Swiss manufacturing AI
AI for Swiss Manufacturers Without Sending Data Abroad
Short answer
Swiss precision manufacturers can ground AI on their SAP, Infor, or Oracle ERP without routing production or customer data through a foreign cloud region, satisfying both the revised Federal Act on Data Protection (revFADP) and the confidentiality expectations Swiss customers carry over from the country's banking sector. This page covers how to keep model inference inside Switzerland, or a contractually defined and properly assessed boundary, while still getting a grounded, ERP-aware AI layer.
- ERP
- SAP S/4HANA, Infor LN, Oracle EBS
- Industries
- Manufacturing, Precision Engineering, Medical Devices
- Written for
- CIO
Switzerland's manufacturing base is built around precision engineering: watchmaking, medical devices, specialty machine tools, and the broader MEM sector represented by Swissmem. Larger multinationals typically run SAP, and a meaningful share of machine tool and industrial equipment makers carry an Infor LN heritage from Baan. Most of these companies operate with lean IT teams that have real interest in AI but limited appetite for open-ended cloud dependencies.
The revised Federal Act on Data Protection, revFADP, came into force in September 2023 and tightened rules around cross-border data transfer. Switzerland is not an EU or EEA member, so it maintains its own adequacy framework under the Federal Data Protection and Information Commissioner, the FDPIC, which is closely related to but distinct from the EU's own adequacy list. Assuming an EU-hosted deployment automatically satisfies revFADP is a mistake worth avoiding early.
Beyond the legal minimum, Swiss industry carries a cultural expectation of discretion that traces back to the banking sector but shows up across precision manufacturing too. Watchmaking and specialty machine tool customers, in particular, often contractually require data to stay within Switzerland's borders even where the law itself would permit a wider boundary, because discretion is part of what they are buying alongside the product.
This page sets out the architecture and deployment options for keeping AI inference inside Switzerland or a properly assessed boundary. It describes mechanisms and options, not Swiss legal advice; get counsel for anything specific to your own contracts or regulatory position.
What usually gets in the way
The problems we hear most from cio teams running SAP S/4HANA.
Switzerland is not in the EU, so 'EU region' is not automatically the right answer
Cross-border transfer rules under revFADP require assessing the destination country's adequacy separately from the EU's own list, and Switzerland's FDPIC list is not identical to it.
Precision manufacturers guard product data as tightly as banks guard account data
Watchmaking, medtech, and specialty machine tool companies treat BOMs, tolerances, and supplier data with a confidentiality standard that public cloud AI defaults don't meet.
SME manufacturers can't justify a full on-prem GPU cluster for one AI use case
A Swiss precision shop with fifty to three hundred staff has real appetite for AI but not the volume to make dedicated infrastructure obviously worthwhile without a clear cost model.
Multilingual documentation fragments the knowledge base
Work instructions and supplier documents written in German, French, Italian, and English make a single grounded AI layer harder to build well.
revFADP's newer, stricter obligations push legal toward a conservative default
With stricter breach notification and higher individual liability for some violations, legal teams are pushing IT toward defaults that avoid cross-border ambiguity altogether.
Where AI earns its place in SAP S/4HANA
Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.
Tolerance and specification Q&A for precision parts
Engineers and quality staff ask about tolerance history and specification revisions for a part without digging through PLM and ERP separately.
Touches: Item master, engineering change records, quality specification tables
Outcome: Cuts the time spent reconstructing a part's specification history for a customer query or audit.
Supplier confidentiality-aware Q&A
Procurement asks pricing or capacity questions about a supplier relationship, with the agent respecting confidentiality flags on sensitive contracts.
Touches: Supplier master, contract terms, purchase order history
Outcome: Keeps sensitive supplier terms visible only to authorized procurement staff, matching existing ERP role restrictions.
Multilingual work instruction retrieval
Shop-floor staff ask questions in German, French, or Italian and get answers grounded in whichever language the underlying work instruction was actually written in.
Touches: Routing and work instruction documents, linked to production orders
Outcome: Reduces time operators spend searching for the right version of a work instruction.
Export classification support
An agent helps engineering and trade compliance staff check whether a part or assembly is likely subject to dual-use export control classification, based on its BOM and description, flagging cases for a compliance officer's review.
Touches: Item master, BOM, export classification fields where maintained
Outcome: Surfaces likely dual-use flags earlier in the quoting process instead of at shipment.
Small-batch job costing explanation
For make-to-order precision shops, an agent explains how a quoted or actual job cost was built up from labor, material, and machine time.
Touches: Job cost tables, labor and machine time transactions, material issues
Outcome: Speeds up cost review conversations between the shop floor and finance.
Regulatory document drafting for medtech suppliers
Quality engineers supplying Swiss medtech OEMs get a first-draft technical file section or supplier quality data sheet, grounded in ERP quality records.
Touches: Quality management records, inspection and test results
Outcome: Reduces drafting time on routine sections of supplier quality documentation.
