In predictive analytics and machine learning, the term 'concept drift' (or 'data drift') means that the properties of the target variables (i.e. the themes and concepts underlying each of the labels in Re:infer), that the model is trying to predict, change over time in unforeseen ways.
Essentially, more recent data coming into the dataset will, over time, become increasingly different to the original data on which the model was trained.
This causes problems because the predictions become less accurate as time passes and the variables that the model is trying to predict become increasingly different to the training data.
Concept drift is one of the key reasons why it's important to properly maintain Models used in production use-cases, e.g. automations, by doing a small amount of exception training on a scheduled basis.