Current studies show that AI tools can save lawyers between 40 and 70 percent of time spent on document analysis, research, and contract review. A Vals AI benchmark study from October 2025 found that AI systems outperformed lawyers in 150 of 200 standardised legal questions. The largest gains appear where repetitive document work meets high volumes.
What studies on AI time savings for lawyers actually prove
The evidence base has solidified over the past two years, and the figures no longer come exclusively from software vendors. The most methodologically transparent study is from Vals AI (published in beck-aktuell, October 2025): 200 legal questions drawn from eight major US law firms - including Reed Smith, Paul Hastings, and Paul Weiss - were answered by four AI systems and human lawyers. The result: AI averaged 74-78 percent of the total score; lawyers averaged 69 percent. On accuracy, AI systems reached 80 percent versus 71 percent for lawyers.
At the same time, lawyers dominated four of ten question types - specifically those requiring complex context, cross-border legal comparison, or judgment-heavy analysis. Their lead in those categories averaged around 9 percentage points. Anyone using AI time savings as a blanket argument is missing part of the picture.
An older but widely cited study concludes that lawyers spend roughly 60 percent of their working time on research, document review, and administrative tasks - and only 40 percent on actual advisory work. That ratio is the real driver behind AI adoption in law firms.
Time savings vs. revenue gains: why not every hour saved creates value
This is the point most studies avoid, and it matters especially for Swiss firms billing by the hour.
When an associate saves six hours, that does not automatically generate six billable hours of additional work. Three scenarios:
Scenario 1 - Capacity goes into mandate work. The recovered hours flow into more complex advisory tasks that had been deferred. This produces real revenue uplift.
Scenario 2 - Capacity goes into business development. The time feeds pipeline, not billing. Sensible in the medium term; no immediate revenue effect.
Scenario 3 - Capacity goes unused. The time saving evaporates because no mandate expansion is planned and no resource planning sits behind it.
Whether a firm actually captures value from AI time savings depends on a deliberate decision about how freed capacity is deployed. That sounds obvious. It is rarely planned explicitly in practice.
Where the time effect is largest - and where it is not
High-volume document review delivers the clearest results. A widely cited case study (Control Risks / Relativity): 100 contract lawyers needed 20 weeks to process 2 million documents at a cost of around 4 million euros. The same task with AI support: 2 weeks, 99.9 percent accuracy, roughly 25 percent of the original cost.
For individual contracts or complex negotiation situations, the savings are smaller and human judgment remains hard to replace.
For day-to-day practice in Swiss law firms, this means AI pays off most in due diligence processes with large document sets, in initial risk analysis of standard contracts, and in checking compliance with internal playbook standards across many contract positions.
A Zurich M&A boutique used the CASUS AI Data Room to analyse 340 purchase agreements from an asset deal mandate in around 4 hours - compared with an estimated 38 hours of manual work for the same clause mapping covering liability, IP, SLA, and termination periods. That represents a reduction of over 89 percent for the extraction step. The legal assessment of the results remained with the team.
What this means concretely for Swiss law firms
Since 14 June 2024, the Swiss Bar Association (SAV) has had a guidance document on the use of AI in legal practice. It states that using AI tools is generally permissible, but that professional due diligence under Art. 12 BGFA remains fully intact. AI outputs must be reviewed - this is a professional obligation, not a recommendation.
The revised Data Protection Act (revDSG), in force since 1 September 2023, places additional requirements on the use of cloud-based tools, particularly where personal data in mandate documents is processed. For firms serving FINMA-regulated clients - in M&A or capital markets work, for example - further confidentiality and data residency requirements apply.
Using AI tools that process data in US data centres is not categorically prohibited, but it requires justification. Tools hosted in Switzerland or the EU with a verifiable zero-data-retention architecture reduce this risk structurally.
Three operational observations from practice - and what they actually mean
Anyone who has accompanied Swiss law firms through AI tool adoption encounters patterns that rarely appear in studies.
