Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining gratitude scores for a plurality of people who have interacted with an organization. In one aspect, a method comprises: obtaining, for each of a plurality of people who have interacted with an organization, history data characterizing previous interactions of the person with the organization and third party data characterizing aspects of the person outside of their previous interactions with the organization; and processing, for each of the plurality of people, the history data and the third party data for the person using a gratitude prediction machine learning model to generate a gratitude score for the person, wherein the gratitude score for the person characterizes a likelihood that the person will take a specified action on behalf of the organization in the future.