It's not that bad in practice ;-)... Let's assume random distribution of used features among your customers. If you shoot randomly at your users with a 2% hit

It's not that bad in practice ;-)... Let's assume random distribution of used features among your customers. If you shoot randomly at your users with a 2% hit rate [alienate them by dropping a feature], your user will stay with your product with 98% probability. Do it 50 times, any user will have a 0.98^50 = 36% chance of not having been hit [having at least one important feature taken away from him that makes him stop using your product...], so only (100%-36%) 64% of users will drop your product and go elsewhere.