Customer-facing status updates without exposing internal detail
An agent drafts order status updates for customers from the ERP without revealing internal cost or supplier detail not meant to leave the company.
Touches: Sales order and shipment or delivery tables
Outcome: Keeps customer communication accurate and current without manual redrafting by a planner.
Reference architecture
The architecture defaults to keeping inference physically inside Switzerland, avoiding the cross-border adequacy question rather than relying on a contractual workaround.
- 1
ERP connectors
Connects to SAP via OData/BAPI, to Infor LN via BODs and ION, or to Oracle EBS/JDE via their standard interfaces, without direct database access.
- 2
Data and semantic layer
Handles German, French, Italian, and English field names and documents, applying confidentiality flags already present in the ERP before data reaches the model.
- 3
Model serving
Runs on GPUs physically located in Switzerland, or on-prem at the plant, avoiding the cross-border adequacy question entirely rather than relying on a contractual workaround.
- 4
Retrieval and agents
Retrieval-augmented generation grounds answers in ERP and PLM data; agents proposing write-back actions route through the ERP's own approval workflow.
- 5
Governance and audit
Query and action logs are kept per user, giving IT a clear record for internal governance and for answering customer data-location questionnaires.
Integration notes for your ERP team
- Connects to SAP via OData/BAPI, Infor LN via BODs and ION, or Oracle EBS/JDE via their standard interfaces, without direct database access.
- Handles German, French, Italian, and English source documents and field labels without requiring the ERP itself to be re-localised.
- Confidentiality flags already present on supplier or contract records in the ERP are respected by the retrieval layer, not just the ERP's own screens.
- Model serving and any vector store are deployed on infrastructure physically located in Switzerland when that is the requirement, named explicitly in the contract.
- Export-classification support is read-only and advisory; it never auto-classifies or blocks an order without human review.
- Document grounding indexes PLM, quality, and supplier documents from existing systems, respecting current folder or system permissions.
Deployment options
Air-gapped on-prem
Precision shops, such as watchmaking, medtech, or specialty machine tools, where customers or internal policy expect data to never leave the building
Model serving runs on a server at the plant, sized for realistic usage rather than hyperscale capacity.
Private / sovereign cloud in a Swiss data center
Manufacturers wanting to avoid hardware ownership while keeping data inside Switzerland specifically
Dedicated GPU capacity in a Swiss data center, with a contract naming Switzerland, not just EU, as the processing location, directly answering the revFADP cross-border question.
Hybrid
Groups with Swiss headquarters and manufacturing sites elsewhere in the EU
Swiss entity data stays on Swiss infrastructure; other-country plants use their own region-appropriate deployment under the same governance model.
Compliance and data control
How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.
Revised Federal Act on Data Protection (revFADP / nLPD / nDSG)
Keeping model serving physically inside Switzerland removes the cross-border adequacy assessment question; where data must leave, the FDPIC's own list is checked rather than assuming EU adequacy applies automatically.
Swiss export control law and Switzerland's participation in dual-use regimes
AI-assisted classification flags are treated as a first pass for a human compliance officer, not a determination, keeping the actual export decision inside your existing process.
Customer confidentiality expectations beyond legal minimums
A Switzerland-only deployment lets you answer a customer's data-location question with a specific location, not a regional description, which matters to Swiss watchmaking and medtech buyers in particular.
ISO 13485 for medtech suppliers and related quality system requirements
AI-drafted quality documents stay inside the same document control and retention process your QMS already runs; the AI layer produces drafts, not approved records.
Where Netray fits
ERPray
Grounded question answering across SAP, Infor, or Oracle fits the everyday need for engineers, quality, and procurement staff to query the ERP directly.
Custom build
Export classification support and multilingual work-instruction retrieval are specific enough to precision manufacturing to warrant a tailored build.
How an engagement runs
Phase 1 . 2-3 weeks
Discovery
- -ERP and document source inventory
- -Data residency decision, Switzerland-only versus EU-acceptable, documented
- -Priority use case shortlist
Phase 2 . 6-8 weeks
Pilot
- -Working Q&A or agent for one department or plant
- -Deployment confirmed inside Switzerland or the agreed boundary
- -Multilingual grounding validated with real users
Phase 3 . 4-6 weeks
Production
- -Rollout to additional departments
- -Monitoring and access review in production
- -Documentation for internal or customer data-location questionnaires
Phase 4 . ongoing
Scale
- -Additional ERP or PLM sources connected
- -Quarterly review of usage and accuracy
- -New use cases prioritised with the business
Questions to ask any vendor, including us
A short list that separates real SAP S/4HANA AI work from a chatbot demo.
- Can you confirm, in the contract, the physical country where inference and any vector store run?
- How do you handle the revFADP cross-border transfer question if any component is hosted outside Switzerland?
- Does the system respect confidentiality flags already set on supplier or contract records, or only the ERP's own screen-level restrictions?
- Can the system ground answers in German, French, and Italian documents, not just English?
- What happens to export-classification flags - does the system make a determination or just surface a flag for review?
