Most business applications mutate relational data. Online inference is often done on this mutable data, so training data should reflect the state at the prediction's "point in time" for each object. There are a number of data architecture / domain modeling patterns which solve this issue, but they only work from implementation date onwards.
In this talk we'll suggest how to use the "point in time" as a first-class citizen in your ML Platform, while still striving to maximize the use of your older messier data.