A method, apparatus and computer program product group records based upon a prediction of the content of the records. In the context of a method, data associated with respective subjects of the records is received and a threshold of a machine learning model is adjusted to satisfy an accuracy requirement for record categorization. In response to analyzing the data, but not the records, by the machine learning model, the method separates, using the machine learning model, the records into the first and second groups with the first group including records that the associated data indicates are more likely to support the addition of a code and the second group including records that the associated data indicates are less likely to support the addition of a code. The method also includes subsequently processing the records in different manners depending upon whether the records are included in the first or second group.