Merit aid is rightly controversial but "everybody does it." In its current form we rely almost entirely on its pecuniary qualities. There is an emotional component insofar as a student feels recognized for her achievements, but mostly we expect it to have positive yield consequences because it lowers the cost of attendance beyond what need-based aid accomplishes. In a not dissimilar move, we award work-study money with no suggestion of what the work might be and the work often enough is working in a clerical position somewhere in the college. WHY NOT exploit the relational and identity dimensions by attaching this aid to a concrete programmatic offering. Example: "We would also like to offer you a $4,000 fellowship to participate in our first year "Innovation Exploration" program or "Our offer also includes a $3,000 research assistantship you can apply after your first year to work as a research assistant to a professor."
This is a gigantic underutilization of the leverage value of these resources. Why not dedicate part of what we call merit aid and work-study to targeted designations and programmatic features of the college. Rather than a letter that says "we would like to offer you an additional $4,000 and call you a "dean's scholar," we write one that says "based on our reading of your application we would like to invite you to be a part of a new program at Mills, the "Social Innovation Corps" or the "Student-Faculty Research Collaborative" or the "Mills Oakland Urban Research Partnership." We then structure something like an LLC around these and deploy some part of an intro course to the project, etc.
Another possibility is to offer students a "research fellowship" which would designate them as a faculty research assistant from day 1. The letter could give them a choice based on a number of faculty who have expressed interest in them based on their record. We faculty would work together to coordinate a good experience and ways of cross training students so that it was actually helpful rather than just more work supervising.
All of these things could be done in random experiments to see if they are effective. Example: we identify candidates in the admissions pool and then randomly make the offer to some and then we track the yield on the two groups. Alternatively, we could do this further upstream - perhaps when inquiries are first made we follow up with an introduction to a program like this - "based on your profile we thought you might be interested in being considered for this program at Mills" - and then we see what our conversion rate is for those we do this for and those we do not.