A transaction parameter date set comprising input variables characterizing the transaction conditions are collected when an electronic payment is initiated on an electronic payment system. The collected transaction input variables are used to calculate an expected future transaction value using a decision function. The expected future transaction value identifies high utility transaction based on a comparison of the expected future transaction value to a threshold value established while training the decision function on a training data set of known outcome. A high utility transaction can include behavior that if rewarded is likely to retain current users or grow the electronic payment system network by adding new users. If the expected future transaction value exceeds the threshold value the sender of the electronic payment is rewarded by the system.