Improved techniques for proactive identification of errors are provided. Documents are analyzed, using natural language processing (NLP) models, to identify entities in the documents. The documents are processed with the NLP models to identify relationships among the entities, and each of the entities is stored in a data store, where the data store further includes an indication of the identified relationships among the entities. A draft correspondence is received, and it is determined that the draft correspondence includes a first entity and a second entity, where the first entity and the second entity have a first relationship. Upon determining that the first relationship conflicts with at least one of the relationships stored in the data store, a suggested revision for the draft correspondence is generated.