Ambulance Fleet Optimizer

Early EMS planning simply divided a city into fixed response districts, one ambulance per district, regardless of how demand actually clustered in space or time. This "one unit, one zone" approach — inherited from fire-department box-alarm thinking — wastes capacity: districts with low call volume tie up a unit that a busier neighboring district needs, and a district's sole ambulance is unavailable the moment it is already on a call. Operations researchers reframed the problem as a location-covering problem in the 1970s. The Location Set Covering Problem (LSCP, Toregas et al. 1971) asked: what is the minimum number of ambulances, and where should they sit, so that every demand point is within a target response distance of at least one unit? LSCP is elegant but often demands unrealistic fleet sizes, because it insists on covering the very last, hardest-to-reach demand point. The Maximal Covering Location Problem (MCLP, Church & ReVelle, 1974) relaxed this: given a fixed, budget-constrained number of facilities, choose their locations to maximize the population or call volume covered within the time standard, accepting that some low-demand areas may fall outside range. This reframing — optimize expected coverage subject to a resource constraint, rather than demand perfect coverage at any cost — became the template for virtually all subsequent EMS location models. MCLP's key weakness is that it assumes a covering ambulance is always available. In reality, ambulances are frequently busy on other calls, so "covered" does not mean "will respond in time." The Maximum Expected Coverage Location Model (MEXCLP, Daskin, 1981) fixed this by introducing a fleet-wide "busy fraction" q — the probability that any given ambulance is unavailable at a random moment — and computing the probability that at least one of several overlapping ambulances near a demand point is free to respond. Placing units so that high-demand areas are covered by multiple overlapping ambulances (redundant coverage) directly raises the expected fraction of calls reached in time, which is what patients and dispatchers actually experience. MEXCLP and its later stochastic descendants (MALP, AMEXCLP, dynamic dispatch simulation models) remain the conceptual backbone of most modern ambulance location software.

System Status Management (SSM), pioneered by Jack Stout in the 1980s, is the operational practice built on top of coverage models: instead of assigning ambulances to permanent stations, a fleet is treated as a pool of mobile resources continuously redeployed among a larger set of "staging posts" — street corners, parking lots, or hospital bays — based on which posts are predicted to best cover the next hour's expected call pattern. A dispatch center runs a "move-up" or "status plan": a pre-computed table (often dozens of plans, one per hour-of-week or per number-of-units-available) specifying where each idle unit should sit given the current time and how many units remain in service. When a unit goes out of service to answer a call, the plan may direct a *different* idle unit to move up into the gap that unit's departure created — filling coverage holes proactively rather than waiting for the next call to expose them. • Demand forecasting inputs: multi-year historical call logs binned by hour-of-day, day-of-week, and geographic cell (often hexagonal or 1 km² grid cells), sometimes refined with weather, local events, and epidemiological signals (e.g., flu season, holidays). • Posting logic: posts are chosen to maximize expected coverage (MEXCLP-style) against the forecast demand surface for that specific time window — a downtown financial-district post might be prioritized 8am–6pm on weekdays, and a nightlife-district post prioritized Friday/Saturday 10pm–3am. • Real-time trigger: computer-aided dispatch (CAD) systems continuously track each unit's live GPS position and status (available, en route, on-scene, transporting, out-of-service) and automatically prompt or auto-dispatch move-ups whenever the number or geographic distribution of available units drifts from the plan. Critics of aggressive SSM note the human cost: crews can spend long stretches of a shift parked in a truck at a street-corner post rather than a station, which is associated with higher reported fatigue and lower job satisfaction — a real operational tradeoff against the coverage gains.

Modern computer-aided dispatch does more than log addresses — it continuously re-solves a small assignment problem: given all currently available units' live GPS positions, expected travel times (from historical road-speed models, not straight-line distance), and each new call's priority and location, which unit should be sent to minimize expected response time system-wide, not just for this one call? Sending the geographically nearest unit is not always optimal if doing so strips coverage from a zone about to generate its own call — so advanced CAD/AVL (automatic vehicle location) systems increasingly incorporate look-ahead heuristics from the same MEXCLP family used for static planning. Performance is measured against fractile response-time standards. The classic benchmark popularized by many large US municipal EMS contracts is "90% of Code 3 (lights-and-sirens, life-threatening) calls answered within 8 minutes 59 seconds," often shortened to the "9-minute standard," with a secondary, looser target for lower-acuity Code 2 calls. NFPA 1710, the National Fire Protection Association standard governing career fire-department EMS, specifies a similar cascade: 60 seconds turnout time and a 4-minute travel-time goal for the first-arriving unit (90th percentile), plus 8 minutes for a full ALS assemblage — numbers that assume adequate unit density and are frequently missed in practice. In the field, compliance with these fractile targets typically runs 80–95% in well-resourced systems on an average day, but degrades sharply during system-wide surges — mass-casualty events, severe weather, or simply an unlucky clustering of simultaneous calls — when every nearby unit is already committed and the dispatcher must pull from progressively farther away. This is precisely the scenario dynamic repositioning is designed to buffer against: keeping a probabilistic reserve of coverage distributed across the service area rather than concentrated at a handful of stations.

Dense urban systems can hold a large share of their fleet at a handful of central posts because travel distances are short and road networks are redundant — a unit two kilometers away is often only three or four minutes out even in traffic, so modest repositioning yields large coverage gains. Rural and "frontier" EMS systems face the opposite geometry: a single ambulance may be responsible for hundreds or thousands of square kilometers, home-to-scene distances routinely exceed 15–30 minutes, and there may be no second unit within a practical mutual-aid radius at all. In these systems, positioning optimization is less about minute-level repositioning and more about long-range strategic siting of a small number of stations, first-responder co-location (fire, sheriff, or volunteer first-responder programs that can reach a patient minutes before the ambulance), and formal mutual-aid agreements with neighboring agencies. Across both settings, the marginal value of adding one more ambulance to a system follows a classic diminishing-returns curve. The first few units added to an under-resourced area close large coverage gaps; each additional unit beyond that increasingly overlaps existing coverage, because MEXCLP-style probabilistic coverage saturates — a demand point already covered by three or four overlapping ambulances gains very little expected-coverage improvement from a fifth. Because ambulances are expensive to staff around the clock (24/7 crewing of a single unit typically costs several times the vehicle's purchase price annually in labor alone), most systems find that *reallocating* existing units according to a good SSM/MEXCLP plan produces a larger response-time improvement per dollar than purchasing additional trucks once a baseline fleet size is reached. This is the central cost/coverage tradeoff the simulation above lets you explore directly: watch how enabling dynamic repositioning changes the response-time and coverage metrics at a fixed fleet size — often rivaling the effect of adding two or three more static-post units.