> By training the model on successful applications from previous years, La Caixa says it can find and advance proposals that have the best chance of success more efficiently.

This is the wrong approach. What you should do is analyze the successful proposals with a generic model against some criteria. Maybe you use existing criteria or you identify new ones as part of the analysis. Then you analyze new applications against the criteria itself, not against previous proposals. Then a human reviews the analysis results and adds in some discretionary insights. All of that is used to determine which applications pass on.