I've had 5 different tunes on my non-M, and 3 tunes on each of my Z4M's. The only time there were 20+ whp gains were when the tune accompanied a large hardware change, especially on the M's with alpha-n tuning. I improved 21.1 whp on a supercharged Z4M after swapping in catless stepped headers in place of the stock North American catted headers, and a stepped s-pipe with high flow metallic cats in place of the large OEM 400 cell cats. So, these were huge exhaust system changes, including different O2 sensor locations and going from 4 x 400 cell large cats to 2 x 100 cell tiny cats, on a car with alpha-n tuning (much less adaptive, where the car is essentially flying blind). And going from a canned supercharged tune (buffered for any car in any condition) to a custom dyno tune only netted a 5% gain in power. No way I believe a tuner can honestly pull 20 whp out of an otherwise stock non-M that still utilizes a MAF sensor.
Every time I made a new hardware change, power was left on the table until the tuning was optimized to reflect what was on the car. Tuning itself didn't create any power. And tuning is less important on cars utilizing a MAF sensor (pretty much every non-supercharged or non-carbon airbox car out there).
One of my non-M tunes was the Conforti Shark Injector, which was great due to the intuitive throttle remap and 7k rpm redline, but 5 whp is barely noticeable.
It sucks, but the way it typically works is a tuner will send a canned file they purchased off software developers, manipulate the dyno graphs to flatter the ego of the customer, and the customer can then feel good thinking his car has an extra ___ horsepower over the average car.