Building a yield model your enrollment committee will trust
A 45-minute session with two VPs of enrollment on moving from historical averages to institution-specific yield probability — and how to present it to a sceptical committee.
Key takeaways
- Explainability matters more than accuracy for adoption
- Three to five historical cycles is the practical minimum
- Score at the file level, plan at the segment level
- Publish model error against actual class each cycle
Why committees reject models
Enrollment committees do not reject models because they are inaccurate. They reject them because nobody can explain why a particular applicant scored the way they did, which makes it impossible to defend an aid decision.
Every score in this session is shown with its contributing factors, which is the single change that moved both institutions from pilot to operational use.
From score to action
Scores drive outreach prioritisation and melt prevention; they should not drive admission decisions. Both panelists draw that line explicitly and describe how they communicated it internally.
Keeping trust over time
Publish the model’s error against the actual class every cycle, including the cycles where it was wrong. Both institutions credit this practice for the model surviving a leadership change.