First: AI systems reliably stumble over non-standard definitions. Contracts that define core terms such as "damage" or "breach" in locally deviating ways cause automatic risk analyses to misclassify findings or miss them entirely. The fix is not better AI, but a more precise review instruction: giving the model the local definition yields noticeably more reliable results. This step is routinely skipped in practice.
Second: the biggest time loss occurs after the analysis, not during it. In teams that work without a structured handover workflow, the person doing the review frequently spends longer transferring AI outputs into the document than the analysis itself would have taken. Direct integration into the working environment - via a Word add-in, for example - eliminates that bottleneck meaningfully.
Third: benchmark checks against playbooks work well, but only when the playbook is maintained. A common early disappointment: the internal playbook exists nominally but has not been updated in two years. The AI benchmark then measures against outdated standards - and produces false confidence. The real task is less about AI integration and more about playbook governance.
How CASUS maps this workflow
The CASUS Risk and Quality Review analyses contracts in a structured way by party perspective and prioritises findings by severity. The Benchmark workflow checks documents against internal playbooks or best-practice standards and surfaces missing clauses, deviations, and a percentage match score.
For legal research in the Swiss context, the Legal Research module searches over 660,000 cantonal and federal court decisions as well as statutory provisions, and returns results that are source-based, structured, and traceable.
Firms that want to test these workflows in their own mandate practice can start with CASUS for free at app.getcasus.com/signup without prior registration. The platform is hosted in Switzerland and the EU, does not transfer data to the US, and does not retain documents permanently.
FAQ
How much time do AI tools save lawyers on average?
It depends heavily on the task. For high-volume document analysis, studies show savings of 70 to 90 percent compared with manual processing. For complex legal questions the advantage is considerably smaller; the Vals AI study from October 2025 shows that lawyers outperform AI on context-intensive questions.
Are study findings transferable to Switzerland?
Partly. The major efficiency studies come from the US and the UK. Swiss firms work with different legal sources (CO, CC, Federal Court case law), different mandate structures, and additional requirements from the revDSG and professional secrecy rules. The order of magnitude of time savings on standard tasks is broadly transferable; the accuracy of AI outputs on questions of Swiss law needs to be assessed separately.
What does the Swiss Bar Association (SAV) say about using AI?
The SAV adopted a guidance document on AI use on 14 June 2024 (published in Anwaltsrevue 9/2024). It holds that AI tools may be used, but that the duty of care under Art. 12 BGFA remains fully intact. AI outputs must be reviewed by the lawyer.
Which tasks are best suited to AI in law firms?
Legal research (cited by 66 percent of surveyed lawyers as the top use case in the LexisNexis study), document analysis in due diligence, contract review against playbooks, and structured information extraction from large document sets. Complex judgment calls and cross-border legal analysis remain domains of human expertise.
May a Swiss law firm upload mandate documents to AI tools?
Yes, provided data protection and professional secrecy requirements are met. The key frameworks are the professional secrecy obligation under Art. 13 BGFA, the requirements of the revDSG on data processing by third parties, and - for mandates with EU exposure - the GDPR. Tools hosted in Switzerland or the EU, with zero data retention and no data transfer to the US, reduce legal risk structurally.
How reliable are AI tools for legal questions?
The Vals AI study shows accuracy of 80 percent for AI systems versus 71 percent for lawyers on standardised legal questions. For source citability, legal AI reached 76 percent. These results do not give a blank cheque: hallucinations occur, and on complex contextual questions lawyers retain the upper hand. Outputs must be reviewed.
What risks exist when using AI in legal practice?
Hallucinations (incorrect or fabricated content), data protection breaches from inappropriate tooling, and false confidence from unreviewed AI outputs. There is also a governance problem: when internal standards - playbooks, definitions - are not maintained, the AI measures against outdated reference points.
Does the EU AI Act affect AI use in law firms?
The EU AI Act (in force since 1 August 2024) applies to companies operating in the EU and will be relevant to Swiss firms with EU clients. Art. 4 EU AI Act has required operators of AI systems to implement AI literacy measures since 2 February 2025. High-risk classifications for legal systems take effect from 2026.
