Media, methods, and systems are disclosed for optimizing performance of a running application in connection with a group-based communication system. Log data is collected regarding prior metrics for applications that have encountered performance events. Application state information is monitored and a machine-learning model mapping application metrics to performance outcomes predicts whether the running application will encounter a performance event. The machine-learning model mapping application metrics to performance outcomes is trained based on the collected logs. Based on whether a degradation outcome will be impactful, an application performance parameter may be degraded.