The scope of the outcome to consider is the whole.
Stealing a wallet can be considered good, if only the short term and only immediate economic benefits are considered. But we all know that stealing is bad because the total, short+long term consequences (bad reputation, eventual prison, social isolation and rejection, increment of the risk of being caught, lose of the money earning capabilities, losing time, etc.) and the total economic benefit (easy money at the beginning, difficults to get money in the long term due to all previous collateral social damage) are negative.
This is precisely the typical error when assessing the validity of utilitarianism: to consider only the short term and immediate outcome. According to such wrong approach, stealing a wallet can be considered good "because the outcome is positive for the subject".
Now, considering the whole outcome benefits of an act is evidently difficult. Is it good to legalize marihuana? To perform human cloning? But the assessment of such difficulty is a different problem. That is not a problem of utilitarianism.
I believe it is Moore who proposed this approach: not only the sum of the parts must be consider (which is already a nice reference), but the whole as an organic unity.
See the Ideal Utilitarianism entry on the SEP.
Consider also that when confronting the benefits of the subject against the benefits of the whole society, society has the priority due to simple majority. If I like to kill and society rejects murder, it is more probable for me to get the worst total outcome.