AI-assisted contract review and manual review by lawyers differ on more than just speed. Experienced lawyers bring negotiation context and client strategy; AI delivers structured risk analysis in minutes rather than hours, without fatigue at high document volumes. Using either method well requires a clear understanding of where each one stops.
Two methods, two logics
Manual contract review works linearly: a lawyer reads a document, marks risks, and formulates recommendations. For a mid-sized supplier agreement of 40 pages, this takes between four and eight hours depending on complexity and experience level. In a due diligence project covering 80 contracts, that effort multiplies accordingly.
AI-based review works differently. The model analyzes the entire document at once, assigns risks to the relevant parties, prioritizes by severity, and proposes concrete drafting alternatives. Processing time is not linear – it barely increases with document length.
A practical example: the in-house legal team of a Geneva-based pharmaceutical company uploaded 47 supplier agreements into CASUS AI Data Room. Extracting liability clauses, termination periods, and data protection provisions – fields that would otherwise occupy three associates for two weeks – was completed and ready to export in approximately four hours. A single senior counsel then handled the legal assessment of the most notable deviations.
Time and capacity: the often underestimated dimension
The time advantage of AI is measurable, but tends to be framed incorrectly in practice. The question is not "AI or lawyer" but: how many hours of qualified legal time are being spent on work a system can complete in minutes?
For a Zurich M&A team reviewing 30 to 40 vendor agreements each quarter, AI-assisted first-pass analysis shifts the associate hours toward actual negotiation points. The initial check – completeness of standard clauses, missing definitions, open placeholders – is done in a few minutes before the lawyer even opens the document.
The CASUS Benchmark workflow, for instance, shows not only whether a clause is missing but also the percentage match against the internal playbook. That gives the reviewer a clear prioritization before reading a single line.
Accuracy and completeness
Honesty matters here. AI systems reliably detect formal patterns: missing liability caps, undefined terms, contradictory termination periods, open placeholders such as "[●]" or "TBD". Anyone who has ever missed an inconsistency between the use of a term on page 3 and a different spelling on page 47 of a 90-page contract knows how fatigue-prone manual review of long documents can be.
Where AI is weaker: assessing whether a contractually permissible clause is actually acceptable for this specific client in this specific situation. A liability clause can be correctly drafted and still represent a strategic risk in a particular contractual relationship that is only visible with knowledge of the negotiation history.
A concrete example from Swiss IT services contracts: the regulation of source code access in the event of the provider's insolvency is missing from the majority of incoming standard agreements. AI reliably identifies the gap and can suggest a clause. Whether that clause should be negotiated, at what cost, and whether the client can enforce it, remains a legal judgment call.
Liability: who bears the risk when AI-assisted review is used?
This question is rarely answered clearly in practice, even though it matters considerably. AI tools are aids, not legal advisors. Liability for a review result stays with the responsible lawyer – just as it stays with the supervising lawyer when a paralegal drafts a document.
Art. 321 StGB extends the criminal professional secrecy obligation explicitly to auxiliaries. That means anyone using AI tools in a mandate relationship must ensure that the systems used do not jeopardize professional secrecy. The Swiss Bar Association (SAV) addressed this directly in its "Guidance on the Use of AI", adopted in June 2024 and published in the Anwaltsrevue 9/2024, which explicitly identifies problems with tools hosted on US server infrastructure.
For law firms and in-house teams, this produces a concrete requirement: hosting in Switzerland or the EU, no data transfer to the US, no storage of client data after the session. CASUS meets these requirements through Switzerland and EU hosting combined with Zero Data Retention – documents are not retained after each session. Details are on the CASUS security page.
What AI still does not do – observed in practice
Beyond strategic assessment, there are further limitations that surface regularly in practice.
AI systems do not detect implicit expectations from the course of negotiations. If a party verbally agreed in the last negotiation round to accept a particular clause, and the document deviates from that, the AI finds the deviation only if it receives the baseline document as a reference.
AI does not determine whether a clause is valid under Swiss law when the question requires a case-by-case assessment. The CASUS Legal Research function searches across more than 660,000 cantonal and federal court decisions for relevant case law – but the application of that law to the specific facts remains legal work.
And: AI does not catch false statements in contract documents that are internally consistent but factually incorrect. A wrongly typed annual revenue figure will not trigger any alert from any AI tool.
Comparison table: AI review vs. manual review
Criterion | Manual review | AI-assisted review |
|---|---|---|
Processing time (single contract) | 4–8 hours | 5–15 minutes |
Scalability (volume) | linear, resource-limited | near-constant |
Formal completeness | fatigue-dependent | consistently high |
Strategic assessment | high | low |
Negotiation context | present | absent |
Swiss hosting / data protection | depends on tool | CASUS: CH/EU, Zero Data Retention |
Playbook comparison | manual, time-intensive | automated with percentage score |
Cost efficiency at volume | high (hourly billing) | significantly lower |
What to look for when choosing a tool
Selecting a legal AI tool is not a technology question – it is a compliance and liability question. Swiss law firms face different requirements than US or UK firms using platforms such as Harvey or Spellbook.
Concrete points to check: Where is the data processed? Is the model trained on client data? Is there an abuse monitor, and can it be disabled? Does the tool support working directly in Word, or does it require switching between platforms?
A Basel-based in-house team reviewing cross-border SaaS agreements under Swiss law with international counterparties has different needs than a Lausanne boutique firm handling mid-market acquisition contracts. A tool that supports both meaningfully must combine analytical flexibility with clear data sovereignty.
Teams starting with AI-assisted contract review can test CASUS through the web app or the Microsoft Word add-in without uploading client data before internal compliance questions are resolved. A free account is available at app.getcasus.com/signup.
FAQ
How does AI-assisted contract review differ from manual review?
AI review analyzes documents in a structured and fast way: it identifies missing clauses, contradictions, and deviations from a standard in minutes. Manual review brings negotiation context, client-specific judgment, and strategic assessment – the two approaches complement each other.
Can AI fully replace manual legal review?
No. AI reliably detects formal risks but cannot handle client-specific negotiation goals, implicit agreements, or legal case-by-case assessments. Legal responsibility remains with the lawyer.
Who is liable if an AI-assisted review misses an error?
Liability stays with the responsible lawyer. AI tools are aids, comparable to a paralegal draft. Art. 321 StGB and the SAV guidance from 2024 specify the obligations when using third-party tools in a mandate relationship.
Is AI-assisted review compatible with Swiss professional secrecy rules?
Yes – provided the tool meets the relevant requirements: hosting in Switzerland or the EU, no data transfer to the US, no training on client data, and Zero Data Retention.
How accurately does AI detect missing contract clauses?
For clearly defined review fields, the detection rate is high. AI reliably identifies the absence of a liability cap, a data protection provision, or a termination period. Whether that absence is acceptable in a specific contract remains a legal question.
How long does an AI-assisted first review of a contract take?
Depending on document length and complexity, a structured first analysis takes between five and fifteen minutes. A manual review of the same document by an experienced lawyer typically takes four to eight hours.
Which Swiss legal areas do current legal AI tools cover?
Modern legal AI tools are language model-based and are not restricted to specific areas of law. CASUS additionally includes a Legal Research function covering more than 660,000 cantonal and federal court decisions, enabling source-based preliminary research under Swiss law.
Is AI-assisted review worthwhile even for small contract volumes?
Yes. Even for individual contracts, AI-driven first-pass analysis significantly reduces time spent on formal review steps. The benefit scales upward at higher volumes – due diligence projects and standard contract portfolios in particular.
