In some examples, a computing device may predict, using a machine learning module, scenarios and transactions associated with a usage of a software package. The computing device may select at least a portion of the scenarios and the transactions to cover a predetermined percentage of a predicted usage of the software package. The computing device may select a subset of unit test cases (e.g., created by software designers to test software units that are components of the software package) and execute the test cases to generate test results to determine whether the software package is ready to be deployed to customers. The computing device may train the machine learning module using at least one of the test results, the portion of the scenarios and the transactions, or the test cases. The test results may be evaluated to determine an effectiveness of the set of test cases.