How to Run Ai-assisted and Online Mediation in Switzerland (2026): Process, Compliance Steps & Checklist

AI-assisted mediation Switzerland projects are entering a decisive phase in 2026, as the staged enforcement of the EU AI Act begins to intersect with Switzerland’s revised Federal Act on Data Protection and the rapid uptake of online dispute resolution platforms. For in-house counsel, dispute resolution managers and mediation providers, the practical question is no longer whether AI tools can play a role in settlement processes, but how to deploy them in a compliant, auditable and confidentiality-preserving way. This guide sets out a stepwise operational framework, scoping, procurement, data protection, running the session and record-keeping, mapped to Swiss and EU obligations as they stand in 2026.

It is written as a regulator-style procedural guide, with explicit timelines, document checklists and cost ranges rather than high-level commentary.

Overview, what this guide covers and when to use it

Quick summary

This guide explains how to operationalise AI-assisted mediation Switzerland workflows end to end. It covers project scoping and consent, vendor selection and technical evaluation, data protection impact assessment, pre-session testing, live session management, and post-session audit logging. Each step identifies the primary owner and a realistic duration. The guidance ties procedural decisions to the obligations that matter most in practice: transparency and logging requirements derived from the EU AI Act where EU parties are involved, and the data protection expectations set out under the revised Swiss Federal Act on Data Protection and the guidance of the Federal Data Protection and Information Commissioner (FDPIC). It is general information, not legal advice.

Who should read this

Read this if you are responsible for designing, procuring or supervising dispute resolution that uses online mediation Switzerland platforms or AI assistance such as automated transcription, summary drafting or suggested-settlement features. It is aimed at corporate dispute managers, procurement and IT functions, data protection officers, and mediation providers building ODR Switzerland offerings. It assumes you want a defensible compliance record, not merely a working tool. Where legal interpretation is required for a specific matter, seek specialist legal or regulatory advice.

Eligibility, when AI-assisted mediation Switzerland is appropriate

Types of disputes suitable

Commercial contract disputes, supplier and distribution disagreements, and many cross-border B2B matters are well suited to AI-assisted or online mediation. Higher-volume, lower-value consumer disputes also lend themselves to ODR Switzerland models, where automation can reduce cost and accelerate resolution. The common factor is that the parties can give informed consent, the subject matter does not depend on in-person credibility assessment, and the data involved can be processed lawfully.

High-risk features to avoid

Certain matters warrant caution or exclusion from AI tooling:

• Sensitive personal data. Disputes involving health, biometric, religious, trade-union or similar special-category data raise elevated obligations under Swiss data protection law and may trigger a mandatory data protection impact assessment.
• Criminal or quasi-criminal matters. Where conduct may expose a party to criminal liability, automated processing and transcript retention can create disproportionate risk.
• Automated decision-making on outcomes. Tools that effectively propose or impose binding outcomes without meaningful human control raise heightened transparency and oversight concerns and should be treated conservatively.
• Vulnerable parties. Where a power imbalance exists, AI-generated “suggested solutions” can distort voluntariness and should be disabled or tightly supervised.

Step-by-step process for AI-assisted mediation Switzerland

The full workflow, from scoping to retention, typically takes six to ten weeks of preparation before the first live session, with the mediation itself running over one to three days. The table below summarises owners and durations; the detailed steps follow. Timescales are indicative and will vary by matter and organisation.

Table A, Step, owner and duration

Step 1: Project scoping and consent framework

Owner: in-house counsel / dispute manager / mediator. Duration: 1–2 weeks.

Begin by defining exactly what the AI will and will not do. Produce four deliverables:

1. Scope statement. Set out the dispute type, the parties, the mediation objective and the precise role of any AI assistance (for example transcription and summary drafting only, with no outcome suggestions).
2. AI features inventory. List every AI function the platform offers and mark each as enabled or disabled for the matter. This inventory later feeds the risk classification and transparency disclosures.
3. Consent forms. Draft express, informed consent covering the use of AI features, the data processed, hosting locations, retention and the parties’ rights. In AI-assisted mediation Switzerland practice, consent should be documented before any data is uploaded.
4. Jurisdictional check. Confirm which parties are established in the EU, as this may determine whether EU AI Act transparency and logging obligations apply to the deployment, and whether cross-border transfer rules engage.

