A method, system, and/or computer program product predicts trouble during a transition phase of a current project. Past operational data and past trouble data for past healthy projects and past troubled projects before transition phases is collected. A trouble correlation between the past operational data and the past trouble data is then determined. A set of key metrics that describes a current health of a current project before transitioning from a current phase to a future phase is defined, and values of the set of key metrics for the current project are identified. Identified values of the set of key metrics for the current project are compared to the past operational data for past healthy and troubled projects in order to generate a pattern comparison. This pattern comparison, along with the trouble correlation, is used to generate a predicted quality of transition for the current project.