We need to understand the specific problem and use an appropriate model to solve that specific problem. Based on the specific category of the problem, we need to experiment with various algorithms before making the decision to use a particular model in production. In a specific category of algorithms, it is possible to deploy an incorrect model that does not produce accurate results on the new dataset.

We need to understand the data pattern over a period of time before finally choosing the model that is deployed. This is especially important in the case of mission-critical AI applications that augment human capabilities.