Admissions Intelligence Build
An applicant-level yield model replaces the flat historical assumption six campuses had been offering against, letting enrollment and aid decisions track actual likelihood to enroll.
The challenge
The admissions office had rich applicant data and no way to use it before offers went out:
- Offer volumes were set from a single flat yield assumption carried over from the prior year
- Scholarship dollars were allocated by program tradition, not by predicted sensitivity to aid
- Six campuses ran six different informal rules of thumb for how much to over-offer
- Enrollment misses meant late-cycle scrambles to reopen waitlists or over-enroll a program
- No visibility into which admitted students were actually likely to enroll until the deposit deadline
How it works
Modelling yield at the applicant level, not the cohort level
Five admissions cycles of applicant history were enough to build a model that outperforms a flat assumption:
- 01
Assembled five cycles of applicant-level data: program, aid offered, competing-offer signals, portal engagement, campus visits
- 02
Trained a yield-propensity model scoring every admitted student's individual enrollment probability
- 03
Validated the model against two held-out cycles before go-live
- 04
Rebuilt the offer-and-aid planning workbook around modelled yield instead of a flat historical rate
- 05
Gave each of the six campuses a live yield forecast they could adjust offer volume against in real time
- 06
Tracked actual deposits weekly against the model's forecast through the entire cycle
What we built
Key capabilities
Applicant-level yield scores
Every admitted student carries an individual enrollment probability, not a cohort-wide average.
Aid targeted at sensitivity
Scholarship dollars go where they actually shift a decision, not where tradition says they should.
One model, six campuses
Every campus forecasts against the same model, replacing six incompatible rules of thumb.
Weekly tracking through cycle
Actual deposits are checked against the forecast every week, not just at the deadline.
Before vs after
What changed in the admissions cycle
- Yield rate
- 55% → 65%
- Forecasting method
- Flat historical rate → Applicant-level model
- Aid allocation basis
- Program tradition → Predicted aid sensitivity
- Late-cycle waitlist reopenings
- Frequent → Rare
- Forecast visibility
- At deposit deadline → Weekly through cycle
Business impact
What it changed
18% relative yield improvement
Yield against modelled offers rose from a 55% baseline to 65% in the very next admissions cycle — an 18% relative gain.
Aid dollars that change outcomes
Scholarship allocation shifted from program tradition to predicted sensitivity, without increasing the total aid budget.
No more late-cycle scrambles
Weekly forecast tracking gave all six campuses time to adjust offer volume before the deposit deadline, not after it.
Technology stack
“Yield stopped being a number inherited from last year and became a number computed for this applicant — and the offer strategy followed it.”
Keep reading