Definition
A method of controlled experimentation to compare the performance of two different versions (A and B) of a model or system in a live production environment.
In short: A method of controlled experimentation to compare the performance of two different versions (A and B) of a model or system in a live production environment.