Step 2: Vendor selection and technical evaluation

Owner: procurement / IT / compliance. Duration: 2–4 weeks.

Evaluate mediation platforms Switzerland options against both functional and compliance criteria. Do not assess convenience alone; assess auditability, data residency, subprocessor transparency and the vendor’s ability to produce logs of AI inputs and outputs. The comparison below frames the trade-offs between traditional, online-only and AI-assisted models.

Require shortlisted vendors to supply a model description, a data processing addendum, a subprocessor list and evidence of encryption and access controls. A vendor that cannot describe its AI functions in writing cannot support a compliant AI-assisted mediation Switzerland deployment.

Step 3: Data protection and DPIA / AI risk assessment

Owner: DPO / compliance / adviser. Duration: 2–3 weeks.

Under the revised Swiss Federal Act on Data Protection, a data protection impact assessment is required where processing is likely to result in a high risk to the personality or fundamental rights of data subjects, for example large-scale processing of sensitive data or systematic profiling. Mediation transcripts, uploaded evidence and AI-generated summaries may meet that threshold. The FDPIC provides guidance on when an assessment is required and what mitigation it should document. Run the DPIA in parallel with an AI risk assessment that classifies each enabled AI function and records transparency, logging and human-oversight controls. Where residual high risk remains after mitigation, Swiss law may require prior consultation with the FDPIC before processing begins, unless an exception applies.

Step 4: Pre-session testing, security and user training

Owner: IT / mediator / provider. Duration: 1 week.

Before any live session, test the platform under realistic conditions. Verify encryption in transit and at rest, confirm role-based access controls, and run a failover rehearsal so that a connectivity or AI-service outage does not halt the mediation. Train the mediator and tech operator on how the AI assistant behaves, how to disable features mid-session, and how outputs are labelled. Record the training and have participants acknowledge it in writing, this evidence matters if the process is later challenged.

Step 5: Running the session, moderators, tech failover and AI-assistant rules

Owner: mediator and tech operator. Duration: 1–3 days per mediation.

During the session the mediator retains full control. Agree ground rules at the outset: when the AI assistant is active, how AI-generated summaries are presented and corrected, and that no AI output is treated as advice or as a binding proposal. The tech operator monitors system health and is ready to switch to a no-AI fallback. Transparency expectations mean parties should always know when they are interacting with or viewing AI-generated content, which is central to defensible AI-assisted mediation Switzerland practice.

Step 6: Post-session records, logs and audit trail

Owner: provider / parties. Duration: 1–7 days.

After the session, compile the complete record: the audit log of AI inputs and outputs with timestamps, the session summary, the signed agreement where settlement is reached, and the consent and training documentation. Apply the retention schedule agreed in Step 1 and confirm deletion triggers. A well-maintained audit trail is the single most useful artefact when demonstrating compliance with both data protection and EU AI Act logging expectations.

Table B, Required documents

Table C, Costs and fees (indicative ranges, CHF)

The figures below are illustrative estimates only and should be validated against current market quotations; they are not fixed or regulated fees.

Required documents and templates

Consent form template elements

A consent instrument for AI-assisted mediation should be specific and readable. At a minimum, include:

• Identification of AI features. Name each enabled function (transcription, summarisation, translation) and expressly state what is disabled.
• Data categories and purposes. Describe what personal data is processed, why, and by whom, including subprocessors.
• Hosting and transfers. State where data is hosted and whether any cross-border transfer occurs, together with the safeguard relied upon.
• Retention and deletion. Give the retention period and the deletion trigger.
• Transparency acknowledgment. Confirm the parties understand when they are viewing AI-generated content and that no AI output is binding.
• Rights of data subjects. Explain how a party can exercise their data protection rights and switch to a no-AI process.

