Right, I agree completely that the two are often very different things. I think though that this point raises another ethical question (and one to which I don't know that there's a clear answer): is the "best" program one that successfully targets the people most likely to be successful, or that targets the most needy? For example, in your evaluation of a cash/in-kind transfer to micro-enterprises in Ghana (with Chris Woodruff, Marcels Fafchamps and Simon Quinn) you find that when given grants, women with initial profits below the median are on average not profitable when given the transfer. In contrast, it is the wealthier households who tend to benefit the most. However, I'm not sure from that that the logical conclusion is that free in-kind transfers should only be given to wealthier households. I think there's perhaps a case to be made that poorer households are equally deserving of such an intervention, even if they are on average less likely to be able to use it successfully.
Given how little we often know about who will be most successful, I think the hypothetical scenario I raise only affects a limited number of interventions. However, even in an ideal world where we know who is most likely to benefit from a program, I'm not entirely certain that should be our primary method of targeting.