Industries / Government, education and community / Government / Emergency services / Ambulance and medical transport
Dispatch
Dispatch is deciding, for each call, whether an ambulance goes, how fast it must arrive, and where idle ambulances should wait for the next call.
5
Milestones achieved, from 1883 to 2017.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1883
A rescue station opens
On 8 December 1881 a fire in Vienna's Ringtheater killed at least 386 people; staff did not use the alarm line to the nearest fire station, and the fire brigade was called only after eleven minutes. The next day Jaromír Mundy, Hans Wilczek and Eduard Lamezan-Salins founded the Vienna Voluntary Rescue Society (Wiener Freiwillige Rettungsgesellschaft). It made its first official patient transport on 24 April 1882 in one of three ambulance wagons that the coachbuilder Lohner built to Mundy's plans, and on 1 May 1883, after Mundy had trained 30 medical students in carrying patients, it opened its first rescue station. The city took the service over in 1938; by 2003 it kept about 100 vehicles ready at all times, and it had twelve district stations.
Lessona disaster in which nobody was organized to rescue anyone produced a standing service with stations.
Credit
Jaromír Mundy, Hans Wilczek and Eduard Lamezan-Salins; Vienna Voluntary Rescue Society
1971
Stations placed by formula
Constantine Toregas, Ralph Swain, Charles ReVelle and Lawrence Bergman of Cornell University treated the siting of emergency stations as a "set covering problem": choose the fewest sites so that every neighbourhood lies within a set time or distance of at least one, and solve it with linear programming. It turned station placement into a calculation that planners could repeat. A 2026 review of 150 modelling studies of ambulance services found that placing and moving ambulances is still the most studied topic, with response time the main measure.
Credit
Constantine Toregas, Ralph Swain, Charles ReVelle and Lawrence Bergman
2005
Eight minutes questioned
Many US systems judged themselves by an 8-minute response standard, which was drawn from studies of cardiac arrest. In Charlotte, North Carolina, T. H. Blackwell and J. S. Kaufman found no difference in median response time between survivors and non-survivors among 5,424 urgent transports, but lower deaths when the crew arrived within 5 minutes (0.51% against 1.58%). In Denver, P. T. Pons and colleagues studied 9,559 patients from 1998 and found that arriving within 8 minutes was not linked to survival once illness severity was allowed for, while arriving within 4 minutes was, for patients at intermediate or high risk of dying.
Lessonone clock for every call does not track who survives; only a few conditions gain from minutes saved.
Credit
T. H. Blackwell and J. S. Kaufman, Carolinas Medical Center; P. T. Pons and colleagues, Denver Health
2007
Nurses answer before ambulances
Tokyo's Emergency Telephone Consultation Center, reached by dialling #7119, opened in 2007 as Japan's first telephone triage service: nurses tell callers whether they need an ambulance, should see a doctor within an hour, or can wait 6 or 24 hours. Other prefectures copied it. An independent analysis of public data from 1989 to 2022 found no clear fall in ambulance dispatches per 1,000 people in most of the 16 prefectures that had adopted it; Tokyo's own estimate was uncertain.
Lessonan advice line on its own did not stop the rise in ambulance use.
Credit
Tokyo's Emergency Telephone Consultation Center, staffed with the Tokyo Medical Association; Atsushi Sakurai and colleagues (evaluations); Y. Sekizawa (2026 analysis)
2017
England rewrites its targets
From October 2015 to April 2017 NHS England tested two changes in its ambulance services: giving call handlers more time, up to 240 seconds, to assess calls that were not plainly life-threatening before sending a vehicle, and sorting calls into four categories by clinical need. Category 1, life-threatening calls, must be reached in 7 minutes on average and 15 minutes for 9 calls in 10; Category 2, such as stroke, in 18 minutes on average and 40 minutes for 9 in 10. The independent evaluators wrote that the old goal, an 8-minute response for at least half of callers, was "simply not achievable". The new standards were in force across England by November 2017.
Credit
NHS England; Janette Turner, Jon Nicholl and colleagues, School of Health and Related Research, University of Sheffield (evaluation)
Watch for
AI steers calls to nurses
The Seattle Fire Department has used live prompts from Corti's AI since December 2023 to suggest which 911 medical calls dispatchers should pass to a nurse line run by its ambulance contractor instead of sending an ambulance at once. A 2024 Corti release quoted the department's medical director crediting the prompts with a 50% rise in calls sent to the nurse line; the department later told The Seattle Times the rise was 32%. The department says it has no public measures of how the AI affects callers.
