In this paper, we describe a method to compute multiple improvement scenarios for Multi-Agent Simulation using parameter search by metaheuristic. Specifically, we try to construct multiple scenarios to acquire more customers by using modeling of taxi driver’s behavior from probe car data. The main proposal of this paper is not to find an optimal solution but an approach to lead multiple suboptimal solutions. We construct a taxi agent model that can represent individual traffic behaviors and apply the model to urban traffic multi-agent simulation where general vehicles, bus, and taxis co-exist in the realistic road network of the city of Kyoto, Japan. Then, we try to construct multiple improvement scenarios concerning the sales strategy by multipoint local search method and clustering of parameters. The proposed method works usefully in the problem that the solution property with wide solution space is unknown. In addition, this method has scalability, and it is possible to obtain a better optimum solution by increasing the number of cluster calculators and increasing the number of searching points.