Example implementations may relate to a cloud service that stores a detection metric corresponding to a maintenance request for a particular component. In particular, the cloud may receive sensor data from various robotic systems each having the particular component. The cloud may then determine, based on the sensor data, performance data for the particular component over time at the various robotic systems. The cloud may also determine various maintenance events for the particular component. Based on the performance data, the cloud may determine that at least one maintenance event occurs at other metrics that are different from the detection metric. Responsively, the cloud may adjust the detection metric based on a difference between the detection metric and the other metrics. The cloud may then detect operation of a particular robotic system at the adjusted detection metric and may responsively request maintenance for the particular component at the particular robotic system.