How a European private bank consolidates AI under one governed deployment
A 120-seat European private bank consolidates fragmented SaaS AI tooling onto a single governed Operayde deployment — keeping client data under the bank's control and its model usage fully auditable.
- SaaS copilots consolidated
- 3 → 1
- Indicative AI spend reduction
- ~40%
- Time-to-enrol
- under 30 min
- Prompt data to third-party vendors
- none
Why this bank chose on-prem
This private bank runs 120 relationship managers across Luxembourg and Geneva. By late 2025 they were paying three different SaaS vendors for copiloting, summarisation, and KYC triage — each of which silently retained prompt data for up to 30 days, each with a different DPA, each routable to a different cloud region depending on the quarter.
The compliance team's audit concluded that "three separate transfer impact assessments, renewed annually, is not a sustainable posture for a regulated private bank." The engineering lead looked at two paths: build a self-hosted stack on their own Kubernetes cluster, or bring in an appliance.
What we deployed
A single Operayde Professional appliance — 2x L40S GPUs, installed in the Luxembourg DR room — serves all 120 seats. Every request, every document, every audit line stays on the bank's LAN. The central plane holds only metadata: which appliance is healthy, which policy bundle version is active, which operator signed off on yesterday's model rotation.
Federation to the bank's Entra ID tenant took one afternoon. The legacy SaaS copilots continued to run in parallel for two weeks so managers could A/B their own prompts; the switchover was unremarkable.
Outcomes at 60 days
Three SaaS contracts terminated with effect from the renewal boundary. Monthly AI spend dropped from roughly EUR 38k (SaaS) to EUR 22k (Operayde + amortised hardware) — a 41% reduction with the appliance paid off inside 19 months at the current usage curve.
No prompt data has left the bank's network since cutover. The compliance team replaced three transfer impact assessments with one internal DPIA.