The default should be to set the edge to 0. Only use a nonzero edge if you have a clear reason to pass up a profitable push according to the ICM, or to make a push which loses according to the ICM.
To get good advice from SNG Wizard, you must adjust the ranges it gives your opponents. The default ranges SNG Wizard gives are often too tight against small blind shoves, and too loose against shoves from outside the blinds. The result is that if you don't adjust the default ranges, SNG Wizard very often says to push ATC from the small blind, but to push very tightly outside the blinds, particularly from the CO and earlier.
Initially, you will not know how to adjust SNG Wizard's ranges. As you get better at putting your opponents on a range, SNG Wizard will become more valuable. Some steps you can take toward learning to put your opponents on good ranges include the following:
1) Ask what the game-theoretically optimal range is. Use a tool like PokerStrategy's ICM Trainer, or the HoldemResources.net Nash equilibrium calculator. It is very valuable to learn how the game-theoretically optimal ranges vary even when you play against real opponents.
2) Ask what you would push/call in that situation. Of course, your opponents do not play exactly as you do, but put yourself in the big blind's position. What would you call? Then find out what the small blind should push against a player who plays that way.
3) Ask what you believe your typical opponents call in situations like that. Some players are just good at remembering how often they get called. Others remember particular hands. Be careful that you will see some outliers, and you don't get to see a hand when an opponent folds. If one player calls with 98o, and another folds QTo, you get to see the call and not the fold, and you should not assume that all players call with 98o and anything better. You can use a database program to investigate calling frequencies in particular situations once you have a large sample.
4) Take notes on particular opponents. Develop better stereotypes of players (e.g., the tight-early aggressive late multitabler, or the 70/30 early casual player), and better models of particular players.