■ SAFE RISK ■ Public Service & Government
Not replaced, but restructured. Machines are already triaging non-emergency traffic and listening in on live calls, while the human keeps the part where someone is screaming and a decision must be made anyway.
“AI routes calls. But calming a panicking caller while dispatching help? That's human multitasking at its finest.”
Our AI replacement risk score — how we score jobs
A dispatcher runs several channels at once. They answer a call, work a structured protocol to classify the emergency, extract a location from someone who does not know where they are, and start units rolling before the interrogation is finished. Simultaneously they are typing into a CAD system, monitoring radio traffic from responding crews, updating the incident record, and giving pre-arrival instructions — CPR compressions counted aloud, choking manoeuvres, childbirth coaching — to a caller in the worst minutes of their life. Shifts are twelve hours, the queue never empties, and the job's defining feature is holding composure while a stranger's crisis plays out in your ear.
Automation has moved in fast. Non-emergency lines are increasingly handled by conversational systems that take reports and route requests. Real-time speech analysis assists cardiac-arrest recognition, catching arrests callers do not identify. Automated location technology pulls precise handset coordinates. Text-to-911 triage, translation for non-English callers, transcription, CAD-integrated call classification and automated resource recommendation are all live in some centres. Some agencies use models to prioritise the queue during surges.
The residual is stubborn. Callers lie, panic, whisper, or speak while being assaulted; classification often depends on background sounds and inference rather than the words. Overriding a protocol because something is wrong, coaching a bystander through compressions while they sob, and carrying legal responsibility for the dispatch decision all stay human. Our score of 7 says the console changes and the seat does not empty — though centres will run leaner per call. The union and liability picture matters too. Emergency communications is a regulated public function with certification requirements, recorded lines and litigation exposure, so agencies adopt assist tools far faster than they adopt anything that removes a certified human from the loop on a life-safety call.
Automatability: our editorial assessment of current and near-term AI capability
Serious change already underway, but not elimination. Conversational systems for non-emergency lines and AI-assisted call analysis are deploying now, and by around 2030 most large centres will run a hybrid console where software handles routine traffic and flags high-acuity signals. Chronic dispatcher shortages and heavy turnover mean these tools mostly absorb overflow. Expect the role to become more supervisory and higher-acuity rather than to disappear.
For non-emergency lines, yes — several jurisdictions route administrative and low-priority calls to conversational systems to free human dispatchers for emergencies. On live emergency calls, AI runs in the background: speech analysis assisting cardiac-arrest recognition, automatic transcription, translation and location retrieval. The human still owns the call. Full automated handling of a genuine emergency is not deployed anywhere serious.
Headcount pressure is real but starts from a deficit. Emergency communications centres are chronically understaffed with brutal turnover, so the first effect of automation is filling gaps rather than cutting posts. Over time expect fewer dispatchers per call volume, with the remaining roles skewed toward high-acuity calls, supervision and quality assurance. That is compression, not extinction.
The information is unreliable by nature. Callers panic, whisper, give wrong addresses, minimise, or cannot speak at all; dispatchers infer from breathing, background noise and hesitation. Pre-arrival instructions require coaching a terrified bystander through a physical procedure in real time. And someone must be legally accountable for the dispatch decision when it goes wrong, which agencies are unwilling to assign to software.
Become the person who handles what the system escalates. Get emergency medical dispatch and crisis-intervention certifications, learn the analytics and CAD tooling well enough to train others, and move toward QA, supervision or centre technology roles. Also stay sceptical of the assist tools — recognising when the automated classification is wrong is becoming the core skill of the job.