Because of the self-selection nature of a loyalty program, we have never been able to measure ROI by looking at value differences between members and non-members. Control groups and looking at the performance difference pre- and post-enrollment cohorts (i.e. Amazon) is valid, but unfortunately impractical for many (most?) retailers.
We have found valid ROI measurements of reward redemption in creating incremental visits and increased lifetime value (profits). Customer churn and margin metrics can also be attributed to the program. Softer metrics like engagement and attitudinal feedback help as well.
Definitely agree that the understanding the measurement parameters prior to launch, developing a logical ROI pro forma, creating the infrastructure, and consistently following it is the right approach.