You've shipped a model that is not meeting expectations, and the team wants a structured plan to improve it. You need to decide whether the issue comes from data, model choice, generalization, or decision threshold.
How would you optimize a model that is underperforming?