Higher education runs its enrollment marketing on Cost Per Inquiry and Cost Per Enrolled Student. The institution runs on retained, graduating students and the tuition revenue, outcomes and alumni value that follow them. The space between those two numbers is where online program managers spend a partner's budget filling a cohort that melts by week six, a regional university pays for inquiries in programs it is quietly closing, and the provost reads an enrollment report that says "starts up 18 percent" in the same year first-to-second-year retention falls.
Open the dashboard at any enrollment marketing team in 2026 (a university's central marketing office, an OPM running twenty partner programs, an online college) and the spine of the reporting is the same. Impressions, clicks, inquiries, applications, admits, deposits, enrolled students. A cost per inquiry and a cost per start, probably split by program and by term. Somewhere in a different office, a retention and completion report the registrar owns and the marketing team reads once a year.
That structure made sense when the inquiry was the scarce thing. The prospective student searched, clicked, submitted an inquiry form, and an admissions counselor worked it to a start. The marketer's job ended at the inquiry. Persistence, completion and outcomes belonged to student success.
That separation no longer holds. Performance Max bids on signals the marketer never sees, toward whichever conversion the account defines. Demand Gen and YouTube put the program in front of a prospect months before the inquiry, on a surface the reporting cannot isolate. Feed either a start target and the platform finds the cheapest starts: the students most likely to enroll and least likely to persist, in the programs with the lowest completion rates, from the sources that produce volume. The platform does what it is told. It is the metric that is wrong.
What Cost Per Enrolled Student actually optimizes for
Cost Per Enrolled Student treats every start as identical at the moment the student is counted in the census. A student who completes the program, finishes in the expected term count, and becomes an alumnus the institution can point to is worth the same in CPES as a student who attends two weeks, stops out before the add-drop deadline, and shows up in the retention report as a loss the institution absorbed. The dashboard cannot tell the difference. The platform optimizes toward whichever audience produces the start cheapest.
The downstream effect is predictable in enrollment accounts of any size. The cheapest starts cluster where persistence is worst: the broadest degree-seeking terms, the lead-aggregator sources, the programs with the lowest barrier to entry and the highest melt. Marketing hits the start number. Student success watches first-term persistence slide. Finance sees cost per completed credit hour drift upward and the accreditor notices the completion rate. Everyone's report is correct.
Run the arithmetic on a regional university's online programs, with illustrative numbers. Say enrollment marketing produces 2,000 starts a year at a blended $1,200 per start. Starts from direct search and program-page inquiries cost $1,700 and persist past the first year at 72 percent, in programs where completion runs 60 percent. Starts from lead aggregators and accelerated-start campaigns cost $800 and persist at 41 percent, in programs where completion runs 30 percent. At a 50/50 mix by starts, the cheap half produces roughly 300 eventual graduates and the expensive half roughly 600, and the expensive half cost about twice as much per start to buy, so the two halves land at nearly the same cost per graduate before retention and completion are weighted, and far apart after. Move a third of the aggregator budget to direct channels at the higher cost per start and total starts fall, which the enrollment report records as a loss, while completed credit hours, net tuition revenue and the completion rate the accreditor reads all rise. The institution measured the start because the start was the thing it could count in the fall.
Last-click attribution had this same shape. Meridian MMM exists because enterprise marketers finally got tired of the Search-attribution illusion. Cost Per Enrolled Student is the higher-ed version of the same illusion, and the gap between the marketing event and the real decision is longer here than in any other vertical in this series.
The metric that actually runs the institution
Cost Per Graduate is the marketing investment required to produce one student who persists to completion, weighted by the expected net tuition revenue and outcome value of that student. It is not a new number. Every CFO in higher education runs net tuition revenue per student, and every institutional research office runs persistence and completion by cohort. The novelty is putting it on the marketing dashboard, in the bidding signal and in the OPM contract, instead of letting it live in a report that arrives three years after the campaign ended.
Three components matter. The persistence rate: what share of starts from this source and program survive the first term, the first year and the degree. The expected net tuition revenue per student: tuition less discount, by program and by modality. The outcome value: the placement and alumni value the institution uses in its own accountability reporting.
The math is simple. The lag is the problem. A graduate is two to six years from the inquiry, which is why no bidding algorithm can learn from the graduate event directly. The practical unit is the leading indicator the institution's own data says predicts completion: first-term persistence past the census date, or second-term re-enrollment. The data lives in the student information system, the enrollment CRM and the ad platforms; joining them at the inquiry level, by source, program and term, is the actual work.
What changes when you optimize on Cost Per Graduate
Run the same media plan on both metrics against the same inquiry data and the budget shifts in three predictable ways.
Program and audience mix moves toward completion
CPES is indifferent to the program. Cost Per Graduate is not. The programs and audiences with the highest persistence are rarely the cheapest to buy an inquiry in, because they are the ones every competitor is also recruiting for. Cost Per Graduate moves spend toward the programs that complete, pulls it off the high-melt programs that were producing cheap starts, and treats academic capacity and support staffing as a marketing input. The first thing most institutions discover is that they have been paying to fill programs the provost was already reviewing.
