Question
During failure analysis, a log-target revenue model consistently underpredicts high-revenue accounts after predictions are exponentiated. The team averaged log-scale predictions and then exponentiated the average for reporting. What should they investigate?