We introduce a new way of selecting among model forms in automated ETS forecasting routines, here addressed as treating. The approach operates by subsetting the pool of competing models based on the information delivered by their prediction intervals. An application to exponential smoothing formulations gave rise to alternative forecasting methods, the Treated ETS and the Treated AICc weights. By the same token, we also proposed a pruning strategy that can be used to enhance the accuracy of forecasts arising from any forecast combination method, provided that the models to be combined are able to generate prediction intervals to their point forecasts.