In the race to save lives, every second counts for firefighters. Indeed, chances to rescue people trapped in a fire are optimal if the response time is below 5 minutes. The ability to deploy units to the scene of an incident has always been a critical measure of performance for fire departments. Hence, everywhere in the world, fire services have conducted several analysis of what could impact their responses and seek strategies to decrease their response times. With the opportunities that big data and artificial intelligence are bringing in different areas, firefighting will not be spared by this revolution. Imagine a manager of the fire service who could leverage the data to optimize the response times of the department. Every morning, a software would inform him about what areas are at risk of fires and need to be inspected and what are the estimated response times for different areas of the city. Depending on the weather conditions and traffic, the predictions vary in real-time showing clear information on a map. Thanks to the new tools brought by Data Science, this scenario is now realistic. Although the predictions are not always prefect, the way managers of fire departments will take decisions will be surely impacted. In this thesis, using data from the fire department of Montreal, I present a complete framework to predict the response times and discuss what the challenges are for bringing Data Science into the management of fire services.Firefighters usually divide the response times of their soldiers in 2 main parts: the turnout time, which corresponds to the seconds elapsed while the firefighters prepare themselves in the fire base station, and the travel time, which refers to the time taken by the vehicle to arrive at the location of the incident. Several studies indicate that the time of the day, the type of incidents and the station layout impact the turnout times but use mainly basic descriptive statistics methods. Previous research demonstrates that the travel time of the firefighter is affected by the time of the day and the location of the stations, although here, tools to predict it exist and are documented in the literature. In the first part of this thesis, using raw data from the Fire Services of Montreal (Service des Incendies de Montréal, SIM), I use a Data Science pipeline, which involves cleaning, feature engineering and algorithm tuning, to predict the turnout time and the response time. Interestingly, predictions present a significant improvement over baseline models and existing solutions used by the fire department.Then, I discuss the possible usage for this prediction engine. Based on existing documentations and personal observations about the management of the fire department, I show that this framework could be ideal for the strategic planning of the response. The versatility and modularity of this tool could help to build a complete simulation engine of events happening in the city, which could benefit the fire department.Finally, because fire departments are already well organized structure, I show that bringing new tools that modify the information system chain and the decision-making process is not an easy task. The city of Atlanta has recently developed system called Firebird to predict the buildings at risk of catching fires using a complete Data Science pipeline. The development of the tool revealed that the main challenge was to collect and organize the data from the different actors. Implementing Data Science solutions requires large amount of data to be able to depict a complete picture of the reality. However, the quality of this data and its storage are still the main issues today. I show that having an external government expertise to manage the information is a good option as it is the most central actor between the fire department and other public services or external companies. In this organizational and policy part, I also insist on the training and support that managers need to receive so that they could transition to tools using artificial intelligence. In conclusion, because several fire departments are now beginning to better exploit their data, I show that the prediction framework presented here could be used by other services and that the city of Montreal would largely benefit from a collaboration.