According to recent theories of financial intermediation, one of a bank’s main roles is to serve as a relationship lender.1 Relationships offer banks comparative advantages in lending through the accumulation of private information, which can arise from the length of the relationship over time and its breadth across multiple products. In turn, relationships benefit bank customers through increased credit availability, such as greater amounts of and lower prices for credit (Boot and Thakor 1994).2 Empirical studies of the benefits of relationship banking have largely focused on the benefits to customers, and corporate customers in particular.3 Only limited empirical research has been conducted on the benefits of relationships to banks. In his review of relationship banking literature, Boot (2000) concludes that ―existing empirical work is virtually silent on identifying the precise sources of value in relationship banking.‖ Mester, Nakamura, and Renault (2007) study a sample of 100 Canadian smallbusiness borrowers, and find that information about customers’ collateral, and their inventory and accounts receivable in particular—which might not be available to banks outside of a relationship—is useful for loan monitoring. Also, changes in transaction account balances are informative about changes in this collateral. In subsequent work, Puri, Steffen, and Sascha (2017) provide external validation to our work and show that retail customers in Germany who have a relationship with their savings bank prior to applying for a loan default significantly less often than customers with no prior relationship. This paper studies the economic benefits of relationship banking to banks in the context of retail banking. 4 Credit cards provide a good setting for analyzing retail relationship banking, since it is easier to identify the information actually used by credit card issuers to manage their accounts. Specifically, our paper examines the implications of bank relationships for key aspects of credit card behavior, such as default, attrition, and utilization rates. 5 We use a unique, representative dataset of about 100,000 credit card accounts that are linked to information about other relationships account holders have with the bank that issued the card. Our dataset includes ―public‖ credit bureau information that is available to all potential lenders, lenders’ ―private‖ within-account (as opposed to cross-account) information about past behavior of the accounts at issue, and lenders’ ―private‖ cross-account relationship information. Given the information used by banks to manage their accounts, this paper can more cleanly test whether additional information—in this case, relationship information—provides additional predictive power.