Vendor SLA checklist and data processing addendum

The service agreement should bind the provider to concrete, verifiable commitments:

• Security standards. Encryption, access control and incident response obligations.
• Subprocessor control. A current subprocessor list, change-notification rights and flow-down obligations.
• Logging and audit access. Guaranteed export of AI input/output logs with timestamps.
• Breach notification. Defined notification windows and content requirements.
• Deletion on exit. Certified deletion or return of data at contract end.

Timeline and deadlines, typical project planning

Sample timeline

A first-time AI-assisted mediation Switzerland deployment typically runs on the following indicative schedule, which aligns with Table A. Scoping and the consent framework occupy weeks one and two. Vendor selection and procurement run across weeks two to five, overlapping slightly with scoping. The DPIA, AI risk assessment and data processing addendum negotiations occupy weeks four to six. Pre-session testing and training take place in week six or seven. The mediation itself is scheduled for one to three days thereafter. Post-session record compilation and audit logging complete within one to seven days of the final session. Retention and deletion then follow the agreed policy, which should reflect any applicable statutory retention requirements and ongoing enforcement exposure.

Where the same framework is reused for subsequent disputes, preparation compresses to one to two weeks because vendor, DPIA and template work is already in place.

Costs and fees, budget line items

Budgeting for AI-assisted or online mediation Switzerland should separate one-off set-up costs from per-case running costs. The largest one-off items are usually the external technical security assessment and the initial compliance advisory work. The data protection impact assessment, if outsourced, rises with complexity and the sensitivity of the data. Per-case running costs are typically dominated by mediator fees and platform and AI usage charges, which together vary considerably depending on features and volume. Translation, transcription and long-term secure storage are usually modest but recurring. A contingency line for ad hoc compliance queries is prudent. All figures should be validated against current vendor quotations and professional fee schedules before you finalise a budget.

Confidentiality, privilege and record-keeping in Switzerland

Swiss confidentiality rules and mediator duties

Confidentiality is foundational to mediation. Parties and mediators ordinarily agree in writing that statements made during mediation are confidential and are not to be relied upon in later proceedings, and lawyers acting as mediators are subject to professional confidentiality duties. The Swiss Civil Procedure Code also addresses mediation in the context of civil proceedings, including that mediation is confidential and separate from the court. When AI tooling is introduced, these duties do not weaken, but the surface area for inadvertent disclosure widens, because content is now processed and stored by a third-party platform and, potentially, its subprocessors. The contractual and technical controls described above exist precisely to keep the confidentiality promise intact once processing moves into the cloud.

Impact of AI and ODR on confidentiality and transcripts

AI transcription and summarisation create durable records that would not exist in a purely oral, in-person process. That raises two practical issues. First, a transcript or AI summary may later be sought in litigation; the mediation agreement should state clearly that such outputs are confidential and are not to be used as evidence, to the extent permitted by law. Second, AI-generated summaries can contain errors or inferences that were never said; parties should review and correct summaries before they are finalised, and the audit log should preserve the original inputs alongside any corrections. Treat every AI output as a draft subject to human verification, not as an authoritative record.

Cross-border issues

Where EU-based parties participate, confidentiality and data protection arrangements must satisfy both Swiss and EU expectations. Map any data flow between the EU and Switzerland and ensure the transfer safeguard, such as an adequacy decision, standard contractual clauses or equivalent contractual controls, is in place before processing begins.

