Dental: routine cleaning request vs uncontrolled bleeding (page on-call dentist). HVAC: tune-up booking vs gas smell or no-heat-in-January (page on-call tech). Medical: prescription refill vs chest pain or stroke symptoms (911 + page on-call provider). Law: schedule consultation vs domestic violence with immediate danger (transfer to local emergency services + page attorney). MSP: user can't print vs entire network down (page on-call engineer). The framework is the same: AI handles routine, pages humans for emergencies, never tries to autonomously handle anything life/safety-critical.
When AI determines an emergency: (1) immediately reassures the caller with a clear escalation script ('I'm connecting you to the on-call tech right now — please stay on the line'), (2) initiates a parallel call to the designated cell with caller context already pulled ('emergency at [address], gas smell reported, customer is [name]'), (3) bridges the calls in under 60 seconds OR keeps caller on the line until human picks up. Time-to-human for true emergencies: under 90 seconds.
Bad pattern: AI tries to triage too narrowly, waits for keyword match, escalates only for explicit 'emergency' word — misses people who say 'I think something might be wrong.' Bad pattern 2: AI escalates everything, paging the on-call tech at 2am for routine tune-up requests, leading to learned-helplessness where the tech ignores pages. The correct configuration is calibrated thresholds tuned to YOUR specific business's emergency types.
Some vertical insurers (medical malpractice, professional liability) have specific requirements about AI handling of patient calls. Best practice: AI is explicit it's a 'scheduling assistant,' doesn't diagnose, doesn't provide medical advice, doesn't form provider-patient relationships, escalates anything beyond scheduling to the human. Properly configured AI is compliant with most professional liability requirements — your insurance carrier should review the configuration before launch.
Setup includes emergency-flow design + live testing. Cost: included in standard $797/mo Pilot tier — not an add-on. Always test emergency flow yourself before launch: call your own AI from your cell, say something that should trigger emergency, verify your phone rings within 90 seconds. Hear our emergency flow: +1 (877) 640-3761 — ask about a 'gas leak' and listen for what happens. Free audit walks through your specific emergency configuration.