S4 conference

How to deal with high attrition in a small complicated data set?