The article presents a method that combines simulation and metamodels to predict outcomes faster and more accurately. By updating the metamodel with new simulation data, the prediction accuracy for new inputs can be improved. This approach is particularly useful for web-based simulations where quick response times are important. The study focuses on stochastic kriging metamodels and shows that updating them with new data is similar to a Bayesian forecast combination. Using metamodel predictions of variance instead of point estimates can enhance metamodel performance.