details are on the security page.
Anyone wanting to check the fit can test CASUS free for fourteen days and use the procedure described above with their own closed contracts. Pricing is CHF 145 per user per month, or CHF 120 with annual billing (as of August 2026). Sign-up runs through app.getcasus.com.
A broader view of the Swiss market including research tools is in the comparison of legal AI tools for Switzerland.
FAQ
What distinguishes a contract review tool from a general AI assistant?
A contract review tool works on the document: it detects risks, compares against a stored standard and reports missing clauses. A general assistant summarises and rewrites, but knows neither the team's standard nor the contract's version history.
Which legal requirements apply first in Switzerland?
Professional secrecy under Art. 321 of the Criminal Code, which extends by its first paragraph to auxiliary persons, and the duty under Art. 13(2) BGFA to ensure auxiliary persons observe it. In data protection terms, processing on behalf under Art. 9 revDSG applies, and Art. 16 revDSG where data flows abroad.
How can you tell whether a tool reflects Swiss law?
Most reliably through a deliberately flawed clause in the test corpus. A sweeping liability waiver, for instance, is void for intent and gross negligence under Art. 100(1) of the Code of Obligations; a tool that lets it pass or marks it as favourable is reviewing by a foreign standard.
How long does a proper evaluation take?
Two to four weeks are enough when testing with real, closed contracts. Longer pilots usually fail not for technical reasons but because nobody finds time for structured testing.
How many findings are normal in a contract review?
At CASUS, an average review returns around twenty findings in one to three minutes. Substantially fewer suggests a superficial pass; substantially more suggests a list that relocates the read-through rather than shortening it.
Does it matter whether the tool runs inside Word?
Yes, because that is where work on the contract happens. A finding that has to be carried back into the document by hand consumes the time saved. At CASUS, around eighty per cent of usage happens in the Word add-in.
