Well, it's a work in progress, we've decided to start with KPI trees at our squad level since they all take care of a slice of the overall engineering workflow.
What I've noticed is there are two ways to go about it
1. find one north star metric and break that down (think mean time to recovery).
2. if one metric is difficult, think about the different dimensions that are important for the squad. For instance, for the squad that takes care of the Kubernetes infrastructure, we're thinking about performance, security, usability as some dimensions that are important, and we're trying to break those down into smaller units of things