How to read this
A start is a student who actually begins classes (not an applicant, a deposit, or an enrollment). 10 to 40% of admits never start (the "summer melt"; NCAN), so every figure here is cost per start: the milestone that ties to tuition, never the softer enroll or deposit. This synthetic institution is modeled as a nonprofit with a professional and online portfolio; in production the curves and funnels are fit per sector, because for-profit and nonprofit economics differ enough that they are never pooled.
Where your next $1,000 goes
The budget is spent $1K at a time, each chunk going to whichever channel buys the most incremental starts at that moment, respecting seat caps, which mute a channel once its programs fill. Click to add a chunk and watch the marginal table reshuffle; the Optimize tab runs this rule to the end of the budget.
Channel spend / month
Program mix: how spend is apportioned
Weights normalize to 100%. Set a program to 0 to drop it from the buy.
Panel A: starts & cost per start, by program
| Program | Inquiries | Apps | Starts | Seats | Cost / start |
|---|
Marginal economics at the current operating point
| Channel | Spend | % of ceiling | Next $1K buys | Marginal cost/start |
|---|
When the marginal column spans 3x across channels, the last dollars are in the wrong place. A lead score can't see this: it ranks people, not budgets.
Where this lands against the benchmarks
Reference lines: ~$1,505 median cost per enrolled start (undergrad) and ~$3,804 (graduate). Only 43% of higher-ed marketers can compute their own dot on this axis.
Panel B: same budget, reallocated
Greedy hill-climb: the total budget is re-spent $1K at a time, each chunk going to whichever channel buys the most incremental starts at that moment, respecting seat capacity, which mutes a channel once its programs fill.
Current vs optimized allocation
| Channel | Current | Optimized | Shift |
|---|
Hand-allocation on historical averages misplaces an estimated 20-30% of media budgets: this is that number, made specific to one institution.
Panel C: channel response curves
Funnel assumptions (editable)
| Program | Inquiry → app | App → enroll | Seat capacity / mo |
|---|
Every output on every tab recomputes from these cells. In production they come from the SIS join, not a slider.
One-pager: the outcome calculator
An agency repositioning from lead volume to cost-per-enrolled-start economics needs to forecast enrolled starts under different spend allocations; otherwise outcome-based deals can't be quoted, only hoped for. A lead score is not a budget-allocation method. The missing layer is per-channel response ceilings and saturation: roughly 75% of performance marketers report diminishing returns on social spend, yet most plans still extrapolate last quarter's average CPL as if it were the marginal one.
- Each channel × program pair is a concave response curve: inquiries = C · (1 − e^(−spend/k)), where C is the saturation ceiling and k sets how fast the channel bends. Paid Search saturates fastest; CTV is cheaper at the margin but noisier.
- Inquiries flow through each program's own funnel (inquiry→app, app→enroll, seat capacity) to monthly enrolled starts and a blended cost per start, against the published undergrad/grad benchmarks.
- The marginal column (what the next $1K buys in each channel) makes misallocation visible, and a greedy optimizer prices it: same budget, more starts.
The same engine with curves fit from the agency's own $100M+ of historical spend: hierarchical Bayesian or saturating-regression fits per channel-by-vertical, so a small program borrows strength from the portfolio. Refreshed monthly; joined to SIS enrollment records so the response variable is actual starts, not proxied leads. Only 43% of higher-ed marketers track cost per enrolled student. The institutions that can see this screen are the ones that sign outcome deals, and the ones that survive the demographic cliff ahead.
I've sold and built exactly this stack: Director of Solutions at Sparkroom (higher-ed performance marketing), closing and implementing $1.5M/yr in education SaaS; senior data scientist at Meta on ads and media analytics; and prior open-source work on a constrained media-budget optimizer with a CPI-vs-CPS objective toggle (the same allocation math under a different acronym). Jeff Pinto · jeff@jeffpinto.com · jeffpinto.com
Synthetic institution, but every output is calibrated to published 2026 higher-ed benchmarks so the numbers reconcile: average cost per enrolled student $2,849 ($1,505 undergrad, $3,804 graduate) and average cost per inquiry $140 (range $29-$450), Search Influence 2026 benchmarks / UPCEA. The funnel separates enroll from start because 10-40% of admits never start (the "summer melt"). The channel response curves are illustrative concave shapes (inquiries = C·(1-e^(-spend/k))) tuned so the blended cost per start lands inside that $1.5K-$3.8K band and Paid Search saturates before CTV; they are deliberately not channel-accurate spot rates. In production the C and k are fit per channel-by-vertical from the agency's own $100M+ of spend, not assumed.