Ideally, you would pay for the final outcomes which you care about, eg paying a certain amount per unit reduction in child mortality, reduced disease prevalence, improved test scores, etc. If this is too difficult, then you could pay for intermediate results, eg number of children vaccinated, number of people protected by bednets, etc.

A concern, as I see it, is the first part (paying for outcomes) is too difficult to successfully do on a large scale (though social impact bonds do seem promising from my first glances), and the second part (paying for outputs, like number of bed nets) already happens.