You've shipped a model that is underperforming relative to expectations, and the team wants a clear plan to improve it. You need to explain how you would evaluate the current errors, identify likely causes, and decide which changes are most likely to raise performance.
How would you improve the accuracy of a model that is underperforming?
Anchor your answer in a realistic example such as a Grammarly Editor suggestion classifier, where both false positives and false negatives affect user trust and product value.