The melt problem moves from admissions to marketing
When the signal is "start," summer melt and first-term stop-out are an admissions or student-success problem. When the signal is "persisting student," they are a marketing efficiency loss visible by source and program within one term. Marketing, admissions and student success start looking at the same cohorts. The lead-aggregator channel gets judged on persistence, which is usually the end of the lead-aggregator channel. The OPM revenue share gets renegotiated around retained students instead of starts.
Creative shifts from "start next month" to fit
CPES rewards creative that maximizes starts: accelerated start dates, "no application fee," the shortest path to enrollment. The prospect most responsive to that creative is the prospect least likely to persist, because the ad selected for speed, not fit. Cost Per Graduate rewards creative that pre-qualifies: the program's actual workload, the time to completion, the outcomes, the support model. The inquiry count drops. The persistence rate rises. The CFO and the accreditor both notice inside two terms.
"The prospect most responsive to 'start next Monday, no application fee' is the prospect most likely to be gone by week six. That is not a student-success problem. It is a metric problem."
Cost Per Enrolled Student vs. Cost Per Graduate, side by side
How to feed Cost Per Graduate into Google Ads
The implementation is longer here than in any other vertical in this series, because of the lag, and it is still shorter than most institutions expect. Google already built the primitives: offline conversion imports and enhanced conversions for leads carry a persistence event from the SIS back to the click, conversion value rules let value vary by program, and value-based bidding inside Performance Max and Search acts on it. What is missing is the proxy event and the value.
The shape of the build: define the bidding conversion as first-term persistence past census (or second-term re-enrollment, if the institution's own data says it predicts completion better), fired from the SIS or the enrollment CRM and matched back to the inquiry with enhanced conversions for leads. Calculate persistence by source and program from institutional research. Calculate expected net tuition revenue per persisting student by program and modality from finance. Multiply to produce an expected value per persisting student. Pass it as the conversion value, and import the upstream application and deposit events as lower-value conversions so the bidding algorithm is not starved during the lag. Move campaigns to a value-based bid strategy once the persistence events are flowing. Google's AI will reweight toward higher expected persisting-student value within two to four weeks of the data arriving, which in higher ed means planning the build a term ahead of the campaign it is meant to improve.
The harder work is institutional. Cost Per Graduate forces marketing, admissions, student success, institutional research and finance into the same data, and in an OPM relationship it forces the contract to be renegotiated around retained students. That requires a sponsor at the provost or CFO level, and it is also the moment enrollment marketing starts speaking the language in Marketing on the P&L: how the decision reads in net tuition revenue and completion, not in starts.
Where compliance actually sits
Cost Per Graduate is an internal economic metric. It does not appear in recruiting materials, and it does not change what the institution discloses under gainful-employment, misrepresentation or incentive-compensation rules. Two things matter upstream. The expected-value model must not use attributes that would create admissions or recruiting practices the institution could not defend under Title VI, Title IX or state consumer-protection law, and the persistence data flowing to the ad platform must be de-identified and matched in a way that satisfies FERPA and the institution's data-governance policy. Built properly, Cost Per Graduate is more defensible than Cost Per Enrolled Student, because it aligns the recruiting incentive with the student's completion rather than against it, which is the direction every regulator has been pushing for a decade.
What this looks like on the P&L
The financial argument for Cost Per Graduate is the same shape as every argument on this site: the marketing decision and the business decision become the same decision.
An institution running on CPES allocates budget to whatever produces starts cheapest. The cohort arrives. Some fraction persists past census. Some fraction of those returns for a second year. Some fraction of those graduates. The enrollment report looks fine in the fall. By the following fall, net tuition revenue per student is down, the first-to-second-year retention rate has fallen, the accreditor has questions about completion, and the budget conversation is about cutting marketing in the one place it was recruiting students who stayed. Both reports are technically correct.
An institution running on Cost Per Graduate allocates budget against expected persisting-student value at the point of the bid, by program. The cohort arrives already weighted toward students who persist. The annual reconciliation still happens, but the variance is information rather than damage: it improves next year's persistence model instead of indicting this year's spend. The same marketing dollars produce more completed credit hours, more net tuition revenue, a better completion rate and a lower Cost Per Decision, because the program-capacity decision and the media decision are finally the same decision. FinTech has the same shape of metric problem with funding in place of persistence. The lesson generalizes.
This is one vertical instance of a broader argument: every marketing KPI in common use measures activity, not decisions, and agentic AI is the first technology that makes the correction operable rather than aspirational. Higher education is where the lag between the marketing event and the real decision is longest, which is why the gap has been invisible for so long and why closing it is worth the most.
The marketing team is not optimizing wrong. The metric is. Cost Per Enrolled Student was the best an institution could do when the inquiry was scarce, the platform did not bid on its own, and nobody had joined the SIS to the ad account. None of those conditions hold anymore. The institutions that rebuild their KPI around the graduate first will spend the next two enrollment cycles compounding a completion advantage the rest of the sector is still reporting in starts.
Sources & further reading
- About offline conversion imports (Google Ads Help).
- About enhanced conversions for leads (Google Ads Help).
- About conversion value rules (Google Ads Help). The mechanism for varying value by program.
- Value-based bidding best practices (Google Ads Help).
- About Performance Max campaigns (Google Ads Help).
- About Demand Gen campaigns (Google Ads Help). How consideration forms on YouTube and Discover before the inquiry.