[ilyas121]

Maybe go off historical data. If it's consistently off by a factor like 2-3x. Then take the prediction and just scale it?