Generating and applying a prediction model based on blockchain data is disclosed. Based on blockchain data related to historical invoices, a prediction model is generated using machine learning. The prediction model is used to generate a recommendation for a particular entity to obtain a target funding amount by a target funding date. The prediction model is applied to the particular entity's outstanding invoices. Based on blockchain data related to the outstanding invoices, the prediction model determines a predicted cost that the particular entity would incur in obtaining funding, by the target funding date, using each outstanding invoice. A funding type to be used for each invoice is also identified. Invoices are selected based on the predicted costs. An interface presents the recommendation to use the selected invoices to obtain the target funding amount by the target funding date.