The way I’ve tended to approach them is to rather ignore the prescription from the manufacturer.
The problem is that not all cookie bites are alike, some are basic SN losses, some are congenital issues, some are mixed losses and others are due to dead-spots.
Ideally you need to split the loss into 3parts. To start with, a flat prescription to begin with using the two ‘shoulders’ of the loss as the start point, somewhere near 1/3gain. Then consider the actual cookie bite section and apply the 1/3 gain average again. This effectively gives you a Low/Mid/High setting.
If you look at a speech mapped REM, this bubble (inverse cookie bite for output) shape is fairly apparent.
Now, up you can futz about with levels, but broadly speaking, you need to preserve the general shape to maintain the amplification across the board, but the transition frequencies at either edge of the cookie-bite might be pitched somewhere between the two gain levels. Loudness is a massive problem though, people struggle with the amount of energy in the signal.
My historical approach is to follow the steps above, then give multiple programs and volume control refine what works best for the client. Usually you’ll find what’s working best after a couple of steps, make this P1 and put derivative programs in the other slots. Leave the volume control still with the maximum range and cut the MPO’s down to 105 maximum, even where the dynamic range of the loss would demand higher output. This step ensures that you limit the over amplification of adjacent frequencies.