Abstract: Techniques for detecting software regression in computing systems are disclosed. One example technique includes fitting data in a dataset into a multiple variable model to obtain a set of estimated coefficients. The dataset having data representing multiple entries each containing an identification of a combination of multiple payloads included in a build of a software product and a corresponding value of a performance metric of executing the build at a computing device. The payloads individually represent a source code change, a feature enablement, or a configuration modification of the software product. The estimated coefficients individually correspond to one of the payloads. The method further includes in response to determining that a corresponding estimated coefficient of one of the payloads has an absolute value that is greater than a preset threshold, indicating that a software defect is likely present in the corresponding payload.