Currently, companies seek to increase customer satisfaction by offering new interactions with their clients. Among different interactions, retail delivery services, like home delivery or door-to-door delivery, have gained importance in recent years, being now a competitive advantage. However, this interaction is expensive because of the distances that must be traveled, and it could incur in low service levels due to the large number of orders that might not be covered completely. The use of mobile warehouses constitutes a good approach to solve retail delivery logistics problems. Mobile warehouses are vehicles that drive around the city, perform the delivery to the final customer and manage the administrative tasks such as billing and collecting money. This delivery strategy requires to know the number of products that the vehicle should carry in order to satisfy the demand without returning to the distribution center. This study presents a methodology for demand forecasting and inventory management in a mobile warehouse. Different forecasting methods were used, but artificial neural networks showed the best performance to forecast demand using mobile warehouses. Once the demand forecast is determined, a linear programing optimization model is proposed for inventory management with the aim to reduce the total cost of stock out occurrences in terms of traveling distance to the distribution center, time expended, and lost cost sale. A case study from an Ecuadorian company in Quito is presented, in which the number of products that the vehicle should carry is determined minimizing the stock out occurrences.