How would you design an experiment if you cannot randomly assign users to the control group at Betterup?
