Question
A team tunes a delivery-time regression model using MAE and selects a model with acceptable average error. Operations rejects it because occasional very large late-delivery errors are much more costly than small errors. What should the team change in the tuning setup?