Credit scoring is an important part in controlling risk in financial companies. With the high number of nonperforming loans, the assessment of potential new customers in financial companies has become a major focus of the financial industry. High accuracy credit scoring system can give better predictions on new customers and can change the company's economic growth and for better capital. This study uses a real world dataset, where data is obtained directly from a financial company and will be used to feed a random forest model to differentiate between good and bad potential new customers. This contribution of this research is to improve single model with real dataset by using ensembled bagging, bootstrap aggregating. Two methods are implemented, random forest and neural network to see the performance of ensembled using bootstrap aggerating. The output accuracy from the final model of the ensemble methods that resulted from the voting will be compared with the original unmodified single model and another model with similar architecture.The result shows that the modified multiple model surpasses the unmodified single model in terms of accuracy with a tradeoff on duration in the process.