- What is the real cost difference between an on-prem server at our plant and a Swiss private cloud instance?
- Who has access to the model weights, logs, and data if your company is acquired or ceases operating?
Frequently asked questions
Is Switzerland covered by the EU's adequacy decisions for data transfer?
Switzerland is not an EU member and has its own adequacy framework under the revised Federal Act on Data Protection, overseen by the FDPIC. It is not automatically the same as relying on an EU adequacy decision, so cross-border transfers from Switzerland need their own assessment against the FDPIC's list rather than assuming EU rules cover it.
Do we need a data center physically in Switzerland?
Only if your customer contracts, internal policy, or your own risk appetite require it. Many manufacturers are comfortable with a well-documented EU deployment; precision manufacturing and medtech suppliers more often want Switzerland named specifically, which is straightforward to specify in the deployment contract.
Does revFADP apply the same way as GDPR?
revFADP is closely modelled on GDPR but is a distinct Swiss law with its own regulator, the FDPIC, its own cross-border adequacy list, and some differences in scope and individual liability. Treat it as related but not identical, and get Swiss legal advice for anything beyond the general architecture choices this page covers.
Can AI help with export control classification for dual-use precision parts?
It can surface likely classification flags based on the part description and BOM, which speeds up the first pass, but the actual classification decision should stay with your trade compliance officer; treat the AI output as a prompt for review, not a determination.
How does this compare to using SAP Joule or a similar vendor AI feature?
Vendor AI features are convenient where you are fully on that vendor's cloud, but many Swiss precision manufacturers run mixed or on-prem environments specifically to avoid that dependency; a private layer alongside the ERP gives you comparable AI capability without requiring a move to the vendor's multi-tenant cloud.
What is a realistic pilot timeline for a Swiss precision manufacturer?
A single-department pilot, after two to three weeks of discovery to confirm the data residency boundary, typically takes six to eight weeks to reach a working system with real users.
Do small Swiss SMEs really need dedicated GPU hardware?
Not necessarily. A private cloud instance in a Swiss data center avoids the capital cost of on-prem hardware while still keeping data inside Switzerland; on-prem becomes more attractive once usage volume or specific customer requirements justify it.
Related guides
AI for SAP S/4HANA, Running On-Prem or in Your Private Cloud
Run AI on SAP S/4HANA without sending ERP data to a public API. On-prem and private-cloud architecture, CDS views, OData, and honest deployment trade-offs.
Infor LN + AIAI for Infor LN: Sessions, BODs, and Engineer-to-Order Work
Add grounded AI to Infor LN 10.x or CloudSuite: natural-language answers over sessions and BODs, agents for project and engineer-to-order work, on-prem options.
GDPR + AI on ERP dataBuilding GDPR-Compliant AI on Top of Your ERP
Design AI on ERP data that satisfies GDPR: lawful basis, DPIA, data minimisation, and an on-prem architecture that avoids the Schrems II transfer problem.
SAP + private AI in GermanyAI for SAP in the German Mittelstand: On-Prem and DSGVO-Ready
On-prem AI for SAP S/4HANA and Business One built for German Mittelstand manufacturers: DSGVO-aligned, Betriebsrat-ready, data stays on your own servers.
RAG + SQL + permissionsA Private LLM Grounded on Your ERP Data
How a private LLM answers questions on your ERP data: RAG plus text-to-SQL, role-based permissions inherited from the ERP, and where each fits.
Digital thread + AIAI Across PLM and ERP: Teamcenter, Windchill, and Your ERP
AI that reads Teamcenter or Windchill alongside your ERP to trace engineering change impact, BOM sync gaps, and where-used questions, on-prem.
Plan it with numbers
Sovereign AI Readiness Assessment
Score your organization across eleven dimensions of sovereign AI readiness, from data residency and model provenance to cleared personnel and air-gapped operations.
Free ToolOn-Prem LLM Total Cost of Ownership Calculator
Model the full multi-year cost of running LLMs on your own hardware, including GPU capex, power, cooling, support contracts, and operations staffing.
Free ToolERP Data Quality for AI Assessment
Score item master, customer, vendor, and transaction data quality to find out whether your ERP is ready to ground an AI copilot, forecast, or chatbot.
GuideERP GDPR Data Protection Compliance Guide
Achieve GDPR compliance in ERP systems with data mapping, consent management, right-to-erasure implementation, and data protection impact assessments.
GuideA Data Governance Framework for Private AI
A practical data governance framework for private AI: classification, access controls, retention, and audit patterns that satisfy CMMC 2.0 and ITAR.
GuideAI Strategy for the Mid-Market Manufacturer
AI strategy for the mid-market manufacturer: how $20M-$500M firms compete with AI without enterprise budgets, staff, or multi-year transformations.
Talk it through with an engineer who knows SAP S/4HANA
Bring one real question your team cannot answer from the ERP today. We will map the data path, the model, and where it runs, and tell you honestly if AI is the wrong tool for it.