Achieved whenan outside evaluation publishes how many callers the prompts sent to the nurse line, how long they waited, and how many later needed emergency care, against calls handled without prompts.
Claimed by
Seattle Fire Department and Corti, 2024, as reported by The Seattle Times in June 2026
Scottish calls teach a model
The Scottish Ambulance Service, which takes over 1.5 million calls a year, gave Corti copies of 999 call audio from April to August 2025 to train a model on how people in Scotland describe illness, with the AI switched off during live calls. Corti's model, the service says, was already trained on over 10 million emergency calls.
Achieved whenthe service publishes how often the model's severity judgment matched the final clinical outcome on Scottish calls, compared with call handlers, before any live use.
Claimed by
Scottish Ambulance Service, 2025
Open
01
A leaderboard for ambulance placement
A fixed set of test problems for deciding where ambulances wait and when to move them, built from open city models and open call records, with one scoring script and a public table that ranks published methods on the same problems. All runs use past or simulated calls, never live ones.
The briefCredit · unclaimed02
Map where ambulances wait
An open map of how many people live within 8, 15 and 30 minutes' drive of an ambulance station, country by country outside the United States, with a measured check of how complete the mapped stations are against the services' own station lists.
The briefCredit · unclaimed
Open 01 · the brief
A leaderboard for ambulance placement
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A fixed set of test problems for deciding where ambulances wait and when to move them, built from open city models and open call records, with one scoring script and a public table that ranks published methods on the same problems. All runs use past or simulated calls, never live ones.
Why it is still open
A 2026 review of 150 modelling studies found placement and relocation the most studied problem, but only "a limited but growing number" of authors share code and cooperation with ambulance services "remains limited", so methods are tested on different private data. The University of Auckland's open simulator JEMSS.jl ships models of Auckland, Edmonton, Manhattan and Utrecht but no shared results table. Other open simulators also stand alone: qambo, by Michael Allen, Kerry Pearn and Tom Monks (2021), is a training ground for reinforcement learning agents that place ambulances; AmbuSim is a simulator for moving ambulances between posts; and an online tool by Vincent Guigues and colleagues (2024) runs dispatch methods on Rio de Janeiro data and on data any service uploads. Zindi's 2020–21 Nairobi challenge scored only six virtual ambulances, by distance to past crashes. A November 2026 search of GitHub found dozens of one-off student and research projects and no common benchmark.
europepmc.orggithub.comarxiv.orggithub.comarxiv.orgzindi.africa
Built with what exists
JEMSS.jl; New York City's open EMS Incident Dispatch Data, which gives call times, severity and response seconds by zip code; open routing engines such as OSRM or openrouteservice; solvers such as HiGHS or Google OR-Tools for covering models.
First step
Run JEMSS's Auckland example and score two placement methods on the same simulated week.
Done looks like
At least 20 fixed problems from at least four cities, at least five published methods (from a static covering model to a learned relocation policy) scored on mean and 90th-percentile response time and on moves per shift, with all code and the results table public.
Open 02 · the brief
Map where ambulances wait
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open map of how many people live within 8, 15 and 30 minutes' drive of an ambulance station, country by country outside the United States, with a measured check of how complete the mapped stations are against the services' own station lists.
Why it is still open
The United States has been mapped: the Maine Rural Health Research Center found that in 41 states in 2021–2022, 4.5 million people lived more than 25 minutes' drive from the nearest ambulance station, in what it calls ambulance deserts. Elsewhere, the open maps of access to care start from hospitals, not from the stations where crews wait. A 2026 preprint by Johan Emilsson routes drive times from all 217,841 hospitals in OpenStreetMap, and the Heidelberg Institute for Geoinformation Technology's Open Healthcare Access Map does the same for health facilities. OpenStreetMap holds 16,454 objects tagged as ambulance stations (taginfo, 19 November 2026), with completeness unknown. A November 2026 search found no multi-country station coverage map.
Built with what exists
OpenStreetMap ambulance station data; openrouteservice or OSRM for drive-time areas; the GHS-POP or WorldPop population grids; station lists published by national or regional ambulance services.
First step
Download the mapped stations for one country that publishes its own station list, compare the two, and compute 15-minute coverage.
Done looks like
Coverage at 8, 15 and 30 minutes for at least 10 countries other than the US, mapped stations compared with official lists in at least 5 of them (share found, share missing, share wrongly placed), all published as open data.