Before go-live¶
Local clinical validation completed on a sample that represents your own patients
Local performance meets the minimum thresholds you set in advance
Technical integration tested end to end in the production environment
All intended users trained, in line with the AI literacy duty (EU AI Act Art. 4)
Governance structure set up: roles, responsibilities and escalation routes written down
Regulatory documentation complete and up to date (CE marking, or in-house exemption under MDR Art. 5(5))
Incident reporting route set up and communicated to staff
Rollback plan written down (how to switch the AI off and what staff do instead)
Ongoing (after go-live)¶
Performance monitoring dashboard running
Alert thresholds set and responses agreed
Logging running, logs kept for the required period and audit trail accessible
Periodic audits scheduled (for example once or twice a year)
Way for users to give feedback in place
Post-market surveillance plan in place (for medical devices)
Plan for model updates and retraining reviewed