The typical client
Swiss or Swiss-active insurers (CH, FL, possibly EU subsidiaries) with claims volumes from about 50,000 cases per year. You are evaluating or already operating an agentic triage system for first review, letter generation or subrogation checks — and face the question of how that squares with FINMA Circular 2018/3 (outsourcing) and the FADP.
Outsourcing topology per FINMA Circular 2018/3
Data and models stay in Switzerland or in the EU — never in a US-incorporated entity. Outsourcing to Dxzio is a technical outsourcing, not an outsourcing into a hyperscaler. We are a Swiss data processor, not a white-label wrapper.
Data cards per use case
Every pipeline ships a data card (DCAT format, machine-readable) as an audit artifact. No separate compliance pipeline — the card is part of the CI build and checked in the PR diff.
Model governance
Models that prepare claims decisions are versioned, signed and reproducible. A decision from 2026-04-22 can still be recomputed in 2031 with the model of that day, the same input and the same policy state.
Audit chain with Ed25519
Every agent action is signed, hash-chained and tamper-evident. We can hand FINMA, internal audit or the data protection officer a complete audit path for any single claim in seconds — without them needing access to our system.
Evidence from practice
We currently operate this pattern for a Swiss tier-1 insurer in the Liechtenstein business (about 140,000 claims per year). Architecture review submitted to FINMA and FMA Liechtenstein, no open findings. Go-live in Q4 2026. Reference call on request.
What we do not do
We are not a white-label wrapper around OpenAI or Vertex AI. We are not a Salesforce systems integrator. We do not deliver compliance consulting — we deliver architecture that makes the compliance argument easier.