What an AIR engagement looks like in practice

Situation

A 2,000-clinician health system rolled out ambient AI documentation across its hospitals. Within a year, clinician schedules and staffing assumed AI-assisted note completion. Nobody had classified the service in the continuity plan.

Engagement

• AIR Assessment inventoried 23 AI services; the documentation platform scored ACI 5 with an unstated AI-RTO of 'never down'
• Business impact analysis with clinical leadership set a realistic AI-RTO of 2 hours and defined a paper-plus-catch-up degraded mode
• Design phase specified secondary-provider failover for drafting and site-local dictation capture that queues during outages

Outcome

First quarterly recovery drill simulated a full vendor outage during a Monday clinic block. Failover completed inside the 2-hour AI-RTO, and the health system's continuity plan now covers every ACI 4+ AI service.

Situation

Facing an operational resilience examination, a bank realized its important business services relied on AI tooling — client service assistants and compliance surveillance — that appeared nowhere in its third-party risk or resilience documentation.

Engagement

• Assessment produced an AI Service Inventory with vendor concentration analysis: 84% of AI-assisted workflows traced to a single provider
• AI dependencies were mapped into the bank's existing important-business-service documentation, with AI-RTO/RPO targets aligned to regulatory tolerances
• Active-active model routing was deployed for the two highest-criticality services, eliminating the single-provider dependency

Outcome

The bank entered its review with mapped AI dependencies, documented recovery objectives, and drill evidence — converting an expected finding into a demonstrated capability.

Situation

A B2B software company embedded AI features into its product on a single provider's API. Two provider incidents in one quarter produced customer-visible degradation and an uncomfortable pattern: their SLA was effectively their vendor's SLA.

Engagement

• Assessment separated the estate into product-embedded AI (ACI 5) and internal productivity AI (ACI 2–3), with distinct strategies for each
• Product AI moved to active-active routing across two providers, with an evaluation suite validating output quality on both model families
• Internal copilot dependencies received a documented degraded mode instead of infrastructure spend — criticality didn't justify more

Outcome

The next provider incident shifted routing weights automatically with no customer-visible impact. Engineering now runs quarterly failover drills as part of normal reliability practice.