Simulation can be used for analysis, prediction, and optimization of business processes. Nevertheless, process models often differ from reality. Data mining techniques can be used for improving these models based on observations of process and resource behavior from detailed event logs. More accurate process models can be used not only for analysis and optimization, but for prediction and recommendation as well. This paper analyses process model in manufacturing company and its historical performance data. Based on that observation, simulation model is automatically created and used for analysis, prediction and for dynamic optimization.