For example, if we decide to run an experiment on invitations, we're aiming to validate an assumption that semi-automated invites will support players inviting their friends. The metrics we establish for this impact define how many invitations we expect, over what period etc. We build the feature, measure the results, compare them to expectations and learn from this to decide on the next feature. We apply this to the larger picture as well: improve invitations, measure the rate at which invited friends are becoming new players, learn from this and decide where to invest next.
