During a military career, frequently exercised skills appreciate into expertise, while infrequently exercised skills can decay. Decay can be caused by inattention to the skill, which in turn can be caused by infrequent tracking. Although trainee skill states are systematically measured and monitored during formal training (e.g., school house, Initial Qualification Training (IQT), and Mission Qualification Training (MQT)), once trainees are qualified and assigned to operational missions, assessment is less frequent. Training sustainment programs intended to maintain skill proficiency (e.g., Continuation Training (CT)) only require that tasks be accomplished without systematically measuring, storing, or analyzing skill proficiency data. Thus, the problem this paper addresses is that trainee data is not sufficient to determine the nature and magnitude of the skill decay, making it difficult to know the true skill state of military operators at any given time. Fortunately, military operational databases are filled with information related to missions executed, tasks accomplished, tools/platforms used, etc., and can be a rich source of data from which operator skill states can be inferred. In this work, we describe a suite of machine learning data mining algorithms that operate not only on training data stored in Learning Management Systems (LMSs), but also on operational databases, to make inferences about operator skills states that can be used to personalize learning to ensure that only deficient skills are trained. This innovative approach to leverage operational mission data will allow keen insights into operational learning, or the learning that occurs when formal training ends.