PurposeThis study aims to investigate which variables can predict clients closing their checking accounts in a commercial bank and to validate the model used in this research.Design/methodology/approachThe theoretical basis of the study has contemplated satisfaction, loyalty, change mediators, switching costs and client retention. A total of 2,000 account holders plus 2,000 clients who closed their checking accounts were selected by simple probabilistic sample. The samples were probabilistic, without repetition, with an error margin of approximately 2.2 percent and trust interval of 95 percent with random drafting of the account holders' database from a large institution based in all Brazil. The variables were tested using the Binary Logistic Regression and the variables between the account holder and former account holder groups with indication of checking account closing request were compared.FindingsAccount holder's age, time of the account, investor/taker profile, internal relationship, long term assets contracts, risks in other banks, quantity of products, product canceling, average amount of entries and the existence of a joint client are the variables which better identify clients' propensity to end their relationship.Practical implicationsThe obtained results make it possible to identify the clients' propensity to abandonment and contribute to directing future marketing actions.Originality/valueThe model is relevant on the theoretical point of view because it can measure the importance of each independent variable and provides elements to compare all of them in the same base (meta‐analysis). It deals with 14 independent variables and had a good adjustment once was capable to classify correctly 89.8 percent of the account holders and 87.2 percent of the former account holders. The K‐S value was 79 percent with p<0.005 and value of ROC curve was 0.947. Models like this are unknown.