Abstract: Techniques are described for determining a value metric for each of a plurality of payment cards, and recommending payment card(s) based on their respective value metrics. Implementations provide a recommendation engine that analyzes user data and card data, and recommends payment card(s) that may be suitable for a user. The engine consumes usage data and credit score and uses that information, along with characteristics of various available cards, to develop a value metric for each payment card that indicates how valuable owning the card would be to the particular user, such as how much money the user can expect to make or lose every year if they use the card as indicated, given the card's interest rate, rewards, fees, and so forth. The engine can rank the cards according to their value metrics, and the top-ranked card(s) are presented to the user as recommendations in a user interface.