Compliance with the EU AI Act and Swiss data protection, practical mapping

Which EU AI Act obligations matter for mediation platforms

The EU AI Act (Regulation (EU) 2024/1689), whose text is published on EUR-Lex and summarised in the European Commission’s regulatory framework for AI, classifies AI systems by risk and attaches graduated obligations that apply on a phased timetable. For mediation deployments the key questions are: does the system fall into a high-risk category, and does it trigger transparency duties? Transparency obligations, ensuring people are told when they are interacting with, or viewing content produced by, an AI system, are directly relevant to AI-assisted mediation platforms serving EU parties. Where a system is high-risk, obligations extend to areas such as risk management, data governance, technical documentation, record-keeping and logging, transparency, and human oversight.

Even where a Swiss-only deployment sits outside the Act’s territorial reach, aligning to its transparency and logging standards is a defensible baseline given the cross-border nature of commercial disputes. International guidance from the OECD’s AI principles and UNCITRAL’s work on online dispute resolution reinforces the same themes of transparency, human control and procedural fairness. Classification under the Act for a specific tool should be confirmed with specialist advice.

Swiss data protection overlaps and data transfer considerations

The revised Swiss Federal Act on Data Protection, available through Fedlex, runs in parallel with the EU regime on several points relevant here: the requirement to assess high-risk processing through a data protection impact assessment, the duty to inform data subjects, and the control of cross-border transfers. The FDPIC publishes guidance on transfer mechanisms and on when prior consultation is needed. In practice, a single, well-documented DPIA that also records AI-specific controls can address much of both frameworks, provided it covers transparency, logging, subprocessor oversight and the lawful basis for any EU–Switzerland transfer.

Note that Switzerland is generally treated by the EU as an adequate country for personal data transfers, while transfers from Switzerland abroad must rely on a recognised safeguard. Where interpretation of a specific obligation is uncertain, seek specialist legal or regulatory advice before deployment.

What changes in 2026, enforcement realities and immediate action items

2026 is characterised by the phased application of EU AI Act obligations and increasing scrutiny of AI governance across regulated and unregulated sectors alike. For dispute resolution teams the practical effect, as many industry observers expect, is that transparency and logging will increasingly shift from good practice to baseline expectation, and that vendor due diligence will be tested more rigorously. Three immediate action items follow: first, inventory every AI feature already in use across live mediations and classify its risk; second, confirm that each active deployment has documented consent, a DPIA where required and an exportable audit trail; and third, revisit vendor contracts to secure logging access and subprocessor transparency.

Teams that complete these steps now will find subsequent AI-assisted mediation Switzerland projects substantially faster to stand up. Confirm the precise applicability dates of each EU AI Act obligation against the current official timetable before relying on them.

Common pitfalls and how to avoid them

• Treating AI output as authoritative. Always mark AI summaries as drafts and require human verification before they enter the record.
• Skipping the DPIA. Omitting a data protection impact assessment for high-risk processing is a frequent and avoidable compliance failure.
• Opaque vendors. Accepting a platform that cannot describe its AI functions or export logs leaves you unable to evidence compliance.
• Weak consent. Generic terms-of-use consent does not substitute for specific, informed consent to AI features and data uses.
• Ignoring failover. No contingency for an AI or connectivity outage can derail a live session and the settlement opportunity with it.

Next steps

Running compliant AI-assisted mediation Switzerland processes in 2026 is an operational discipline: scope tightly, choose auditable vendors, document consent and risk, keep a clean audit trail, and retain or delete records on a defined schedule. Use the timeline, document checklist and cost table above to build a reusable framework, and align your transparency and logging practices to EU AI Act expectations even where only Swiss parties are involved. Where a specific matter raises questions of legal interpretation, seek specialist legal or regulatory advice before deploying. You can also explore the Mediation, Switzerland practice area and the GLE lawyer directory, Switzerland mediation for further resources.

Sources

1. Swiss Confederation, federal government portal
2. Fedlex, Swiss Federal Law database
3. Federal Data Protection and Information Commissioner (FDPIC / EDÖB)
4. European Commission, Regulatory framework for AI
5. EUR-Lex, EU law portal
6. OECD, AI Principles and guidance
7. UNCITRAL, ODR resources and technical notes