How to analyze/debug a computer vision problem

Imagine you join a new organization and there are multiple problems in the business being solved using computer vision.
Your task is to improve the performance metric of an existing computer vision pipeline
eg: measures the distance of an object from the camera location using a calibrated camera.

What is the best approach to analyze such a pipeline? Is there any such "better way" of doing it?

analyze computer computer vision computervision vision