Working prototype · public SEC filings only · synthetic per-brand spend split, anchors cited
Covista's $247M advertising line is reported as one number. Three brands hide inside it.
The dollar you move between brands this term sets next year's cost-per-start. This model splits the disclosed FY25 advertising across Chamberlain, Walden, and Medical & Vet, computes an implied cost-per-start for each, and lets you reallocate to find the efficiency frontier, on Covista's own filings, before anyone asks for internal data.
Advertising was $247.4M in FY25 (13.8% of revenue), bundled inside "student services and administrative expense," with no public cost-per-start and no per-brand split. So the split here is yours to set, calibrated to public revenue, enrollment, and growth. The model never claims to know Covista's real allocation; it shows the shape of the cross-brand efficiency question on data nobody can dispute, and names exactly what the public record doesn't carry.
01 · Where the next marketing dollar should go
The answer, before you touch a control.
On the recommended scenario below, here's what the public filings imply for the next term's advertising dollar. Pick a different scenario to see the trade-off; fine-tune the exact split only if you want to.
Recommended · efficiency-first
-
-
-
-
Chamberlain
$725.8M rev · 20.9% op margin · +5.8% enr
-
implied advertising per start
-
Walden
$693.4M rev · 25.7% op margin · +15% enr
-
implied advertising per start
-
Medical & Vet
$369.1M rev · 18.6% op margin · +1% enr
-
implied advertising per start
-
Chamberlain
-
Walden
-
Medical & Vet
-
-
Fine-tune the split (optional) analyst controls: click to expand
The scenarios above set all three at once. If you want to test a specific split, move the sliders and the answer and the cards update live. The basis toggle switches the cost-per-start denominator between enrolled students and net new starts.
FY25 advertising split: $247.4M total
-
Cost-per-start basis
The three result cards and the verdict above this disclosure update as you move the sliders, scroll up to read them, or watch the bar chart. The sliders are clamped to the disclosed $247.4M total; the allocation-balance line flags when your split runs over or under the real number.
02 · What's real, what's set by you
The anchors are public. The split is a knob.
The public anchors, the calibration, and the honest limits click to expand
Public anchors (every cell from SEC EDGAR, CIK 0000730464)
Per-brand spend split (the slider): NOT disclosed. The reset default splits the $247.4M by public revenue share (40.6% / 38.8% / 20.6%): a neutral, defensible starting point, not a claim about Covista's real allocation. You set the real shape.
Per-brand enrollment counts: Chamberlain is disclosed (>40,500). Walden and Med&Vet are split from the 91,780 group total in proportion to IPEDS Fall-2024 institutional headcount (Walden 49,441; Chamberlain 32,944; Med&Vet residual), then scaled to the disclosed group total, labelled as a public-data estimate, not a Covista figure.
Net new starts (marginal basis): each brand's disclosed enrollment-growth rate applied to its estimated base → net adds. Spend ÷ net adds gives a marginal cost-per-start that overstates true CAC (spend also retains/re-enrolls), which is why both bases are shown.
Honest limits
This is not Covista's cost-per-start. Covista discloses none. Every cost-per-start here is implied from public spend ÷ public enrollment, and the true value sits between the two bases shown.
The per-brand spend split is the one input the public record doesn't carry. Getting the real split (and real cost-per-lead → cost-per-start by channel) is exactly what the engagement's data-readiness read scopes: internally, in weeks.
IPEDS and earnings-call enrollment use different cohort definitions and dates. They are not summed; IPEDS is used only for the relative split and online-delivery share.
Regulatory frame, not modelled here: which students a brand recruits has a 90/10 and Gainful-Employment dimension, not just an ROI one: the production version flags it; this prototype does not compute it.