Emergency Medical Services (EMS) system's mission is to provide timely and effective treatment to anyone in need of urgent medical care throughout their jurisdiction. The main goal of most EMS deployment is to reduce mortality, disability, pain and suffering. There are several metrics for level of EMS service, and among them, response time (RT) and call coverage rate are the most popular ones used by EMS providers and researchers. Ability to provide timely response is affected by fleet size and the locations of the ambulances. Hence, literature on ambulance location has been dominated by models which generally maximize or guarantee coverage, minimize mean response time, and alike. Essentially all models, including highly sophisticated queuing embedded optimization models, rely on several simplifying assumptions in order to make them tractable. These include the vehicle busy probabilities calculated a priori, dispatching the nearest ambulance to all incidents, a zone (call demand) being covered (can be reached) if it is within the distance/time threshold as a binary exogenous variable, static unit dispatch, and so on. The default dispatch policy is to send the nearest ambulance to all medical emergencies and it is widely accepted by many EMS providers. However, sending nearest ambulance is not always optimal, often imposes heavy workloads on ambulance crews posted in high demand zones while reducing available coverage or requiring ambulance relocations to ensure high demand zones are covered adequately. In this study we propose a simulation embedded optimization approach for relocating ambulances and determining flexible dispatch policies that balance ambulance crew workloads while meeting fast response times for life-threatening calls. A realistic simulation model allows us to remove most of the simplifying assumptions which are required in analytical approaches such as integer programming models as well as queuing theory based models. We show that this approach provides a much richer output that can be used by EMS managers to estimate lives saved for multiple life threatening situations while providing a detailed statistics on important performance measures such as actual ambulance workloads and response times. We validate our approach with an advanced coverage optimization model using real-life data. We present computational statistics and demonstrate the efficacy of a tiered dispatch policy using real-world data.