Thanks! The major reasons for doing this is to really unbox the ‘black-box’ model’s behavior and prediction decisions. This in turn helps us understand how our model works better (on actual data rather than just having a general conceptual overview of a model) and also explain how the model takes decisions to the business (who might not be very technically savvy or interested in knowing mathematical details of how a model works, but visuals helps in that case). Part 1 of my series talks a bit about this. If folks are in general happy about a model’s performance and don’t want to ‘audit’ the model then we wouldn’t need a whole lot of this!